From 51fd1e24cebce0b09072d27c5a0eaa8c7af945a9 Mon Sep 17 00:00:00 2001
From: Thomas Pinder <tompinder@live.co.uk>
Date: Wed, 12 Oct 2022 20:20:57 +0000
Subject: [PATCH 01/14] Distrax to numpyro

---
 examples/barycentres.pct.py     | 19 +++++++-------
 examples/classification.pct.py  | 32 +++++++++++------------
 examples/graph_kernels.pct.py   |  8 +++---
 examples/haiku.pct.py           |  8 +++---
 examples/kernels.pct.py         | 17 ++++++------
 examples/regression.pct.py      | 15 ++++++-----
 examples/tfp_integration.pct.py |  6 ++---
 examples/yacht.pct.py           | 10 +++----
 gpjax/config.py                 | 11 ++++----
 gpjax/gps.py                    | 46 ++++++++++++++-------------------
 gpjax/likelihoods.py            | 24 +++++++++--------
 gpjax/parameters.py             | 16 +++++-------
 12 files changed, 105 insertions(+), 107 deletions(-)

diff --git a/examples/barycentres.pct.py b/examples/barycentres.pct.py
index b3e0738d..b24c36cc 100644
--- a/examples/barycentres.pct.py
+++ b/examples/barycentres.pct.py
@@ -7,7 +7,7 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.11.5
+#       jupytext_version: 1.11.2
 #   kernelspec:
 #     display_name: Python 3.9.7 ('gpjax')
 #     language: python
@@ -18,18 +18,19 @@
 # # Gaussian Processes Barycentres
 #
 # In this notebook we'll give an implementation of <strong data-cite="mallasto2017learning"></strong>. In this work, the existence of a Wasserstein barycentre between a collection of Gaussian processes is proven. When faced with trying to _average_ a set of probability distributions, the Wasserstein barycentre is an attractive choice as it enables uncertainty amongst the individual distributions to be incorporated into the averaged distribution. When compared to a naive _mean of means_ and _mean of variances_ approach to computing the average probability distributions, it can be seen that Wasserstein barycentres offer significantly more favourable uncertainty estimation.
+#
 
+# %%
 import typing as tp
 
-import distrax as dx
 import jax
 import jax.numpy as jnp
 import jax.random as jr
 import jax.scipy.linalg as jsl
 import matplotlib.pyplot as plt
+import numpyro
 import optax as ox
 
-# %%
 import gpjax as gpx
 
 key = jr.PRNGKey(123)
@@ -136,7 +137,7 @@ def sqrtm(A: jnp.DeviceArray):
 def wasserstein_barycentres(
     distributions: tp.List[dx.Distribution], weights: jnp.DeviceArray
 ):
-    covariances = [d.covariance() for d in distributions]
+    covariances = [d.covariance_matrix for d in distributions]
     cov_stack = jnp.stack(covariances)
     stack_sqrt = jax.vmap(sqrtm)(cov_stack)
 
@@ -156,7 +157,7 @@ def step(covariance_candidate: jnp.DeviceArray, i: jnp.DeviceArray):
 # %%
 weights = jnp.ones((n_datasets,)) / n_datasets
 
-means = jnp.stack([d.mean() for d in posterior_preds])
+means = jnp.stack([d.mean for d in posterior_preds])
 barycentre_mean = jnp.tensordot(weights, means, axes=1)
 
 step_fn = jax.jit(wasserstein_barycentres(posterior_preds, weights))
@@ -167,7 +168,7 @@ def step(covariance_candidate: jnp.DeviceArray, i: jnp.DeviceArray):
 )
 
 
-barycentre_process = dx.MultivariateNormalFullCovariance(
+barycentre_process = numpyro.distributions.MultivariateNormal(
     barycentre_mean, barycentre_covariance
 )
 
@@ -186,9 +187,9 @@ def plot(
     ci_alpha: float = 0.2,
     linewidth: float = 1.0,
 ):
-    mu = dist.mean()
-    sigma = dist.stddev()
-    ax.plot(xtest, dist.mean(), linewidth=linewidth, color=color, label=label)
+    mu = dist.mean
+    sigma = jnp.sqrt(dist.covariance_matrix.diagonal())
+    ax.plot(xtest, dist.mean, linewidth=linewidth, color=color, label=label)
     ax.fill_between(
         xtest.squeeze(), mu - sigma, mu + sigma, alpha=ci_alpha, color=color
     )
diff --git a/examples/classification.pct.py b/examples/classification.pct.py
index 0b10a9b2..876da984 100644
--- a/examples/classification.pct.py
+++ b/examples/classification.pct.py
@@ -7,7 +7,7 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.11.5
+#       jupytext_version: 1.11.2
 #   kernelspec:
 #     display_name: Python 3.9.7 ('gpjax')
 #     language: python
@@ -20,7 +20,6 @@
 # In this notebook we demonstrate how to perform inference for Gaussian process models with non-Gaussian likelihoods via maximum a posteriori (MAP) and Markov chain Monte Carlo (MCMC). We focus on a classification task here and use [BlackJax](https://github.com/blackjax-devs/blackjax/) for sampling.
 
 import blackjax
-import distrax as dx
 
 # %%
 import jax
@@ -28,6 +27,7 @@
 import jax.random as jr
 import jax.scipy as jsp
 import matplotlib.pyplot as plt
+import numpyro
 import optax as ox
 from jaxtyping import Array, Float
 
@@ -101,8 +101,8 @@
 
 predictive_dist = likelihood(latent_dist, map_estimate)
 
-predictive_mean = predictive_dist.mean()
-predictive_std = predictive_dist.stddev()
+predictive_mean = predictive_dist.mean
+predictive_std = jnp.sqrt(predictive_dist.variance)
 
 fig, ax = plt.subplots(figsize=(12, 5))
 ax.plot(x, y, "o", label="Observations", color="tab:red")
@@ -139,7 +139,7 @@
 # The Laplace approximation improves uncertainty quantification by incorporating curvature induced by the marginal log-likelihood's Hessian to construct an approximate Gaussian distribution centered on the MAP estimate.
 # Since the negative Hessian is positive definite, we can use the Cholesky decomposition to obtain the covariance matrix of the Laplace approximation at the datapoints below.
 # %%
-f_map_estimate = posterior(D, map_estimate)(x).mean()
+f_map_estimate = posterior(D, map_estimate)(x).mean
 
 jitter = 1e-6
 
@@ -150,7 +150,7 @@
 L_inv = jsp.linalg.solve_triangular(L, I(D.n), lower=True)
 H_inv = jsp.linalg.solve_triangular(L.T, L_inv, lower=False)
 
-laplace_approximation = dx.MultivariateNormalFullCovariance(f_map_estimate, H_inv)
+laplace_approximation = numpyro.distributions.MultivariateNormal(f_map_estimate, H_inv)
 
 from gpjax.kernels import cross_covariance, gram
 
@@ -161,26 +161,26 @@
 
 
 def predict(
-    laplace_at_data: dx.Distribution,
+    laplace_at_data: numpyro.distributions.Distribution,
     train_data: Dataset,
     test_inputs: Float[Array, "N D"],
     jitter: int = 1e-6,
-) -> dx.Distribution:
+) -> numpyro.distributions.Distribution:
     """Compute the predictive distribution of the Laplace approximation at novel inputs.
 
     Args:
         laplace_at_data (dict): The Laplace approximation at the datapoints.
 
     Returns:
-        dx.Distribution: The Laplace approximation at novel inputs.
+        numpyro.distributions.Distribution: The Laplace approximation at novel inputs.
     """
     x, n = train_data.X, train_data.n
 
     t = test_inputs
     n_test = t.shape[0]
 
-    mu = laplace_at_data.mean().reshape(-1, 1)
-    cov = laplace_at_data.covariance()
+    mu = laplace_at_data.mean.reshape(-1, 1)
+    cov = laplace_at_data.covariance_matrix
 
     Ktt = gram(prior.kernel, t, params["kernel"])
     Kxx = gram(prior.kernel, x, params["kernel"])
@@ -214,7 +214,7 @@ def predict(
     )
     covariance += I(n_test) * jitter
 
-    return dx.MultivariateNormalFullCovariance(
+    return numpyro.distributions.MultivariateNormal(
         jnp.atleast_1d(mean.squeeze()), covariance
     )
 
@@ -226,8 +226,8 @@ def predict(
 
 predictive_dist = likelihood(latent_dist, map_estimate)
 
-predictive_mean = predictive_dist.mean()
-predictive_std = predictive_dist.stddev()
+predictive_mean = predictive_dist.mean
+predictive_std = predictive_dist.variance**0.5
 
 fig, ax = plt.subplots(figsize=(12, 5))
 ax.plot(x, y, "o", label="Observations", color="tab:red")
@@ -274,7 +274,7 @@ def predict(
 # %%
 # Adapted from BlackJax's introduction notebook.
 num_adapt = 500
-num_samples = 500
+num_samples = 200
 
 mll = jax.jit(posterior.marginal_log_likelihood(D, constrainer, negative=False))
 
@@ -338,7 +338,7 @@ def one_step(state, rng_key):
 
     latent_dist = posterior(D, ps)(xtest)
     predictive_dist = likelihood(latent_dist, ps)
-    samples.append(predictive_dist.sample(seed=key, sample_shape=(10,)))
+    samples.append(predictive_dist.sample(key, sample_shape=(10,)))
 
 samples = jnp.vstack(samples)
 
diff --git a/examples/graph_kernels.pct.py b/examples/graph_kernels.pct.py
index 0069ad4e..656983fb 100644
--- a/examples/graph_kernels.pct.py
+++ b/examples/graph_kernels.pct.py
@@ -7,7 +7,7 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.11.5
+#       jupytext_version: 1.11.2
 #   kernelspec:
 #     display_name: Python 3.9.7 ('gpjax')
 #     language: python
@@ -88,7 +88,7 @@
 }
 
 fx = f(true_params)(x)
-y = fx.sample(seed=key).reshape(-1, 1)
+y = fx.sample(key).reshape(-1, 1)
 
 D = gpx.Dataset(X=x, y=y)
 
@@ -145,8 +145,8 @@
 initial_dist = likelihood(posterior(D, params)(x), params)
 predictive_dist = likelihood(posterior(D, learned_params)(x), learned_params)
 
-initial_mean = initial_dist.mean()
-learned_mean = predictive_dist.mean()
+initial_mean = initial_dist.mean
+learned_mean = predictive_dist.mean
 
 rmse = lambda ytrue, ypred: jnp.sum(jnp.sqrt(jnp.square(ytrue - ypred)))
 
diff --git a/examples/haiku.pct.py b/examples/haiku.pct.py
index b2434433..9e014412 100644
--- a/examples/haiku.pct.py
+++ b/examples/haiku.pct.py
@@ -7,7 +7,7 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.11.5
+#       jupytext_version: 1.11.2
 #   kernelspec:
 #     display_name: Python 3.9.7 ('gpjax')
 #     language: python
@@ -19,6 +19,7 @@
 #
 # In this notebook we demonstrate how GPJax can be used in conjunction with [Haiku](https://github.com/deepmind/dm-haiku) to build deep kernel Gaussian processes. Modelling data with discontinuities is a challenging task for regular Gaussian process models. However, as shown in <strong data-cite="wilson2016deep"></strong>, transforming the inputs to our Gaussian process model's kernel through a neural network can offer a solution to this.
 
+# %%
 import typing as tp
 
 import distrax as dx
@@ -31,7 +32,6 @@
 from chex import dataclass
 from scipy.signal import sawtooth
 
-# %%
 import gpjax as gpx
 from gpjax.kernels import Kernel
 
@@ -179,8 +179,8 @@ def forward(x):
 latent_dist = posterior(D, final_params)(xtest)
 predictive_dist = likelihood(latent_dist, final_params)
 
-predictive_mean = predictive_dist.mean()
-predictive_std = predictive_dist.stddev()
+predictive_mean = predictive_dist.mean
+predictive_std = jnp.sqrt(predictive_dist.variance)
 
 fig, ax = plt.subplots(figsize=(12, 5))
 ax.plot(x, y, "o", label="Observations", color="tab:red")
diff --git a/examples/kernels.pct.py b/examples/kernels.pct.py
index adfea990..062ca11c 100644
--- a/examples/kernels.pct.py
+++ b/examples/kernels.pct.py
@@ -7,7 +7,7 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.11.5
+#       jupytext_version: 1.11.2
 #   kernelspec:
 #     display_name: Python 3.9.7 ('gpjax')
 #     language: python
@@ -19,6 +19,7 @@
 #
 # In this guide, we introduce the kernels available in GPJax and demonstrate how to create custom ones.
 
+# %%
 import distrax as dx
 import jax
 import jax.numpy as jnp
@@ -29,7 +30,6 @@
 from jaxtyping import Array, Float
 from optax import adam
 
-# %%
 import gpjax as gpx
 
 key = jr.PRNGKey(123)
@@ -64,7 +64,7 @@
     prior = gpx.Prior(kernel=k)
     params, _, _, _ = gpx.initialise(prior, key).unpack()
     rv = prior(params)(x)
-    y = rv.sample(sample_shape=10, seed=key)
+    y = rv.sample(key, sample_shape=(10,))
 
     ax.plot(x, y.T, alpha=0.7)
     ax.set_title(k.name)
@@ -208,13 +208,12 @@ def _initialise_params(self, key: jr.PRNGKey) -> dict:
 # To define a bijector here we'll make use of the `Lambda` operator given in Distrax. This lets us convert any regular Jax function into a bijection. Given that we require $\tau$ to be strictly greater than $4.$, we'll apply a [softplus transformation](https://jax.readthedocs.io/en/latest/_autosummary/jax.nn.softplus.html) where the lower bound is shifted by $4.$.
 
 
+import numpyro
+
 # %%
 from gpjax.config import add_parameter
 
-bij_fn = lambda x: jax.nn.softplus(x + jnp.array(4.0))
-bij = dx.Lambda(bij_fn)
-
-add_parameter("tau", bij)
+add_parameter("tau", numpyro.distributions.transforms.SoftplusTransform())
 
 # %% [markdown]
 # ### Using our polar kernel
@@ -263,8 +262,8 @@ def _initialise_params(self, key: jr.PRNGKey) -> dict:
 
 # %%
 posterior_rv = likelihood(circlular_posterior(D, final_params)(angles), final_params)
-mu = posterior_rv.mean()
-one_sigma = posterior_rv.stddev()
+mu = posterior_rv.mean
+one_sigma = jnp.sqrt(posterior_rv.variance)
 
 # %%
 fig = plt.figure(figsize=(10, 8))
diff --git a/examples/regression.pct.py b/examples/regression.pct.py
index 62f319e4..51735685 100644
--- a/examples/regression.pct.py
+++ b/examples/regression.pct.py
@@ -6,7 +6,7 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.11.5
+#       jupytext_version: 1.11.2
 #   kernelspec:
 #     display_name: Python 3.9.7 ('gpjax')
 #     language: python
@@ -86,9 +86,9 @@
 parameter_state = gpx.initialise(prior, key)
 prior_dist = prior(parameter_state.params)(xtest)
 
-prior_mean = prior_dist.mean()
-prior_std = jnp.sqrt(prior_dist.covariance().diagonal())
-samples = prior_dist.sample(seed=key, sample_shape=20).T
+prior_mean = prior_dist.mean
+prior_std = jnp.sqrt(prior_dist.covariance_matrix.diagonal())
+samples = prior_dist.sample(key, sample_shape=(20,)).T
 
 plt.plot(xtest, samples, color="tab:blue", alpha=0.5)
 plt.plot(xtest, prior_mean, color="tab:orange")
@@ -196,12 +196,15 @@
 #
 # Equipped with the posterior and a set of optimised hyperparameter values, we are now in a position to query our GP's predictive distribution at novel test inputs. To do this, we use our defined `posterior` and `likelihood` at our test inputs to obtain the predictive distribution as a `Distrax` multivariate Gaussian upon which `mean` and `stddev` can be used to extract the predictive mean and standard deviatation.
 
+# %%
+latent_dist
+
 # %%
 latent_dist = posterior(D, final_params)(xtest)
 predictive_dist = likelihood(latent_dist, final_params)
 
-predictive_mean = predictive_dist.mean()
-predictive_std = predictive_dist.stddev()
+predictive_mean = predictive_dist.mean
+predictive_std = jnp.sqrt(predictive_dist.covariance_matrix.diagonal())
 
 # %% [markdown]
 # With the predictions and their uncertainty acquired, we illustrate the GP's performance at explaining the data $\mathcal{D}$ and recovering the underlying latent function of interest.
diff --git a/examples/tfp_integration.pct.py b/examples/tfp_integration.pct.py
index ba26c468..7e6534a6 100644
--- a/examples/tfp_integration.pct.py
+++ b/examples/tfp_integration.pct.py
@@ -7,7 +7,7 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.11.5
+#       jupytext_version: 1.11.2
 #   kernelspec:
 #     display_name: Python 3.9.7 ('gpjax')
 #     language: python
@@ -202,8 +202,8 @@ def run_chain(key, state):
 
 predictive_dist = likelihood(posterior(D, learned_params)(xtest), learned_params)
 
-mu = predictive_dist.mean()
-sigma = predictive_dist.stddev()
+mu = predictive_dist.mean
+sigma = jnp.sqrt(predictive_dist.variance)
 
 # %% [markdown]
 # Finally, we plot the learned posterior predictive distribution evaluated at the test points defined above.
diff --git a/examples/yacht.pct.py b/examples/yacht.pct.py
index bb5683a1..b549ff27 100644
--- a/examples/yacht.pct.py
+++ b/examples/yacht.pct.py
@@ -1,11 +1,12 @@
 # ---
 # jupyter:
 #   jupytext:
+#     custom_cell_magics: kql
 #     text_representation:
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.11.5
+#       jupytext_version: 1.11.2
 #   kernelspec:
 #     display_name: Python 3.9.7 ('gpjax')
 #     language: python
@@ -17,13 +18,12 @@
 #
 # In this notebook we'll demonstrate the use of GPJax on a benchmark UCI regression problem. Such tasks are commonly used within the research community to benchmark and evaluate new techniques against those already present in the literature. Much of the code contained in this notebook can be adapted to applied problems concerning datasets other than the one presented here.
 
+# %%
 import jax.numpy as jnp
 import jax.random as jr
 import matplotlib.pyplot as plt
 import numpy as np
 import optax as ox
-
-# %%
 import pandas as pd
 from jax import jit
 from sklearn.metrics import mean_squared_error, r2_score
@@ -153,8 +153,8 @@
 latent_dist = posterior(training_data, learned_params)(scaled_Xte)
 predictive_dist = likelihood(latent_dist, learned_params)
 
-predictive_mean = predictive_dist.mean()
-predictive_stddev = predictive_dist.stddev()
+predictive_mean = predictive_dist.mean
+predictive_stddev = jnp.sqrt(predictive_dist.variance)
 
 # %% [markdown]
 # ## Evaluation
diff --git a/gpjax/config.py b/gpjax/config.py
index 13695d9e..7a191503 100644
--- a/gpjax/config.py
+++ b/gpjax/config.py
@@ -1,9 +1,10 @@
-import distrax as dx
-import jax.numpy as jnp
 import jax.random as jr
+import numpyro.distributions as npd
 from ml_collections import ConfigDict
 
 __config = None
+import distrax as dx
+import jax.numpy as jnp
 
 Identity = dx.Lambda(forward=lambda x: x, inverse=lambda x: x)
 Softplus = dx.Lambda(
@@ -45,8 +46,8 @@ def get_defaults() -> ConfigDict:
 
     # Default bijections
     config.transformations = transformations = ConfigDict()
-    transformations.positive_transform = Softplus
-    transformations.identity_transform = Identity
+    transformations.positive_transform = npd.transforms.SoftplusTransform()
+    transformations.identity_transform = npd.transforms.IdentityTransform()
 
     # Default parameter transforms
     transformations.lengthscale = "positive_transform"
@@ -63,7 +64,7 @@ def get_defaults() -> ConfigDict:
     return __config
 
 
-def add_parameter(param_name: str, bijection: dx.Bijector) -> None:
+def add_parameter(param_name: str, bijection: npd.transforms.Transform) -> None:
     """Add a parameter and its corresponding transform to GPJax's config file.
 
     Args:
diff --git a/gpjax/gps.py b/gpjax/gps.py
index db13f2f5..909a94fd 100644
--- a/gpjax/gps.py
+++ b/gpjax/gps.py
@@ -2,10 +2,10 @@
 from abc import abstractmethod, abstractproperty
 from typing import Dict
 
-import distrax as dx
 import jax.numpy as jnp
 import jax.random as jr
 import jax.scipy as jsp
+import numpyro.distributions as npd
 from chex import dataclass
 from jaxtyping import Array, Float
 
@@ -31,16 +31,16 @@
 class AbstractGP:
     """Abstract Gaussian process object."""
 
-    def __call__(self, *args: tp.Any, **kwargs: tp.Any) -> dx.Distribution:
+    def __call__(self, *args: tp.Any, **kwargs: tp.Any) -> npd.Distribution:
         """Evaluate the Gaussian process at the given points.
 
         Returns:
-            dx.Distribution: A multivariate normal random variable representation of the Gaussian process.
+            npd.Distribution: A multivariate normal random variable representation of the Gaussian process.
         """
         return self.predict(*args, **kwargs)
 
     @abstractmethod
-    def predict(self, *args: tp.Any, **kwargs: tp.Any) -> dx.Distribution:
+    def predict(self, *args: tp.Any, **kwargs: tp.Any) -> npd.Distribution:
         """Compute the latent function's multivariate normal distribution."""
         raise NotImplementedError
 
@@ -76,7 +76,7 @@ def __rmul__(self, other: AbstractLikelihood):
 
     def predict(
         self, params: dict
-    ) -> tp.Callable[[Float[Array, "N D"]], dx.Distribution]:
+    ) -> tp.Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Compute the GP's prior mean and variance.
         Args:
             params (dict): The specific set of parameters for which the mean function should be defined for.
@@ -84,16 +84,13 @@ def predict(
             tp.Callable[[Array], Array]: A mean function that accepts an input array for where the mean function should be evaluated at. The mean function's value at these points is then returned.
         """
 
-        def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
+        def predict_fn(test_inputs: Float[Array, "N D"]) -> npd.Distribution:
             t = test_inputs
             n_test = t.shape[0]
             μt = self.mean_function(t, params["mean_function"])
             Ktt = gram(self.kernel, t, params["kernel"])
             Ktt += I(n_test) * self.jitter
-
-            return dx.MultivariateNormalFullCovariance(
-                jnp.atleast_1d(μt.squeeze()), Ktt
-            )
+            return npd.MultivariateNormal(jnp.atleast_1d(μt.squeeze()), Ktt)
 
         return predict_fn
 
@@ -118,7 +115,7 @@ class AbstractPosterior(AbstractGP):
     jitter: tp.Optional[float] = DEFAULT_JITTER
 
     @abstractmethod
-    def predict(self, *args: tp.Any, **kwargs: tp.Any) -> dx.Distribution:
+    def predict(self, *args: tp.Any, **kwargs: tp.Any) -> npd.Distribution:
         """Predict the GP's output given the input."""
         raise NotImplementedError
 
@@ -141,7 +138,7 @@ class ConjugatePosterior(AbstractPosterior):
 
     def predict(
         self, train_data: Dataset, params: dict
-    ) -> tp.Callable[[Float[Array, "N D"]], dx.Distribution]:
+    ) -> tp.Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Conditional on a set of training data, compute the GP's posterior predictive distribution for a given set of parameters. The returned function can be evaluated at a set of test inputs to compute the corresponding predictive density.
 
         Args:
@@ -149,7 +146,7 @@ def predict(
             params (dict): A dictionary of parameters that should be used to compute the posterior.
 
         Returns:
-            tp.Callable[[Array], dx.Distribution]: A function that accepts an input array and returns the predictive distribution as a `distrax.MultivariateNormalFullCovariance`.
+            tp.Callable[[Array], npd.Distribution]: A function that accepts an input array and returns the predictive distribution as a `numpyro.distributions.MultivariateNormal`.
         """
         x, y, n = train_data.X, train_data.y, train_data.n
 
@@ -168,7 +165,7 @@ def predict(
         # w = L⁻¹ (y - μx)
         w = jsp.linalg.solve_triangular(L, y - μx, lower=True)
 
-        def predict(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
+        def predict(test_inputs: Float[Array, "N D"]) -> npd.Distribution:
             t = test_inputs
             n_test = t.shape[0]
             μt = self.prior.mean_function(t, params["mean_function"])
@@ -185,9 +182,7 @@ def predict(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
             covariance = Ktt - jnp.matmul(L_inv_Kxt.T, L_inv_Kxt)
             covariance += I(n_test) * self.jitter
 
-            return dx.MultivariateNormalFullCovariance(
-                jnp.atleast_1d(mean.squeeze()), covariance
-            )
+            return npd.MultivariateNormal(jnp.atleast_1d(mean.squeeze()), covariance)
 
         return predict
 
@@ -227,8 +222,9 @@ def mll(
             L = jnp.linalg.cholesky(Sigma)
 
             # p(y | x, θ), where θ are the model hyperparameters:
-            marginal_likelihood = dx.MultivariateNormalTri(
-                jnp.atleast_1d(μx.squeeze()), L
+
+            marginal_likelihood = npd.MultivariateNormal(
+                jnp.atleast_1d(μx.squeeze()), scale_tril=L
             )
 
             # log p(θ)
@@ -263,7 +259,7 @@ def _initialise_params(self, key: jnp.DeviceArray) -> Dict:
 
     def predict(
         self, train_data: Dataset, params: dict
-    ) -> tp.Callable[[Float[Array, "N D"]], dx.Distribution]:
+    ) -> tp.Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Conditional on a set of training data, compute the GP's posterior predictive distribution for a given set of parameters. The returned function can be evaluated at a set of test inputs to compute the corresponding predictive density. Note, to gain predictions on the scale of the original data, the returned distribution will need to be transformed through the likelihood function's inverse link function.
 
         Args:
@@ -271,7 +267,7 @@ def predict(
             params (dict): A dictionary of parameters that should be used to compute the posterior.
 
         Returns:
-            tp.Callable[[Array], dx.Distribution]: A function that accepts an input array and returns the predictive distribution as a `distrax.MultivariateNormalFullCovariance`.
+            tp.Callable[[Array], npd.Distribution]: A function that accepts an input array and returns the predictive distribution as a `numpyro.distributions.MultivariateNormal`.
         """
         x, n = train_data.X, train_data.n
 
@@ -279,7 +275,7 @@ def predict(
         Kxx += I(n) * self.jitter
         Lx = jnp.linalg.cholesky(Kxx)
 
-        def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
+        def predict_fn(test_inputs: Float[Array, "N D"]) -> npd.Distribution:
             t = test_inputs
             n_test = t.shape[0]
             Ktx = cross_covariance(self.prior.kernel, t, x, params["kernel"])
@@ -296,9 +292,7 @@ def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
             covariance = Ktt - jnp.matmul(Lx_inv_Kxt.T, Lx_inv_Kxt)
             covariance += I(n_test) * self.jitter
 
-            return dx.MultivariateNormalFullCovariance(
-                jnp.atleast_1d(mean.squeeze()), covariance
-            )
+            return npd.MultivariateNormal(jnp.atleast_1d(mean.squeeze()), covariance)
 
         return predict_fn
 
@@ -324,7 +318,7 @@ def marginal_log_likelihood(
 
         if not priors:
             priors = copy_dict_structure(self._initialise_params(jr.PRNGKey(0)))
-            priors["latent"] = dx.Normal(loc=0.0, scale=1.0)
+            priors["latent"] = npd.Normal(loc=0.0, scale=1.0)
 
         def mll(params: dict):
             params = transform(params=params, transform_map=transformations)
diff --git a/gpjax/likelihoods.py b/gpjax/likelihoods.py
index 2793126a..e9ac93c4 100644
--- a/gpjax/likelihoods.py
+++ b/gpjax/likelihoods.py
@@ -1,9 +1,9 @@
 import abc
 from typing import Any, Callable, Dict, Optional
 
-import distrax as dx
 import jax.numpy as jnp
 import jax.scipy as jsp
+import numpyro.distributions as npd
 from chex import dataclass
 from jaxtyping import Array, Float
 
@@ -66,16 +66,18 @@ def link_function(self) -> Callable:
             Callable: A link function that maps the predictive distribution to the likelihood function.
         """
 
-        def link_fn(x, params: dict) -> dx.Distribution:
-            return dx.Normal(loc=x, scale=params["obs_noise"])
+        def link_fn(x, params: dict) -> npd.Distribution:
+            return npd.Normal(loc=x, scale=params["obs_noise"])
 
         return link_fn
 
-    def predict(self, dist: dx.Distribution, params: dict) -> dx.Distribution:
+    def predict(self, dist: npd.Distribution, params: dict) -> npd.Distribution:
         """Evaluate the Gaussian likelihood function at a given predictive distribution. Computationally, this is equivalent to summing the observation noise term to the diagonal elements of the predictive distribution's covariance matrix.."""
         n_data = dist.event_shape[0]
-        noisy_cov = dist.covariance() + I(n_data) * params["likelihood"]["obs_noise"]
-        return dx.MultivariateNormalFullCovariance(dist.mean(), noisy_cov)
+        noisy_cov = (
+            dist.covariance_matrix + I(n_data) * params["likelihood"]["obs_noise"]
+        )
+        return npd.MultivariateNormal(dist.mean, noisy_cov)
 
 
 @dataclass
@@ -94,8 +96,8 @@ def link_function(self) -> Callable:
             Callable: A probit link function that maps the predictive distribution to the likelihood function.
         """
 
-        def link_fn(x, params: Dict) -> dx.Distribution:
-            return dx.Bernoulli(probs=inv_probit(x))
+        def link_fn(x, params: Dict) -> npd.Distribution:
+            return npd.Bernoulli(probs=inv_probit(x))
 
         return link_fn
 
@@ -115,9 +117,9 @@ def moment_fn(
 
         return moment_fn
 
-    def predict(self, dist: dx.Distribution, params: dict) -> Any:
-        variance = jnp.diag(dist.covariance())
-        mean = dist.mean()
+    def predict(self, dist: npd.Distribution, params: dict) -> Any:
+        variance = jnp.diag(dist.covariance_matrix)
+        mean = dist.mean
         return self.predictive_moment_fn(mean.ravel(), variance, params)
 
 
diff --git a/gpjax/parameters.py b/gpjax/parameters.py
index 889c90e7..b44c7e5d 100644
--- a/gpjax/parameters.py
+++ b/gpjax/parameters.py
@@ -4,10 +4,10 @@
 from copy import deepcopy
 from warnings import warn
 
-import distrax as dx
 import jax
 import jax.numpy as jnp
 import jax.random as jr
+import numpyro.distributions as npd
 from chex import dataclass
 from jaxtyping import Array, Float
 
@@ -15,8 +15,6 @@
 from .types import PRNGKeyType
 from .utils import merge_dictionaries
 
-Identity = dx.Lambda(forward=lambda x: x, inverse=lambda x: x)
-
 
 ################################
 # Base operations
@@ -134,7 +132,7 @@ def recursive_bijectors(ps, bs) -> tp.Tuple[tp.Dict, tp.Dict]:
                         transform_type = transform_set[key]
                         bijector = transform_set[transform_type]
                     else:
-                        bijector = Identity
+                        bijector = npd.transforms.IdentityTransform()
                         warnings.warn(
                             f"Parameter {key} has no transform. Defaulting to identity transfom."
                         )
@@ -155,10 +153,10 @@ def build_transforms(params: tp.Dict) -> tp.Tuple[tp.Dict, tp.Dict]:
     """
 
     def forward(bijector):
-        return bijector.forward
+        return bijector.__call__
 
     def inverse(bijector):
-        return bijector.inverse
+        return bijector.inv
 
     bijectors = build_bijectors(params)
 
@@ -198,7 +196,7 @@ def transform(params: tp.Dict, transform_map: tp.Dict) -> tp.Dict:
 # Priors
 ################################
 def log_density(
-    param: Float[Array, "D"], density: dx.Distribution
+    param: Float[Array, "D"], density: npd.Distribution
 ) -> Float[Array, "1"]:
     if type(density) == type(None):
         log_prob = jnp.array(0.0)
@@ -263,9 +261,9 @@ def prior_checks(priors: dict) -> dict:
                 )
         else:
             warnings.warn("Placing unit Gaussian prior on latent function.")
-            priors["latent"] = dx.Normal(loc=0.0, scale=1.0)
+            priors["latent"] = npd.Normal(loc=0.0, scale=1.0)
     else:
-        priors["latent"] = dx.Normal(loc=0.0, scale=1.0)
+        priors["latent"] = npd.Normal(loc=0.0, scale=1.0)
 
     return priors
 

From 11b0d9e9700ceae8b78474b9660dcf7f272f3319 Mon Sep 17 00:00:00 2001
From: Thomas Pinder <tompinder@live.co.uk>
Date: Thu, 13 Oct 2022 10:56:35 +0000
Subject: [PATCH 02/14] Update numpyro

---
 examples/barycentres.pct.py    |  17 +++--
 examples/classification.pct.py |  30 ++++-----
 examples/collapsed_vi.pct.py   |  27 ++++----
 examples/regression.pct.py     |  26 +++-----
 examples/uncollapsed_vi.pct.py |  18 +++--
 gpjax/__init__.py              |   2 +-
 gpjax/config.py                |  10 ---
 gpjax/gps.py                   |   2 +-
 gpjax/likelihoods.py           |   4 +-
 gpjax/variational_families.py  | 118 +++++++++++++++------------------
 gpjax/variational_inference.py |  19 ++----
 setup.py                       |   2 +-
 12 files changed, 116 insertions(+), 159 deletions(-)

diff --git a/examples/barycentres.pct.py b/examples/barycentres.pct.py
index b24c36cc..1a3e3c21 100644
--- a/examples/barycentres.pct.py
+++ b/examples/barycentres.pct.py
@@ -100,24 +100,23 @@ def fit_gp(x: jnp.DeviceArray, y: jnp.DeviceArray):
     D = gpx.Dataset(X=x, y=y)
     likelihood = gpx.Gaussian(num_datapoints=n)
     posterior = gpx.Prior(kernel=gpx.RBF()) * likelihood
-    params, trainables, constrainers, unconstrainers = gpx.initialise(
+    parameter_state = gpx.initialise(
         posterior, key
-    ).unpack()
-    params = gpx.transform(params, unconstrainers)
+    )
+    params, trainables, bijectors = parameter_state.unpack()
+    params = gpx.unconstrain(params, bijectors)
 
     objective = jax.jit(
-        posterior.marginal_log_likelihood(D, constrainers, negative=True)
+        posterior.marginal_log_likelihood(D, negative=True)
     )
 
     opt = ox.adam(learning_rate=0.01)
     learned_params, training_history = gpx.fit(
         objective=objective,
-        trainables=trainables,
-        params=params,
+        parameter_state=parameter_state,
         optax_optim=opt,
         n_iters=1000,
     ).unpack()
-    learned_params = gpx.transform(learned_params, constrainers)
     return likelihood(posterior(D, learned_params)(xtest), learned_params)
 
 
@@ -135,7 +134,7 @@ def sqrtm(A: jnp.DeviceArray):
 
 
 def wasserstein_barycentres(
-    distributions: tp.List[dx.Distribution], weights: jnp.DeviceArray
+    distributions: tp.List[numpyro.distributions.Distribution], weights: jnp.DeviceArray
 ):
     covariances = [d.covariance_matrix for d in distributions]
     cov_stack = jnp.stack(covariances)
@@ -180,7 +179,7 @@ def step(covariance_candidate: jnp.DeviceArray, i: jnp.DeviceArray):
 
 # %%
 def plot(
-    dist: dx.Distribution,
+    dist: numpyro.distributions.Distribution,
     ax,
     color: str = "tab:blue",
     label: str = None,
diff --git a/examples/classification.pct.py b/examples/classification.pct.py
index 876da984..9e22f3e7 100644
--- a/examples/classification.pct.py
+++ b/examples/classification.pct.py
@@ -19,8 +19,6 @@
 #
 # In this notebook we demonstrate how to perform inference for Gaussian process models with non-Gaussian likelihoods via maximum a posteriori (MAP) and Markov chain Monte Carlo (MCMC). We focus on a classification task here and use [BlackJax](https://github.com/blackjax-devs/blackjax/) for sampling.
 
-import blackjax
-
 # %%
 import jax
 import jax.numpy as jnp
@@ -30,7 +28,7 @@
 import numpyro
 import optax as ox
 from jaxtyping import Array, Float
-
+import blackjax
 import gpjax as gpx
 from gpjax.utils import I
 
@@ -72,34 +70,31 @@
 # To begin we obtain a set of initial parameter values through the `initialise` callable, and transform these to the unconstrained space via `transform` (see the [regression notebook](https://gpjax.readthedocs.io/en/latest/nbs/regression.html)). We also define the negative marginal log-likelihood, and JIT compile this to accelerate training.
 # %%
 parameter_state = gpx.initialise(posterior)
-params, trainable, constrainer, unconstrainer = parameter_state.unpack()
-params = gpx.transform(params, unconstrainer)
+params, trainable, bijectors = parameter_state.unpack()
+params = gpx.unconstrain(params, bijectors)
 
-mll = jax.jit(posterior.marginal_log_likelihood(D, constrainer, negative=True))
+mll = jax.jit(posterior.marginal_log_likelihood(D, negative=True))
 
 # %% [markdown]
 # We can obtain a MAP estimate by optimising the marginal log-likelihood with Obtax's optimisers.
 # %%
 opt = ox.adam(learning_rate=0.01)
-unconstrained_params, training_history = gpx.fit(
+learned_params, training_history = gpx.fit(
     mll,
-    params,
-    trainable,
+    parameter_state,
     opt,
     n_iters=500,
 ).unpack()
 
-negative_Hessian = jax.jacfwd(jax.jacrev(mll))(unconstrained_params)["latent"][
+negative_Hessian = jax.jacfwd(jax.jacrev(mll))(learned_params)["latent"][
     "latent"
 ][:, 0, :, 0]
-
-map_estimate = gpx.transform(unconstrained_params, constrainer)
 # %% [markdown]
 # From which we can make predictions at novel inputs, as illustrated below.
 # %%
-latent_dist = posterior(D, map_estimate)(xtest)
+latent_dist = posterior(D, learned_params)(xtest)
 
-predictive_dist = likelihood(latent_dist, map_estimate)
+predictive_dist = likelihood(latent_dist, learned_params)
 
 predictive_mean = predictive_dist.mean
 predictive_std = jnp.sqrt(predictive_dist.variance)
@@ -139,7 +134,7 @@
 # The Laplace approximation improves uncertainty quantification by incorporating curvature induced by the marginal log-likelihood's Hessian to construct an approximate Gaussian distribution centered on the MAP estimate.
 # Since the negative Hessian is positive definite, we can use the Cholesky decomposition to obtain the covariance matrix of the Laplace approximation at the datapoints below.
 # %%
-f_map_estimate = posterior(D, map_estimate)(x).mean
+f_map_estimate = posterior(D, learned_params)(x).mean
 
 jitter = 1e-6
 
@@ -224,7 +219,7 @@ def predict(
 # %%
 latent_dist = predict(laplace_approximation, D, xtest)
 
-predictive_dist = likelihood(latent_dist, map_estimate)
+predictive_dist = likelihood(latent_dist, learned_params)
 
 predictive_mean = predictive_dist.mean
 predictive_std = predictive_dist.variance**0.5
@@ -276,7 +271,7 @@ def predict(
 num_adapt = 500
 num_samples = 200
 
-mll = jax.jit(posterior.marginal_log_likelihood(D, constrainer, negative=False))
+mll = jax.jit(posterior.marginal_log_likelihood(D, negative=False))
 
 adapt = blackjax.window_adaptation(
     blackjax.nuts, mll, num_adapt, target_acceptance_rate=0.65
@@ -334,7 +329,6 @@ def one_step(state, rng_key):
     ps["kernel"]["lengthscale"] = states.position["kernel"]["lengthscale"][i]
     ps["kernel"]["variance"] = states.position["kernel"]["variance"][i]
     ps["latent"] = states.position["latent"][i, :, :]
-    ps = gpx.transform(ps, constrainer)
 
     latent_dist = posterior(D, ps)(xtest)
     predictive_dist = likelihood(latent_dist, ps)
diff --git a/examples/collapsed_vi.pct.py b/examples/collapsed_vi.pct.py
index f80431b2..fe48d87b 100644
--- a/examples/collapsed_vi.pct.py
+++ b/examples/collapsed_vi.pct.py
@@ -6,7 +6,7 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.11.5
+#       jupytext_version: 1.11.2
 #   kernelspec:
 #     display_name: Python 3.9.7 ('gpjax')
 #     language: python
@@ -86,22 +86,21 @@
 # %% [markdown]
 # We now train our model akin to a Gaussian process regression model via the `fit` abstraction. Unlike the regression example given in the [conjugate regression notebook](https://gpjax.readthedocs.io/en/latest/nbs/regression.html), the inducing locations that induce our variational posterior distribution are now part of the model's parameters. Using a gradient-based optimiser, we can then _optimise_ their location such that the evidence lower bound is maximised.
 # %%
-params, trainables, constrainers, unconstrainers = gpx.initialise(sgpr, key).unpack()
+parameter_state = gpx.initialise(sgpr, key)
+params, trainables, bijectors = parameter_state.unpack()
 
-loss_fn = jit(sgpr.elbo(D, constrainers, negative=True))
+loss_fn = jit(sgpr.elbo(D, negative=True))
 
 optimiser = ox.adam(learning_rate=0.005)
 
-params = gpx.transform(params, unconstrainers)
+params = gpx.unconstrain(params, bijectors)
 
 learned_params, training_history = gpx.fit(
     objective=loss_fn,
-    params=params,
-    trainables=trainables,
+    parameter_state=parameter_state,
     optax_optim=optimiser,
     n_iters=2000,
 ).unpack()
-learned_params = gpx.transform(learned_params, constrainers)
 
 # %% [markdown]
 # We show predictions of our model with the learned inducing points overlayed in grey.
@@ -109,10 +108,10 @@
 latent_dist = q.predict(D, learned_params)(xtest)
 predictive_dist = likelihood(latent_dist, learned_params)
 
-samples = latent_dist.sample(seed=key, sample_shape=20)
+samples = latent_dist.sample(key, sample_shape=(20, ))
 
-predictive_mean = predictive_dist.mean()
-predictive_std = predictive_dist.stddev()
+predictive_mean = predictive_dist.mean
+predictive_std = jnp.sqrt(predictive_dist.variance)
 
 fig, ax = plt.subplots(figsize=(12, 5))
 
@@ -166,15 +165,15 @@
 
 # %%
 full_rank_model = gpx.Prior(kernel=gpx.RBF()) * gpx.Gaussian(num_datapoints=D.n)
-fr_params, fr_trainables, fr_constrainers, fr_unconstrainers = gpx.initialise(
+fr_params, fr_trainables, fr_bijectors = gpx.initialise(
     full_rank_model, key
 ).unpack()
-fr_params = gpx.transform(fr_params, fr_unconstrainers)
-mll = jit(full_rank_model.marginal_log_likelihood(D, fr_constrainers, negative=True))
+fr_params = gpx.unconstrain(fr_params, fr_bijectors)
+mll = jit(full_rank_model.marginal_log_likelihood(D, negative=True))
 # %timeit mll(fr_params).block_until_ready()
 
 # %%
-sparse_elbo = jit(sgpr.elbo(D, constrainers, negative=True))
+sparse_elbo = jit(sgpr.elbo(D, negative=True))
 # %timeit sparse_elbo(params).block_until_ready()
 
 # %% [markdown]
diff --git a/examples/regression.pct.py b/examples/regression.pct.py
index 51735685..1e976025 100644
--- a/examples/regression.pct.py
+++ b/examples/regression.pct.py
@@ -146,20 +146,21 @@
 # We can now unpack the `ParameterState` to receive each of the four components listed above.
 
 # %%
-params, trainable, constrainer, unconstrainer = parameter_state.unpack()
+params, trainable, bijectors = parameter_state.unpack()
 pp.pprint(params)
 
 # %% [markdown]
 # To motivate the purpose of `constrainer` and `unconstrainer` more precisely, notice that our model hyperparameters $\{\ell^2, \sigma^2, \alpha^2 \}$ are all strictly positive. To ensure more stable optimisation, it is strongly advised to transform the parameters onto an unconstrained space first via `transform`.
 
 # %%
-params = gpx.transform(params, unconstrainer)
+# params = gpx.constrain(params, bijectors)
+pp.pprint(params)
 
 # %% [markdown]
 # To train our hyperparameters, we optimising the marginal log-likelihood of the posterior with respect to them. We define the marginal log-likelihood with `marginal_log_likelihood` on the posterior.
 
 # %%
-mll = jit(posterior.marginal_log_likelihood(D, constrainer, negative=True))
+mll = jit(posterior.marginal_log_likelihood(D, negative=True))
 mll(params)
 # %% [markdown]
 # Since most optimisers (including here) minimise a given function, we have realised the negative marginal log-likelihood and just-in-time (JIT) compiled this to accelerate training.
@@ -170,10 +171,9 @@
 # %%
 opt = ox.adam(learning_rate=0.01)
 inference_state = gpx.fit(
-    mll,
-    params,
-    trainable,
-    opt,
+    objective = mll,
+    parameter_state = parameter_state,
+    optax_optim = opt,
     n_iters=500,
 )
 
@@ -182,23 +182,13 @@
 
 # %%
 final_params, training_history = inference_state.unpack()
-
-# %% [markdown]
-#
-# The exact value of our learned parameters is often useful in answering certain questions about the underlying process. To obtain these values, we untransfom our trained unconstrained parameters back to their original constrained space with `transform` and `constrainer`.
-
-# %%
-final_params = gpx.transform(final_params, constrainer)
-pp.pprint(final_params)
+final_params
 
 # %% [markdown]
 # ## Prediction
 #
 # Equipped with the posterior and a set of optimised hyperparameter values, we are now in a position to query our GP's predictive distribution at novel test inputs. To do this, we use our defined `posterior` and `likelihood` at our test inputs to obtain the predictive distribution as a `Distrax` multivariate Gaussian upon which `mean` and `stddev` can be used to extract the predictive mean and standard deviatation.
 
-# %%
-latent_dist
-
 # %%
 latent_dist = posterior(D, final_params)(xtest)
 predictive_dist = likelihood(latent_dist, final_params)
diff --git a/examples/uncollapsed_vi.pct.py b/examples/uncollapsed_vi.pct.py
index 6ad8d60a..fca98678 100644
--- a/examples/uncollapsed_vi.pct.py
+++ b/examples/uncollapsed_vi.pct.py
@@ -7,9 +7,9 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.11.5
+#       jupytext_version: 1.11.2
 #   kernelspec:
-#     display_name: Python 3.10.0 ('base')
+#     display_name: Python 3.9.7 ('gpjax')
 #     language: python
 #     name: python3
 # ---
@@ -137,10 +137,15 @@
 #
 # Since Optax's optimisers work to minimise functions, to maximise the ELBO we return its negative.
 # %%
-params, trainables, constrainers, unconstrainers = gpx.initialise(svgp, key).unpack()
-params = gpx.transform(params, unconstrainers)
+parameter_state = gpx.initialise(svgp, key)
+params, trainables, bijectors = parameter_state.unpack()
+params = gpx.unconstrain(params, bijectors)
 
-loss_fn = jit(svgp.elbo(D, constrainers, negative=True))
+loss_fn = svgp.elbo(D, negative=True)
+
+# %%
+b = gpx.abstractions.get_batch(D, 64, key)
+loss_fn(params, b)
 
 # %% [markdown]
 # ### Mini-batching
@@ -152,8 +157,7 @@
 
 inference_state = gpx.fit_batches(
     objective=loss_fn,
-    params=params,
-    trainables=trainables,
+    parameter_state=parameter_state,
     train_data=D,
     optax_optim=optimiser,
     n_iters=4000,
diff --git a/gpjax/__init__.py b/gpjax/__init__.py
index 661c5cb8..2e83fc69 100644
--- a/gpjax/__init__.py
+++ b/gpjax/__init__.py
@@ -47,7 +47,7 @@
 __description__ = "Didactic Gaussian processes in JAX"
 __url__ = "https://github.com/thomaspinder/GPJax"
 __contributors__ = "https://github.com/thomaspinder/GPJax/graphs/contributors"
-__version__ = "0.4.13"
+__version__ = "0.5.0"
 
 
 __all__ = [
diff --git a/gpjax/config.py b/gpjax/config.py
index da585a9b..fa7be8fa 100644
--- a/gpjax/config.py
+++ b/gpjax/config.py
@@ -6,16 +6,6 @@
 import distrax as dx
 import jax.numpy as jnp
 
-Softplus = dx.Lambda(
-    forward=lambda x: jnp.log(1 + jnp.exp(x)),
-    inverse=lambda x: jnp.log(jnp.exp(x) - 1.0),
-)
-
-# TODO: Remove this once 'FillTriangular' is added to Distrax.
-FillTriangular = dx.Chain([tfb.FillTriangular()])
-
-Identity = dx.Lambda(forward=lambda x: x, inverse=lambda x: x)
-
 
 def get_defaults() -> ConfigDict:
     """Construct and globally register the config file used within GPJax.
diff --git a/gpjax/gps.py b/gpjax/gps.py
index 7ac672b4..4d23a1c7 100644
--- a/gpjax/gps.py
+++ b/gpjax/gps.py
@@ -39,7 +39,7 @@
 class AbstractGP:
     """Abstract Gaussian process object."""
 
-    def __call__(self, *args: tp.Any, **kwargs: tp.Any) -> npd.Distribution:
+    def __call__(self, *args: Any, **kwargs: Any) -> npd.Distribution:
         """Evaluate the Gaussian process at the given points.
 
         Args:
diff --git a/gpjax/likelihoods.py b/gpjax/likelihoods.py
index e58d2d3d..0585ba96 100644
--- a/gpjax/likelihoods.py
+++ b/gpjax/likelihoods.py
@@ -33,7 +33,7 @@ class AbstractLikelihood:
     num_datapoints: int  # The number of datapoints that the likelihood factorises over.
     name: Optional[str] = "Likelihood"
 
-    def __call__(self, *args: Any, **kwargs: Any) -> dx.Distribution:
+    def __call__(self, *args: Any, **kwargs: Any) -> npd.Distribution:
         """Evaluate the likelihood function at a given predictive distribution.
 
         Args:
@@ -46,7 +46,7 @@ def __call__(self, *args: Any, **kwargs: Any) -> dx.Distribution:
         return self.predict(*args, **kwargs)
 
     @abc.abstractmethod
-    def predict(self, *args: Any, **kwargs: Any) -> dx.Distribution:
+    def predict(self, *args: Any, **kwargs: Any) -> npd.Distribution:
         """Evaluate the likelihood function at a given predictive distribution.
 
         Args:
diff --git a/gpjax/variational_families.py b/gpjax/variational_families.py
index 0a939fd8..dc02fa94 100644
--- a/gpjax/variational_families.py
+++ b/gpjax/variational_families.py
@@ -16,11 +16,11 @@
 import abc
 from typing import Any, Callable, Dict, Optional
 
-import distrax as dx
 import jax.numpy as jnp
 import jax.scipy as jsp
 from chex import dataclass
 from jaxtyping import Array, Float
+import numpyro.distributions as npd
 
 from .config import get_defaults
 from .gps import Prior
@@ -32,11 +32,21 @@
 DEFAULT_JITTER = get_defaults()["jitter"]
 
 
+def kl_divergence(p, q):
+    N = p.mean.shape[0]
+    iS1 = jnp.linalg.inv(q.covariance_matrix)
+    diff = q.mean - p.mean
+    tr_term = jnp.trace(jnp.matmul(iS1, p.covariance_matrix))
+    det_term = jnp.log(jnp.linalg.det(q.covariance_matrix) / jnp.linalg.det(p.covariance_matrix))
+    quad_term = diff.T @ jnp.linalg.inv(q.covariance_matrix) @ diff
+    return 0.5 * (tr_term + det_term + quad_term - N)
+
+
 @dataclass
 class AbstractVariationalFamily:
     """Abstract base class used to represent families of distributions that can be used within variational inference."""
 
-    def __call__(self, *args: Any, **kwargs: Any) -> dx.Distribution:
+    def __call__(self, *args: Any, **kwargs: Any) -> npd.Distribution:
         """For a given set of parameters, compute the latent function's prediction under the variational approximation.
 
         Args:
@@ -61,7 +71,7 @@ def _initialise_params(self, key: PRNGKeyType) -> Dict:
         raise NotImplementedError
 
     @abc.abstractmethod
-    def predict(self, *args: Any, **kwargs: Any) -> dx.Distribution:
+    def predict(self, *args: Any, **kwargs: Any) -> npd.Distribution:
         """Predict the GP's output given the input.
 
         Args:
@@ -142,12 +152,11 @@ def prior_kl(self, params: Dict) -> Float[Array, "1"]:
         Kzz += I(m) * self.jitter
         Lz = jnp.linalg.cholesky(Kzz)
 
-        qu = dx.MultivariateNormalTri(jnp.atleast_1d(mu.squeeze()), sqrt)
-        pu = dx.MultivariateNormalTri(jnp.atleast_1d(μz.squeeze()), Lz)
-
-        return qu.kl_divergence(pu)
+        qu = npd.MultivariateNormal(jnp.atleast_1d(mu.squeeze()), scale_tril=sqrt)
+        pu = npd.MultivariateNormal(jnp.atleast_1d(μz.squeeze()), scale_tril=Lz)
+        return kl_divergence(qu, pu)
 
-    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distribution]:
+    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Compute the predictive distribution of the GP at the test inputs t.
 
         This is the integral q(f(t)) = ∫ p(f(t)|u) q(u) du, which can be computed in closed form as:
@@ -158,7 +167,7 @@ def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distributi
             params (Dict): The set of parameters that are to be used to parameterise our variational approximation and GP.
 
         Returns:
-            Callable[[Float[Array, "N D"]], dx.Distribution]: A function that accepts a set of test points and will return the predictive distribution at those points.
+            Callable[[Float[Array, "N D"]], npd.Distribution]: A function that accepts a set of test points and will return the predictive distribution at those points.
         """
         mu = params["variational_family"]["moments"]["variational_mean"]
         sqrt = params["variational_family"]["moments"]["variational_root_covariance"]
@@ -170,7 +179,7 @@ def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distributi
         Lz = jnp.linalg.cholesky(Kzz)
         μz = self.prior.mean_function(z, params["mean_function"])
 
-        def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
+        def predict_fn(test_inputs: Float[Array, "N D"]) -> npd.Distribution:
             t = test_inputs
             n_test = t.shape[0]
             Ktt = gram(self.prior.kernel, t, params["kernel"])
@@ -197,9 +206,7 @@ def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
             )
             covariance += I(n_test) * self.jitter
 
-            return dx.MultivariateNormalFullCovariance(
-                jnp.atleast_1d(mean.squeeze()), covariance
-            )
+            return npd.MultivariateNormal(jnp.atleast_1d(mean.squeeze()), covariance)
 
         return predict_fn
 
@@ -230,12 +237,13 @@ def prior_kl(self, params: Dict) -> Float[Array, "1"]:
         sqrt = params["variational_family"]["moments"]["variational_root_covariance"]
         m = self.num_inducing
 
-        qu = dx.MultivariateNormalTri(jnp.atleast_1d(mu.squeeze()), sqrt)
-        pu = dx.MultivariateNormalDiag(jnp.zeros(m))
-
-        return qu.kl_divergence(pu)
+        qu = npd.MultivariateNormal(jnp.atleast_1d(mu.squeeze()), scale_tril=sqrt)
+        pu = npd.MultivariateNormal(
+            jnp.zeros(m), jnp.eye(m)
+        )  # TODO: Implement diagonal MVGaussian.
+        return kl_divergence(qu, pu)
 
-    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distribution]:
+    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Compute the predictive distribution of the GP at the test inputs t.
 
         This is the integral q(f(t)) = ∫ p(f(t)|u) q(u) du, which can be computed in closed form as
@@ -246,7 +254,7 @@ def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distributi
             params (Dict): The set of parameters that are to be used to parameterise our variational approximation and GP.
 
         Returns:
-            Callable[[Float[Array, "N D"]], dx.Distribution]: A function that accepts a set of test points and will return the predictive distribution at those points.
+            Callable[[Float[Array, "N D"]], npd.Distribution]: A function that accepts a set of test points and will return the predictive distribution at those points.
         """
         mu = params["variational_family"]["moments"]["variational_mean"]
         sqrt = params["variational_family"]["moments"]["variational_root_covariance"]
@@ -257,7 +265,7 @@ def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distributi
         Kzz += I(m) * self.jitter
         Lz = jnp.linalg.cholesky(Kzz)
 
-        def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
+        def predict_fn(test_inputs: Float[Array, "N D"]) -> npd.Distribution:
             t = test_inputs
             n_test = t.shape[0]
             Ktt = gram(self.prior.kernel, t, params["kernel"])
@@ -281,9 +289,7 @@ def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
             )
             covariance += I(n_test) * self.jitter
 
-            return dx.MultivariateNormalFullCovariance(
-                jnp.atleast_1d(mean.squeeze()), covariance
-            )
+            return npd.MultivariateNormal(jnp.atleast_1d(mean.squeeze()), covariance)
 
         return predict_fn
 
@@ -360,12 +366,12 @@ def prior_kl(self, params: Dict) -> Float[Array, "1"]:
         Kzz += I(m) * self.jitter
         Lz = jnp.linalg.cholesky(Kzz)
 
-        qu = dx.MultivariateNormalTri(jnp.atleast_1d(mu.squeeze()), sqrt)
-        pu = dx.MultivariateNormalTri(jnp.atleast_1d(μz.squeeze()), Lz)
+        qu = npd.MultivariateNormal(jnp.atleast_1d(mu.squeeze()), scale_tril=sqrt)
+        pu = npd.MultivariateNormal(jnp.atleast_1d(μz.squeeze()), scale_tril=Lz)
 
-        return qu.kl_divergence(pu)
+        return kl_divergence(qu, pu)
 
-    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distribution]:
+    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Compute the predictive distribution of the GP at the test inputs t.
 
         This is the integral q(f(t)) = ∫ p(f(t)|u) q(u) du, which can be computed in closed form as
@@ -378,7 +384,7 @@ def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distributi
             params (Dict): The set of parameters that are to be used to parameterise our variational approximation and GP.
 
         Returns:
-            Callable[[Float[Array, "N D"]], dx.Distribution]: A function that accepts a set of test points and will return the predictive distribution at those points.
+            Callable[[Float[Array, "N D"]], npd.Distribution]: A function that accepts a set of test points and will return the predictive distribution at those points.
         """
         natural_vector = params["variational_family"]["moments"]["natural_vector"]
         natural_matrix = params["variational_family"]["moments"]["natural_matrix"]
@@ -408,7 +414,7 @@ def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distributi
         Lz = jnp.linalg.cholesky(Kzz)
         μz = self.prior.mean_function(z, params["mean_function"])
 
-        def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
+        def predict_fn(test_inputs: Float[Array, "N D"]) -> npd.Distribution:
             t = test_inputs
             Ktt = gram(self.prior.kernel, t, params["kernel"])
             Kzt = cross_covariance(self.prior.kernel, z, t, params["kernel"])
@@ -433,9 +439,7 @@ def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
                 + jnp.matmul(Ktz_Kzz_inv_L, Ktz_Kzz_inv_L.T)
             )
 
-            return dx.MultivariateNormalFullCovariance(
-                jnp.atleast_1d(mean.squeeze()), covariance
-            )
+            return npd.MultivariateNormal(jnp.atleast_1d(mean.squeeze()), covariance)
 
         return predict_fn
 
@@ -486,12 +490,8 @@ def prior_kl(self, params: Dict) -> Float[Array, "1"]:
         Returns:
             Float[Array, "1"]: The KL-divergence between our variational approximation and the GP prior.
         """
-        expectation_vector = params["variational_family"]["moments"][
-            "expectation_vector"
-        ]
-        expectation_matrix = params["variational_family"]["moments"][
-            "expectation_matrix"
-        ]
+        expectation_vector = params["variational_family"]["moments"]["expectation_vector"]
+        expectation_matrix = params["variational_family"]["moments"]["expectation_matrix"]
         z = params["variational_family"]["inducing_inputs"]
         m = self.num_inducing
 
@@ -510,12 +510,12 @@ def prior_kl(self, params: Dict) -> Float[Array, "1"]:
         Kzz += I(m) * self.jitter
         Lz = jnp.linalg.cholesky(Kzz)
 
-        qu = dx.MultivariateNormalTri(jnp.atleast_1d(mu.squeeze()), sqrt)
-        pu = dx.MultivariateNormalTri(jnp.atleast_1d(μz.squeeze()), Lz)
+        qu = npd.MultivariateNormal(jnp.atleast_1d(mu.squeeze()), scale_tril=sqrt)
+        pu = npd.MultivariateNormal(jnp.atleast_1d(μz.squeeze()), scale_tril=Lz)
 
-        return qu.kl_divergence(pu)
+        return kl_divergence(qu, pu)
 
-    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distribution]:
+    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Compute the predictive distribution of the GP at the test inputs t.
 
         This is the integral q(f(t)) = ∫ p(f(t)|u) q(u) du, which can be computed in closed form as
@@ -528,14 +528,10 @@ def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distributi
             params (Dict): The set of parameters that are to be used to parameterise our variational approximation and GP.
 
         Returns:
-            Callable[[Float[Array, "N D"]], dx.Distribution]: A function that accepts a set of test points and will return the predictive distribution at those points.
+            Callable[[Float[Array, "N D"]], npd.Distribution]: A function that accepts a set of test points and will return the predictive distribution at those points.
         """
-        expectation_vector = params["variational_family"]["moments"][
-            "expectation_vector"
-        ]
-        expectation_matrix = params["variational_family"]["moments"][
-            "expectation_matrix"
-        ]
+        expectation_vector = params["variational_family"]["moments"]["expectation_vector"]
+        expectation_matrix = params["variational_family"]["moments"]["expectation_matrix"]
         z = params["variational_family"]["inducing_inputs"]
         m = self.num_inducing
 
@@ -554,7 +550,7 @@ def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], dx.Distributi
         Lz = jnp.linalg.cholesky(Kzz)
         μz = self.prior.mean_function(z, params["mean_function"])
 
-        def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
+        def predict_fn(test_inputs: Float[Array, "N D"]) -> npd.Distribution:
             t = test_inputs
             Ktt = gram(self.prior.kernel, t, params["kernel"])
             Kzt = cross_covariance(self.prior.kernel, z, t, params["kernel"])
@@ -579,9 +575,7 @@ def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
                 + jnp.matmul(Ktz_Kzz_inv_sqrt, Ktz_Kzz_inv_sqrt.T)
             )
 
-            return dx.MultivariateNormalFullCovariance(
-                jnp.atleast_1d(mean.squeeze()), covariance
-            )
+            return npd.MultivariateNormal(jnp.atleast_1d(mean.squeeze()), covariance)
 
         return predict_fn
 
@@ -611,20 +605,18 @@ def _initialise_params(self, key: PRNGKeyType) -> Dict:
             self.prior._initialise_params(key),
             {
                 "variational_family": {"inducing_inputs": self.inducing_inputs},
-                "likelihood": {
-                    "obs_noise": self.likelihood._initialise_params(key)["obs_noise"]
-                },
+                "likelihood": {"obs_noise": self.likelihood._initialise_params(key)["obs_noise"]},
             },
         )
 
     def predict(
         self, train_data: Dataset, params: Dict
-    ) -> Callable[[Float[Array, "N D"]], dx.Distribution]:
+    ) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Compute the predictive distribution of the GP at the test inputs.
         Args:
             params (Dict): The set of parameters that are to be used to parameterise our variational approximation and GP.
         Returns:
-            Callable[[Float[Array, "N D"]], dx.Distribution]: A function that accepts a set of test points and will return the predictive distribution at those points.
+            Callable[[Float[Array, "N D"]], npd.Distribution]: A function that accepts a set of test points and will return the predictive distribution at those points.
         """
         x, y = train_data.X, train_data.y
 
@@ -658,11 +650,9 @@ def predict(
         Lz_inv_Kzx_diff = jsp.linalg.cho_solve((L, True), jnp.matmul(Lz_inv_Kzx, diff))
 
         # Kzz⁻¹ Kzx (y - μx)
-        Kzz_inv_Kzx_diff = jsp.linalg.solve_triangular(
-            Lz.T, Lz_inv_Kzx_diff, lower=False
-        )
+        Kzz_inv_Kzx_diff = jsp.linalg.solve_triangular(Lz.T, Lz_inv_Kzx_diff, lower=False)
 
-        def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
+        def predict_fn(test_inputs: Float[Array, "N D"]) -> npd.Distribution:
             t = test_inputs
             Ktt = gram(self.prior.kernel, t, params["kernel"])
             Kzt = cross_covariance(self.prior.kernel, z, t, params["kernel"])
@@ -684,9 +674,7 @@ def predict_fn(test_inputs: Float[Array, "N D"]) -> dx.Distribution:
                 + jnp.matmul(L_inv_Lz_inv_Kzt.T, L_inv_Lz_inv_Kzt)
             )
 
-            return dx.MultivariateNormalFullCovariance(
-                jnp.atleast_1d(mean.squeeze()), covariance
-            )
+            return npd.MultivariateNormal(jnp.atleast_1d(mean.squeeze()), covariance)
 
         return predict_fn
 
diff --git a/gpjax/variational_inference.py b/gpjax/variational_inference.py
index 16e1e604..97ec0b19 100644
--- a/gpjax/variational_inference.py
+++ b/gpjax/variational_inference.py
@@ -104,9 +104,7 @@ def elbo_fn(params: Dict, batch: Dataset) -> Float[Array, "1"]:
 
         return elbo_fn
 
-    def variational_expectation(
-        self, params: Dict, batch: Dataset
-    ) -> Float[Array, "N 1"]:
+    def variational_expectation(self, params: Dict, batch: Dataset) -> Float[Array, "N 1"]:
         """Compute the expectation of our model's log-likelihood under our variational distribution. Batching can be done here to speed up computation.
 
         Args:
@@ -119,15 +117,14 @@ def variational_expectation(
         x, y = batch.X, batch.y
 
         # q(f(x))
+        print(self.variational_family.predict(params))
         predictive_dist = vmap(self.variational_family.predict(params))(x[:, None])
         mean = predictive_dist.mean().val.reshape(-1, 1)
         variance = predictive_dist.variance().val.reshape(-1, 1)
 
         # log(p(y|f(x)))
         log_prob = vmap(
-            lambda f, y: self.likelihood.link_function(
-                f, params["likelihood"]
-            ).log_prob(y)
+            lambda f, y: self.likelihood.link_function(f, params["likelihood"]).log_prob(y)
         )
 
         # ≈ ∫[log(p(y|f(x))) q(f(x))] df(x)
@@ -176,9 +173,7 @@ def elbo_fn(params: Dict) -> Float[Array, "1"]:
             Kzz = gram(self.prior.kernel, z, params["kernel"])
             Kzz += I(m) * self.variational_family.jitter
             Kzx = cross_covariance(self.prior.kernel, z, x, params["kernel"])
-            Kxx_diag = vmap(self.prior.kernel, in_axes=(0, 0, None))(
-                x, x, params["kernel"]
-            )
+            Kxx_diag = vmap(self.prior.kernel, in_axes=(0, 0, None))(x, x, params["kernel"])
             μx = self.prior.mean_function(x, params["mean_function"])
 
             Lz = jnp.linalg.cholesky(Kzz)
@@ -224,12 +219,10 @@ def elbo_fn(params: Dict) -> Float[Array, "1"]:
             diff = y - μx
 
             # L⁻¹ A (y - μx)
-            L_inv_A_diff = jsp.linalg.solve_triangular(
-                L, jnp.matmul(A, diff), lower=True
-            )
+            L_inv_A_diff = jsp.linalg.solve_triangular(L, jnp.matmul(A, diff), lower=True)
 
             # (y - μx)ᵀ (Iσ² + Q)⁻¹ (y - μx)
-            quad = (jnp.sum(diff**2) - jnp.sum(L_inv_A_diff**2)) / noise
+            quad = (jnp.sum(diff ** 2) - jnp.sum(L_inv_A_diff ** 2)) / noise
 
             # 2 * log N(y; μx, Iσ² + Q)
             two_log_prob = -n * jnp.log(2.0 * jnp.pi * noise) - log_det_B - quad
diff --git a/setup.py b/setup.py
index 79fa846c..13af1caa 100644
--- a/setup.py
+++ b/setup.py
@@ -16,7 +16,7 @@ def parse_requirements_file(filename):
     "jaxlib>=0.1.47",
     "optax",
     "chex",
-    "distrax>=0.1.2",
+    "numpyro",
     "tensorflow-probability>=0.16.0",
     "tqdm>=4.0.0",
     "ml-collections==0.1.0",

From b40f883bf59f28b86980b8406a1d6bf046bc01a2 Mon Sep 17 00:00:00 2001
From: Daniel Dodd <daniel_dodd@icloud.com>
Date: Fri, 14 Oct 2022 14:57:40 +0100
Subject: [PATCH 03/14] Improve kl divergence efficiency.

---
 gpjax/variational_families.py | 139 +++++++++++++++++++++++++++-------
 1 file changed, 110 insertions(+), 29 deletions(-)

diff --git a/gpjax/variational_families.py b/gpjax/variational_families.py
index dc02fa94..7e9a5877 100644
--- a/gpjax/variational_families.py
+++ b/gpjax/variational_families.py
@@ -18,9 +18,9 @@
 
 import jax.numpy as jnp
 import jax.scipy as jsp
+import numpyro.distributions as npd
 from chex import dataclass
 from jaxtyping import Array, Float
-import numpyro.distributions as npd
 
 from .config import get_defaults
 from .gps import Prior
@@ -32,16 +32,6 @@
 DEFAULT_JITTER = get_defaults()["jitter"]
 
 
-def kl_divergence(p, q):
-    N = p.mean.shape[0]
-    iS1 = jnp.linalg.inv(q.covariance_matrix)
-    diff = q.mean - p.mean
-    tr_term = jnp.trace(jnp.matmul(iS1, p.covariance_matrix))
-    det_term = jnp.log(jnp.linalg.det(q.covariance_matrix) / jnp.linalg.det(p.covariance_matrix))
-    quad_term = diff.T @ jnp.linalg.inv(q.covariance_matrix) @ diff
-    return 0.5 * (tr_term + det_term + quad_term - N)
-
-
 @dataclass
 class AbstractVariationalFamily:
     """Abstract base class used to represent families of distributions that can be used within variational inference."""
@@ -154,9 +144,11 @@ def prior_kl(self, params: Dict) -> Float[Array, "1"]:
 
         qu = npd.MultivariateNormal(jnp.atleast_1d(mu.squeeze()), scale_tril=sqrt)
         pu = npd.MultivariateNormal(jnp.atleast_1d(μz.squeeze()), scale_tril=Lz)
-        return kl_divergence(qu, pu)
+        return kld_dense_dense(qu, pu)
 
-    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
+    def predict(
+        self, params: Dict
+    ) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Compute the predictive distribution of the GP at the test inputs t.
 
         This is the integral q(f(t)) = ∫ p(f(t)|u) q(u) du, which can be computed in closed form as:
@@ -235,15 +227,13 @@ def prior_kl(self, params: Dict) -> Float[Array, "1"]:
         """
         mu = params["variational_family"]["moments"]["variational_mean"]
         sqrt = params["variational_family"]["moments"]["variational_root_covariance"]
-        m = self.num_inducing
 
         qu = npd.MultivariateNormal(jnp.atleast_1d(mu.squeeze()), scale_tril=sqrt)
-        pu = npd.MultivariateNormal(
-            jnp.zeros(m), jnp.eye(m)
-        )  # TODO: Implement diagonal MVGaussian.
-        return kl_divergence(qu, pu)
+        return kld_dense_white(qu)
 
-    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
+    def predict(
+        self, params: Dict
+    ) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Compute the predictive distribution of the GP at the test inputs t.
 
         This is the integral q(f(t)) = ∫ p(f(t)|u) q(u) du, which can be computed in closed form as
@@ -369,9 +359,11 @@ def prior_kl(self, params: Dict) -> Float[Array, "1"]:
         qu = npd.MultivariateNormal(jnp.atleast_1d(mu.squeeze()), scale_tril=sqrt)
         pu = npd.MultivariateNormal(jnp.atleast_1d(μz.squeeze()), scale_tril=Lz)
 
-        return kl_divergence(qu, pu)
+        return kld_dense_dense(qu, pu)
 
-    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
+    def predict(
+        self, params: Dict
+    ) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Compute the predictive distribution of the GP at the test inputs t.
 
         This is the integral q(f(t)) = ∫ p(f(t)|u) q(u) du, which can be computed in closed form as
@@ -490,8 +482,12 @@ def prior_kl(self, params: Dict) -> Float[Array, "1"]:
         Returns:
             Float[Array, "1"]: The KL-divergence between our variational approximation and the GP prior.
         """
-        expectation_vector = params["variational_family"]["moments"]["expectation_vector"]
-        expectation_matrix = params["variational_family"]["moments"]["expectation_matrix"]
+        expectation_vector = params["variational_family"]["moments"][
+            "expectation_vector"
+        ]
+        expectation_matrix = params["variational_family"]["moments"][
+            "expectation_matrix"
+        ]
         z = params["variational_family"]["inducing_inputs"]
         m = self.num_inducing
 
@@ -513,9 +509,11 @@ def prior_kl(self, params: Dict) -> Float[Array, "1"]:
         qu = npd.MultivariateNormal(jnp.atleast_1d(mu.squeeze()), scale_tril=sqrt)
         pu = npd.MultivariateNormal(jnp.atleast_1d(μz.squeeze()), scale_tril=Lz)
 
-        return kl_divergence(qu, pu)
+        return kld_dense_dense(qu, pu)
 
-    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
+    def predict(
+        self, params: Dict
+    ) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Compute the predictive distribution of the GP at the test inputs t.
 
         This is the integral q(f(t)) = ∫ p(f(t)|u) q(u) du, which can be computed in closed form as
@@ -530,8 +528,12 @@ def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], npd.Distribut
         Returns:
             Callable[[Float[Array, "N D"]], npd.Distribution]: A function that accepts a set of test points and will return the predictive distribution at those points.
         """
-        expectation_vector = params["variational_family"]["moments"]["expectation_vector"]
-        expectation_matrix = params["variational_family"]["moments"]["expectation_matrix"]
+        expectation_vector = params["variational_family"]["moments"][
+            "expectation_vector"
+        ]
+        expectation_matrix = params["variational_family"]["moments"][
+            "expectation_matrix"
+        ]
         z = params["variational_family"]["inducing_inputs"]
         m = self.num_inducing
 
@@ -605,7 +607,9 @@ def _initialise_params(self, key: PRNGKeyType) -> Dict:
             self.prior._initialise_params(key),
             {
                 "variational_family": {"inducing_inputs": self.inducing_inputs},
-                "likelihood": {"obs_noise": self.likelihood._initialise_params(key)["obs_noise"]},
+                "likelihood": {
+                    "obs_noise": self.likelihood._initialise_params(key)["obs_noise"]
+                },
             },
         )
 
@@ -650,7 +654,9 @@ def predict(
         Lz_inv_Kzx_diff = jsp.linalg.cho_solve((L, True), jnp.matmul(Lz_inv_Kzx, diff))
 
         # Kzz⁻¹ Kzx (y - μx)
-        Kzz_inv_Kzx_diff = jsp.linalg.solve_triangular(Lz.T, Lz_inv_Kzx_diff, lower=False)
+        Kzz_inv_Kzx_diff = jsp.linalg.solve_triangular(
+            Lz.T, Lz_inv_Kzx_diff, lower=False
+        )
 
         def predict_fn(test_inputs: Float[Array, "N D"]) -> npd.Distribution:
             t = test_inputs
@@ -679,6 +685,81 @@ def predict_fn(test_inputs: Float[Array, "N D"]) -> npd.Distribution:
         return predict_fn
 
 
+# TODO: Abstract these out to a KL divergence that accepts a linear operator to facilate structured covarainces other than dense.
+def kld_dense_dense(
+    q: npd.MultivariateNormal, p: npd.MultivariateNormal
+) -> Float[Array, "1"]:
+    """Kullback-Leibler divergence KL[q(x)||p(x)] between two dense covariance Gaussian distributions
+             q(x) = N(x; μq, Σq) and p(x) = N(x; μp, Σp).
+
+    Args:
+        q (npd.MultivariateNormal): A multivariate Gaussian distribution.
+        p (npd.MultivariateNormal): A multivariate Gaussian distribution.
+
+    Returns:
+        Float[Array, "1"]: The KL divergence between the two distributions.
+    """
+
+    q_mu = q.loc
+    q_sqrt = q.scale_tril
+    n = q_mu.shape[-1]
+
+    p_mu = p.loc
+    p_sqrt = p.scale_tril
+
+    diag = jnp.diag(q_sqrt)
+
+    # Trace term tr(Σp⁻¹ Σq), and alpha for Mahalanobis term
+    alpha = jsp.linalg.solve_triangular(p_sqrt, p_mu - q_mu, lower=True)
+    trace = jnp.sum(jnp.square(jsp.linalg.solve_triangular(p_sqrt, q_sqrt, lower=True)))
+
+    # Mahalanobis term: μqᵀ Σp⁻¹ μq
+    mahalanobis = jnp.sum(jnp.square(alpha))
+
+    # log|Σq|
+    logdet_qcov = jnp.sum(jnp.log(jnp.square(diag)))
+    two_kl = mahalanobis - n - logdet_qcov + trace
+
+    # log|Σp|
+    log_det_pcov = jnp.sum(jnp.log(jnp.square(jnp.diag(q_sqrt))))
+    two_kl += log_det_pcov
+
+    return two_kl / 2.0
+
+
+def kld_dense_white(q: npd.MultivariateNormal) -> Float[Array, "1"]:
+    """Kullback-Leibler divergence KL[q(x)||p(x)] between a dense covariance Gaussian distribution
+        q(x) = N(x; μq, Σq), and white indenity Gaussian p(x) = N(x; 0, I).
+
+        This is useful for variational inference with a whitened variational family.
+
+      Args:
+        q (npd.MultivariateNormal): A multivariate Gaussian distribution.
+
+    Returns:
+        Float[Array, "1"]: The KL divergence between the two distributions.
+    """
+
+    q_mu = q.loc
+    q_sqrt = q.scale_tril
+    n = q_mu.shape[-1]
+
+    diag = jnp.diag(q_sqrt)
+
+    # Trace term tr(Σp⁻¹ Σq), and alpha for Mahalanobis term:
+    alpha = q_mu
+    trace = jnp.sum(jnp.square(q_sqrt))
+
+    # Mahalanobis term: μqᵀ Σp⁻¹ μq
+    mahalanobis = jnp.sum(jnp.square(alpha))
+
+    # log|Σq| (no log|Σp| as this is just zero!)
+    logdet_qcov = jnp.sum(jnp.log(jnp.square(diag)))
+    two_kl = mahalanobis - n - logdet_qcov + trace
+
+    return two_kl / 2.0
+
+
 __all__ = [
     "AbstractVariationalFamily",
     "AbstractVariationalGaussian",

From 70a400e4a3a113ebe91f76a3310709d50d828af6 Mon Sep 17 00:00:00 2001
From: Daniel Dodd <daniel_dodd@icloud.com>
Date: Fri, 14 Oct 2022 19:07:48 +0100
Subject: [PATCH 04/14] Fix variational expectation vmap.

---
 gpjax/quadrature.py            |  7 +++----
 gpjax/variational_inference.py | 33 ++++++++++++++++++++-------------
 2 files changed, 23 insertions(+), 17 deletions(-)

diff --git a/gpjax/quadrature.py b/gpjax/quadrature.py
index 8312f6e5..839c6b9d 100644
--- a/gpjax/quadrature.py
+++ b/gpjax/quadrature.py
@@ -26,7 +26,7 @@
 def gauss_hermite_quadrature(
     fun: Callable,
     mean: Float[Array, "N D"],
-    var: Float[Array, "N D"],
+    sd: Float[Array, "N D"],
     deg: Optional[int] = DEFAULT_NUM_GAUSS_HERMITE_POINTS,
     *args,
     **kwargs
@@ -36,15 +36,14 @@ def gauss_hermite_quadrature(
     Args:
         fun (Callable): The function for which quadrature should be applied to.
         mean (Float[Array, "N D"]): The mean of the Gaussian distribution that is used to shift quadrature points.
-        var (Float[Array, "N D"]): The variance of the Gaussian distribution that is used to scale quadrature points.
+        sd (Float[Array, "N D"]): The standard deviation of the Gaussian distribution that is used to scale quadrature points.
         deg (int, optional): The number of quadrature points that are to be used. Defaults to 20.
 
     Returns:
         Float[Array, "N"]: The evaluated integrals value.
     """
     gh_points, gh_weights = np.polynomial.hermite.hermgauss(deg)
-    stdev = jnp.sqrt(var)
-    X = mean + jnp.sqrt(2.0) * stdev * gh_points
+    X = mean + jnp.sqrt(2.0) * sd * gh_points
     W = gh_weights / jnp.sqrt(jnp.pi)
     return jnp.sum(fun(X, *args, **kwargs) * W, axis=1)
 
diff --git a/gpjax/variational_inference.py b/gpjax/variational_inference.py
index 97ec0b19..7d701fd0 100644
--- a/gpjax/variational_inference.py
+++ b/gpjax/variational_inference.py
@@ -104,7 +104,9 @@ def elbo_fn(params: Dict, batch: Dataset) -> Float[Array, "1"]:
 
         return elbo_fn
 
-    def variational_expectation(self, params: Dict, batch: Dataset) -> Float[Array, "N 1"]:
+    def variational_expectation(
+        self, params: Dict, batch: Dataset
+    ) -> Float[Array, "N 1"]:
         """Compute the expectation of our model's log-likelihood under our variational distribution. Batching can be done here to speed up computation.
 
         Args:
@@ -115,20 +117,21 @@ def variational_expectation(self, params: Dict, batch: Dataset) -> Float[Array,
             Array: The expectation of the model's log-likelihood under our variational distribution.
         """
         x, y = batch.X, batch.y
+        link_fn = self.likelihood.link_function
 
-        # q(f(x))
-        print(self.variational_family.predict(params))
-        predictive_dist = vmap(self.variational_family.predict(params))(x[:, None])
-        mean = predictive_dist.mean().val.reshape(-1, 1)
-        variance = predictive_dist.variance().val.reshape(-1, 1)
+        # variational distribution q(f(.)) = N(f(.); μ(.), Σ(., .))
+        q = self.variational_family(params)
+
+        # μ(x) and √diag(Σ(x, x))
+        (mean, sd), _ = vmap(lambda x_i: q(x_i).tree_flatten())(x[:, None])
 
         # log(p(y|f(x)))
-        log_prob = vmap(
-            lambda f, y: self.likelihood.link_function(f, params["likelihood"]).log_prob(y)
-        )
+        log_prob = vmap(lambda f, y: link_fn(f, params["likelihood"]).log_prob(y))
 
         # ≈ ∫[log(p(y|f(x))) q(f(x))] df(x)
-        expectation = gauss_hermite_quadrature(log_prob, mean, variance, y=y)
+        expectation = gauss_hermite_quadrature(
+            log_prob, mean.reshape(-1, 1), sd.reshape(-1, 1), y=y
+        )
 
         return expectation
 
@@ -173,7 +176,9 @@ def elbo_fn(params: Dict) -> Float[Array, "1"]:
             Kzz = gram(self.prior.kernel, z, params["kernel"])
             Kzz += I(m) * self.variational_family.jitter
             Kzx = cross_covariance(self.prior.kernel, z, x, params["kernel"])
-            Kxx_diag = vmap(self.prior.kernel, in_axes=(0, 0, None))(x, x, params["kernel"])
+            Kxx_diag = vmap(self.prior.kernel, in_axes=(0, 0, None))(
+                x, x, params["kernel"]
+            )
             μx = self.prior.mean_function(x, params["mean_function"])
 
             Lz = jnp.linalg.cholesky(Kzz)
@@ -219,10 +224,12 @@ def elbo_fn(params: Dict) -> Float[Array, "1"]:
             diff = y - μx
 
             # L⁻¹ A (y - μx)
-            L_inv_A_diff = jsp.linalg.solve_triangular(L, jnp.matmul(A, diff), lower=True)
+            L_inv_A_diff = jsp.linalg.solve_triangular(
+                L, jnp.matmul(A, diff), lower=True
+            )
 
             # (y - μx)ᵀ (Iσ² + Q)⁻¹ (y - μx)
-            quad = (jnp.sum(diff ** 2) - jnp.sum(L_inv_A_diff ** 2)) / noise
+            quad = (jnp.sum(diff**2) - jnp.sum(L_inv_A_diff**2)) / noise
 
             # 2 * log N(y; μx, Iσ² + Q)
             two_log_prob = -n * jnp.log(2.0 * jnp.pi * noise) - log_det_B - quad

From 9c1607c352d8abf0f071939585af38f7ad3f309b Mon Sep 17 00:00:00 2001
From: Daniel Dodd <daniel_dodd@icloud.com>
Date: Sat, 15 Oct 2022 10:26:00 +0100
Subject: [PATCH 05/14] Fix dense kl bug.

---
 gpjax/variational_families.py | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/gpjax/variational_families.py b/gpjax/variational_families.py
index 7e9a5877..96d9bfa9 100644
--- a/gpjax/variational_families.py
+++ b/gpjax/variational_families.py
@@ -709,11 +709,11 @@ def kld_dense_dense(
 
     diag = jnp.diag(q_sqrt)
 
-    # Trace term tr(Σp⁻¹ Σq), and alpha for Mahalanobis term
-    alpha = jsp.linalg.solve_triangular(p_sqrt, p_mu - q_mu, lower=True)
+    # Trace term tr(Σp⁻¹ Σq)
     trace = jnp.sum(jnp.square(jsp.linalg.solve_triangular(p_sqrt, q_sqrt, lower=True)))
 
     # Mahalanobis term: μqᵀ Σp⁻¹ μq
+    alpha = jsp.linalg.solve_triangular(p_sqrt, p_mu - q_mu, lower=True)
     mahalanobis = jnp.sum(jnp.square(alpha))
 
     # log|Σq|
@@ -721,7 +721,7 @@ def kld_dense_dense(
     two_kl = mahalanobis - n - logdet_qcov + trace
 
     # log|Σp|
-    log_det_pcov = jnp.sum(jnp.log(jnp.square(jnp.diag(q_sqrt))))
+    log_det_pcov = jnp.sum(jnp.log(jnp.square(jnp.diag(p_sqrt))))
     two_kl += log_det_pcov
 
     return two_kl / 2.0

From c1eb3463e7e54d951969657b0a7d3fa951754346 Mon Sep 17 00:00:00 2001
From: Daniel Dodd <daniel_dodd@icloud.com>
Date: Sat, 15 Oct 2022 11:27:06 +0100
Subject: [PATCH 06/14] Fix bijection bug.

---
 gpjax/parameters.py | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/gpjax/parameters.py b/gpjax/parameters.py
index 257d3115..93afa59d 100644
--- a/gpjax/parameters.py
+++ b/gpjax/parameters.py
@@ -178,7 +178,7 @@ def constrain(params: Dict, bijectors: Dict) -> Dict:
     Returns:
         Dict: A transformed parameter set. The dictionary is equal in structure to the input params dictionary.
     """
-    map = lambda param, trans: trans.__call__(param)
+    map = lambda param, trans: trans.inv(param)
 
     return jax.tree_util.tree_map(map, params, bijectors)
 
@@ -194,7 +194,7 @@ def unconstrain(params: Dict, bijectors: Dict) -> Dict:
         Dict: A transformed parameter set. The dictionary is equal in structure to the input params dictionary.
     """
 
-    map = lambda param, trans: trans.inv(param)
+    map = lambda param, trans: trans.__call__(param)
 
     return jax.tree_util.tree_map(map, params, bijectors)
 
@@ -202,7 +202,9 @@ def unconstrain(params: Dict, bijectors: Dict) -> Dict:
 ################################
 # Priors
 ################################
-def log_density(param: Float[Array, "D"], density: npd.Distribution) -> Float[Array, "1"]:
+def log_density(
+    param: Float[Array, "D"], density: npd.Distribution
+) -> Float[Array, "1"]:
     """Compute the log density of a parameter given a distribution.
 
     Args:

From 709c4405b1958843c1faaa68f3f766c4cef32526 Mon Sep 17 00:00:00 2001
From: Thomas Pinder <tompinder@live.co.uk>
Date: Sat, 15 Oct 2022 20:11:47 +0000
Subject: [PATCH 07/14] Numpyro mwe

---
 docs/requirements.txt              |   1 +
 examples/uncollapsed_vi.pct.py     |   7 ++-----
 gpjax/parameters.py                |   4 ++--
 paper.bib => paper/paper.bib       |   0
 paper.md => paper/paper.md         |   0
 paper/paper.pdf                    | Bin 0 -> 209198 bytes
 setup.py                           |   1 -
 tests/test_config.py               |  11 +++++------
 tests/test_gp.py                   |  20 ++++++++++----------
 tests/test_likelihoods.py          |  28 ++++++++++++++--------------
 tests/test_parameters.py           |  20 ++++++++++----------
 tests/test_variational_families.py |  14 +++++++-------
 12 files changed, 51 insertions(+), 55 deletions(-)
 rename paper.bib => paper/paper.bib (100%)
 rename paper.md => paper/paper.md (100%)
 create mode 100644 paper/paper.pdf

diff --git a/docs/requirements.txt b/docs/requirements.txt
index 9f284740..f8305d6e 100644
--- a/docs/requirements.txt
+++ b/docs/requirements.txt
@@ -6,6 +6,7 @@ jinja2
 nbsphinx>=0.8.0
 nb-black==1.0.7
 matplotlib==3.3.3
+tensorflow-probability>=0.16.0
 sphinx-copybutton
 networkx>=2.0.0
 pandoc
diff --git a/examples/uncollapsed_vi.pct.py b/examples/uncollapsed_vi.pct.py
index fca98678..2133c4b9 100644
--- a/examples/uncollapsed_vi.pct.py
+++ b/examples/uncollapsed_vi.pct.py
@@ -139,8 +139,6 @@
 # %%
 parameter_state = gpx.initialise(svgp, key)
 params, trainables, bijectors = parameter_state.unpack()
-params = gpx.unconstrain(params, bijectors)
-
 loss_fn = svgp.elbo(D, negative=True)
 
 # %%
@@ -165,7 +163,6 @@
     batch_size=128,
 )
 learned_params, training_history = inference_state.unpack()
-learned_params = gpx.transform(learned_params, constrainers)
 # %% [markdown]
 # ## Predictions
 #
@@ -176,8 +173,8 @@
 latent_dist = q(learned_params)(xtest)
 predictive_dist = likelihood(latent_dist, learned_params)
 
-meanf = predictive_dist.mean()
-sigma = predictive_dist.stddev()
+meanf = predictive_dist.mean
+sigma = predictive_dist.variance ** 0.5
 
 fig, ax = plt.subplots(figsize=(12, 5))
 ax.plot(x, y, "o", alpha=0.15, label="Training Data", color="tab:gray")
diff --git a/gpjax/parameters.py b/gpjax/parameters.py
index 93afa59d..6a8d8c33 100644
--- a/gpjax/parameters.py
+++ b/gpjax/parameters.py
@@ -285,9 +285,9 @@ def prior_checks(priors: Dict) -> Dict:
     if "latent" in priors.keys():
         latent_prior = priors["latent"]
         if latent_prior is not None:
-            if latent_prior.name != "Normal":
+            if not isinstance(latent_prior, npd.Normal):
                 warnings.warn(
-                    f"A {latent_prior.name} distribution prior has been placed on"
+                    f"A {type(latent_prior)} distribution prior has been placed on"
                     " the latent function. It is strongly advised that a"
                     " unit Gaussian prior is used."
                 )
diff --git a/paper.bib b/paper/paper.bib
similarity index 100%
rename from paper.bib
rename to paper/paper.bib
diff --git a/paper.md b/paper/paper.md
similarity index 100%
rename from paper.md
rename to paper/paper.md
diff --git a/paper/paper.pdf b/paper/paper.pdf
new file mode 100644
index 0000000000000000000000000000000000000000..bd501dd071641b3c7bff33ee96c619c46b7474cc
GIT binary patch
literal 209198
zcmaI7b97{H6fGDV9otFAw(X8>+qP|^W2a-=wr$&XGWC1!t(mvh%)I)iZdI*Y_1&+|
zK4<TJ&Lxu<5u;_KV}&7`T<TjKTAdvnhG8LOAha{IgyG>Kq?aXRCZrd$uy!_a1Uy?C
zIGc!=7}*({5YqoKu{CoxCuCx0U?Ak<BmAEa7$;{(69XF<_w}k&NM)2Y{NbC)qNG`5
z;_Q5OXc<C(2|EzQq9`c|5U@lbqWo3ZD40qLV&XiCdaC`v3Ti?tQB>F{{l04m4dq1_
z2Q(Zi?mj2gZLgE;m*?!4slS^$Z}(Gv*PkcbP7pwr?Ba>PgB5^849iHVW+7wbBt#+{
z1%?rb<w3b{ZEamwKmE=9DOYzeriO%e2_8*<2}Isy#1x>vdMTdVIkf=^NMhQZm`WF6
z_6%^ukRv|;`6n?>A4j;uT#vYSCC|t}0!56ZXOJHl@%1YmK7f6CM|&}{A&rZGM2|3-
zhQDTM1KD$g<kC+KpC8bm`br(6&ZyRcd6`F(pk_UXcqsj)*JenGh`k?bg#$unwZSg^
zV-zbAxfP`ub7bc0QBCPV<r!)?<<#uv;@a$>zgCF;&B-K)FFuMPtzdzkk()a+o>Ej?
z+p0mlxV@AT10kBXmRPAA5O2p>rS(%PrRW==%=j+#ic6UrR)Sg(fQ+CwD1M-Bo+m(I
zoUh<bw9IkIAAngwC38}^R*#zRA9Y_Zf*X(?GCd)5z<~@cPD7g&F?@Kh5Q|DXnJ771
z>CH%mbZA-+l)emFVt7>~UA$)7+#JaX^QxV?FZx~lgZWC&a^*`S+E$};e^BaZEJRsu
zPQWhxFWt<REo?&E=G#N?X9h!7shSWz``k{Mh08{^@r>S9^K<9TPg-x8|F`D>Rj;}=
zr1A#^(AyPM?_3?D<ea2LT}vrbJ}P~l`jx}`m)`S8a-LrmOd7TCcRNC7WR{fR<aW)r
zn$J_cD{wFV$n$PVjqW%b(E18+p2tUpnIfXU6cUilmiAxLo2b&&av($s0dNL?^jXN^
zUXWEF$X$@l0E}BmCqa&I2xxtz-9TqT5af8o)PSWt%;Z2Bed>R}PkBfz0qFLajX?*y
za2$wqLB95&o8XQNAUOeyw-A2>eu)H||H7LOgd<cM#dh>xoy8p`fC(wc1Eau_3L1=u
z(}buBk;;ep2eA&%D#$M|lE*wNY7W#64=-3aOE3$G8BoRmx*Rm6uayq(H9%y~()mj^
z$Z40%9;p+lIuvI&$R5WVq8o0<|LB(49TzwpxnBeh2`4z_m3U8rI5AQp9!&yTF@%MX
zR)Ty!NL%=D)aj5n*1zyTJ_~*t4<k6L&%uGS6|I#?8{U?`8h$0FBNV4UW{>Z{!d<76
zbsL@_I&P$N@8NFfmFiRXQ{j{I6aEwG8$LilA9)g*E|^dVgea&sSzc^a(3~vmH{7pK
z=mQaAg(!2GN5W7PP7%tw^ww<mc=t&66b>P8aYbURcw7nKB9X-O#!zKZE&(pdb}7$4
z8H5#yGGn%f_Dpfw5_IJ3iQHpqqa=sWk=R37`+^ZA#`3C4<&?C^PRU-$YEn{CVp7#o
z2~r@%DpE^QVaaq9@RVJp<w{lmx)({yquUeO<2*AwGv6b{3jfY^nn$ou+K^O5xTZe)
zKg<5Do>W9oJ*KQtu2!*C>Ms8j3aK<K-><5dFo}>#8AnQ?rkYV1D>p6gF2`L$D{m?v
zQMN0Ot(aT7T<TvEu{5!ST%;@KRUj|VEc>W<Sd3bH`IoCgP(tEt)J~r*ttV|)lv^0b
z00rkH#y}FJF#eh|H-fn!zmi^zRH2ZkP~jjXJS~!LI;me0uQAOtiEHeep3gqu1>rXw
zibcy@J?&mmR(MuCwn8&!z0{fiCJ4K9zcRkKUb$55qK8vzQ;b)ZUDKuXk@f6^7aH4v
zpjIk!Qag>OqrM`=F6CBsjlZS4<-qIHCD3Kvb>(UHlykxS)b+^xNc|*t9)~>(e~bwQ
zM+Gm7#UsipZ>JotTDWzZo?mScpFe5eVvxHRoY1BaJ%?SWTKF-HYSipd?ojNIc9$Lz
zAelbKlw_L3U1F@%Y|(zkwt&CTVtQ=qV0tmRK9!Shl>wPa&uGWiV<y;e+c;WhUfXIb
zS!=9is(D>|*%DZlRXJZ}Syin*usppCa)GxD%6!MHV-{(~HrZ(*+;rDG)?m?$Wyq+J
zU&&hOx(us@vLUiD)y&(RdzEz=>>A_hcC~bMvR+en)A%#cHy&^ye__9saE!UzI$t+%
zcSLic@u|UWw&hsB9l}}Vbnf_mc)3sGVC#VI!ssCBTISR?W?D<KYH|r%?XE_!l2SdQ
zv(v%ZA-RRn@pWx`kaOnS`!#0$u=cqY(-6iR&BKA;!F%X5%Q@!cI+fss?diya!$aVS
z<#pHj?8*Dw{5XBfcly;oGA+LEw0KfglX>H~v)~)9m!}t@m)q0s7ws4G69iP~-{^nr
zUnfx7TO3dtbSp?Cm>FaXaRKfWlo>d3+&-zp*TNl)&BR~y`_SN@%0JRZW;ldQoMy7i
zZJq^>eUGa>LLLrp>DMo)hw!ZMTw>@1q=cq~xPr^VRMDr&(iuITpWaXQIlj50IX5F3
zLt{prI;uL!I+;4VgU@}c{lmNFyNLsOa&vMhWG>{0h`I<blo5<q`qg%Q(lN?14h#GI
zxBkEo3=9i;WIC=obrVLjy98rzNA3H7!?-MaSx_?`GLB+8ljw_ytzn<xhkccMu{Zor
znC}DSxW&jtatnM5t$DEp-i3<!Tv%^#*1$9@L98b%JLUoA7ZX{SgWxLSL<$Y(Kb3UT
zsBUIwW@yIn##&|=W)n`Ej?Kr>l0u|jPWhW5ZFq_h&qPbq4e>rCIOCZ~YMC!|{A@oG
z0h$4=c`dW$_QHDyWWdthT9I1gI_pi7_G{-R!V_I-H@d)85*l+&^X?{d;|VGO70)##
zmDAc^57-^7XjZy4F<KiMMQV-Z=C7H(b{SDAQDj5xD{VCudU0Cx>N_oc4g=5dxzg@r
zX;N2G6;k}E4ij`fRzEgRElD^^?9VCOX*Zi9tEAg3t!qvX-!M&)WM~6g{mqRk1u0}H
zH7#`8o^_V0Dqn2lY-?;DF6J&Hu1Qy!&3=`VHEjwPN*kJ`S}pn4R|wm&72I8}&5IXH
zY0G}AWUDS#ZoAjb5DlT_yurL4UihrTY|q{!-_;g1&)f4|xz7<VV=v-YH@1!zmRxyW
zd7W6csjPAMf4EkEx^`ahXxV5xwEk8VhxHPz<(Kl4{xG<69NMoSJtw3R$oN%qOYv*Y
zLo}f^$266;OmHJ}LBvk+E-#&Jdpwda%YDS+;MefxI~0799vRP#jVVdX#^g@(eSQ8}
z|GQT`q~kU-oZGdHx3=X*ceoky0DLm@?EaF~zU#2}D%uh?LRUxMr03ebxqPwc^=MX_
z8~4ol(xl_iHa~Z>zQf7jwR@;CQ(;#9ro-*W`}%%_^I!`HHw?Gk{?}Q_H_Gi|%x3Q@
z&F%jB&!&zGqF>En_xV>jWIH0ZZ{26H$~6r?<01RvhyC|J$H3;ES<mtJyF<|_(v`xJ
zQP7Bt9BvLS0X*iFU;FOaXv3u$%1o4>(nr;G$WRo4-^+WK8;!5^Ve#S7A%RIvqwbX#
z?}f{X``~L3djUI&o%FUv*TzrISNDnVX4nYI8$kp?1;2*x`6urE+jfz;BNO}5+}=b>
zZ#X58Z){RhP*qY=P7Uy4GXW420rv<wxgIJ|xcVFUXda0mjESxB|DvUUqyLC4JH!9R
zaQ{!_mX(u*`TvXDUhz)ElD0X#d`EJvhaJcFvCo?8TY72M+FHTq1?4SZ)+6vTEn^UT
zJ|OTBV;Y;Pnms|m5<qZwsf-<xU`DeC(fG{0?)d$fea)u%{R{+^=}iui7VESQ{zL0M
z+IR`u!B?JMYkSxAuBwSfeCjhp14Z+DZ@D?Xzt=<T+4p-_9DW@L+J<v43HJBAl;=gL
zt=1#p>k?{D|JUmte0H%jj4LEh*(yD%S!_<H*Q57+U)1zf+0>%R{mI;8j+ecBJMmD+
zjI@(+@leRxyh4350Q)WY;WRtG3A;qa3m%hu{DuUA-iwwA!n!xz&-_}d+T=QGCF)*X
zRqOFZbE891wOa<y9$F4$)M1|c^Zpc9%J6cwMEpB?d^@u>D$}1HCI2^R@OhKK0%7F>
zy%SWOT$5mMgZEX#B3i0k&9-jAxB?DI7;sGrGgcV%I5z%>u${?*FLh}kyHbVU+m7Fr
z%riMyUzxf7UwN$x5IPx<?&x`@zf90+ha2^Y%REh-h67p;4SeP1UtB%*KBoFf<3zhW
zbtz(_`Z_>n$<JDoU>w$1NZg#2w6LcpOxr2Yd|F#&&}F%0Re!jQvvMZnY*K8B$i)wH
zaNrl;R}@s^Rm?l)g>4v#R#xu*R8Vgy^QEsJoWRCmi}L?*z>5^q>oWZn6&mxt(U;0W
zPw`~vE_>7vlaQLaXcy8ZxC^PWVWssX_H$F6FD-#gR72fY;nx{T=as^C_2)o9ZxC8&
zFOo*zdcdqL)aJTbVIs7wN|nnD-MwxsEpjMQj#{D%eI;b9UQ;hsWJ%alR<JuDKv`@H
zc8Ra0^7*j7q`NX84b7IWg14cvZ(@iDl?#5{j5$AEbfz<B6Kn_xv%#>dGEKy6R+!#>
z<xZFAeBW6>+DZQ^Ns{J)c)J+L)jza|#&TXiu%Dk&80LvnNyQLP!~pb&UZSkY=x!B)
z2GM8qy$PW`j?ykUrWWCRVjN>#I>cq2Ib}^TG~KCsFY?j8OlxOBPeJ<(xA(<h^d#Dw
z`_4QmGliuhM*vTkKWdRx4#n&m-c3Yw{M7kwtT5uhyj5?1PRdVT!yT&<MA>XYzI)>H
zUzlO^Em5-X+qe-~1<3gb6FP@h)zAc}N@sL4N9fd;-uH<j)o#y-raw$}k2gjL0yO2o
z(bc{GI-k{zo78$KrRY`h2DN9UNj~b=+x=}E7uz}wM<jJ`W7|2C`^WnBqj-gOAg=Vm
zGiCzALosHlDkz_kNiy7GWKNF`{y#Fbu9Sl1jL6{=&Fze3-DBWwW28O-26+347dxYj
zqq^z5raa1-nJTL0_Yce<J|rB!cAihopQ%$&6Qwzc$G0v*zG3r(-H#I-R$Z+U|M!H7
ztCpe~<qsKw`?&WXoNbZPDRIH9rR>(JJS6$?ft&uk2l>Igq<c(D$7QyEbbj$p%$~}#
z$FTo?`8i(8O@H|fv}LmqfIH;uYUidn9J}ppdi>jO#3-rE4^XqX_^$PQ8edZ%X}Bhj
z`#zl8`77>+!*fPLMg0jDGwh>M;`c_~{f7AYaZs#?j-Y+++k+VnBl)cO>!CDEFGS4t
z@RQ@Tt^Uad6fat;e~on@^IR$+=3a?zD*qu-MamY>HaKb2f;yr5lf}viOS+7%0{<?R
z7**R&$>!RgFb!-22gAyL5>z?uf&Add_;W{tb7xtDeJ9qwTaazVJb*hT<oR6ciP+^T
zG~Vxm8Md3jUsY1vTPNW{2gmsb4dfSUdh*rLT!BHn4vqRQQz>$=HbsvQ-#kwFB=`=D
z(yDbm1+~%<-8m@h$8!9$gv@D!f_%jSr9`t?>%VV^S1}$xFFP4*%5-Ah@r`Hji6}7W
zYPvv3D_FYdNN_dn$KR3hH4|deln1F^Cg?pPuu5yE*Ehwe6WZ8SJs1d3e5Y;VVdtb4
z+~ZP<LgJ{5t3lJ6sH^pZ`E=0~(rdxN6K68Sgx)k0Sha8N-@gS4>Ra#_cDQORlYFWy
z6@>&>^2GZt#Wg~<!7nY7Q?g^?VYRpOXE^b(oT)(2ngnUXiP5wbD$O-bDGEl+;?j=-
z?Lcs;yf%OnZett5(MBlBl~jNVCJ*A#vGh9XGeg0lx+xbk5>|(uihYuiVo3iXz(h4G
zR$(a2hL%^g!uedQYdWy6LZQ85mh-NJ7}#t61;YH*mo`;2%{)}&2Hk9wTa-x>7Y+^f
zJC|7(*ON(t`DSpp9GC?3z<6{MM@NLW7%1V#h;DZOk|sOQ0{EpK@kp`7XV~|2t@jDW
zW0V8)Gl?jd8=U#Zw<?p!1mdv!$)~SlU~HO9Whb0A7x|}UN`SP&<<jidkbu^PrewGI
zVBgYooywy5w_VXJ#CQ|<+`!Y@9PW)~HFGV;?rKe=A&5mK%iirInEbj}!=siyY!18A
zJB3LwL{IiEKRkCbJp1#+1Ra`<^!+k+t)4F!Loy8*OKME&9;egQVtt2!@JEU3x?7c8
zwE{-i6GY*Ih{XlPL|2Q&MUC9-u{J5G(qmIhsLxB&4Hm?#dt)zVx>U*6riK2#(QzO!
z_5*B6a&&0pD?i;ONM<4vMWe--9$OiBC*p8sYA&R_V^W*Ak_fGof&EI|1iC(WVI>I_
z5ldOWDc-}XS(S+4$U0Y^?ApadBjpf_)vzutt%XObK8n6ig$trM@D(qCoIFiJ($8%(
z9z5Rcr8uH5Ry!dYA9Ez^{MYm)Kk}?$i)!JK)dLq(M)2y#uH6BQr9ORjaqxXts7fa?
z(I{T~wV9viQ1Q&@;V;~)`I_ud!$Q)({La+$P*lulsU-Kg&1>SezkE)9`;wAB(L|uE
z;@%30h|e6MZ5t~8IzT6tXr=u%=>nRRA6Fv%mt~Nvm{tJ<60A(5obWkJ^KU>SWuA*U
zn*anv!9`hj)VehAZ+JGaBZ#+RR|DAnsvN`tpYh9zh@P6wcdE2Ym+;;?WzuJCX7*tf
zb&F!$w7C=`$0Br^D$7I!b)r`bLF+=L)AgrzVQz<0k>#kWr3v*-stGH~lf)&#%wa55
zZ0Ho;Gi9_n2>c0Nr&RR@r5a=2nB=&PH-W|b<kLiRS6OOM1JMWF-%3UU)%ut{WK+F9
zl)+JcxrpIvv8OHqr3=+0=+La5Iu@)~3zp;=tVENAc**_0d`oDN^P{UoR<&rrh_-{I
zuyG0Vf^t{-woay@&murlM08mcjYvE#BP%I`BICi3r#KL;;^zL$s)JGtEkK|Nx^ihy
z&q-K?AX|8DD|sVt(AOrxW59~T;g~h1qh$R%?izKO&eslg;YzMO`K`et(Ud@M<XU;v
zai66E2@>G@CZxBk!bS*xmEi-L<=xuBR+t-1XSIfz^YJ&?nZ~t27WxahtcZD?rIrnB
zOHG9R<57BjX#yFHg7>hAL>#)s<*j^b^)ywd8OfdiB;1m9NLr2t`HC=92-DE3D{f(M
z<DW^0r2+6J(fQ%;#SL}VHOMOz*&G2G#Ms_g(Hn9p7|)~ahDuQ4N`dyu4se#8Y6fxm
z!1jY933aTBdIq_pL%rh5yvq(g$3Dfe(+YIz*rO_pvk?_`h%K~OdqeyVn&OqMq;^qf
z`7F2-HHjIetzS}te`WapIP*z2vY&3MD+<ve72w$fHy;mGUlopC{gcLFL1H5b8>-1=
z^1Sd8(rz+?@+eBBCRa!HLe6hmbUDB3Wb)4lq0-{>C4+-khp8Zu`i8;MHU3lUN+M__
z2%wGgG}7Fr8dkOGCS3+9+7^GZ*IA>WUDNB}Scyh#y|<Qa59$?XE}YFl=IJ~TZG~1b
zYhITV(CapQ2l}HJD%XDd9_PfSubSw@sl=0}0Y{L6Ag3q<U3g_Bn5)^*R&^^U<yB;G
zAD0p-Q-2HVGHOYOtNnWsa1Zw^c=!Xs0^F{$Laiymq-Is-+FRnl3^vRd_5E!d5*JDN
z)<}%5?@q`{X9@`cNf#<u>;_NOMRCOpp%NiEvdd15x2nj2oQt36$YopN*z#&@7*%!*
z9slXQ275Rvk&EFg{-Ca&2!A)uG|n0Mvzki8N5EdYkg15x7C=8Q;>1fonoQe>0aK_D
zQ*P4HaIDYkyvHu&;l9?<%oC@XBIOg=Y_{Um-KEDgR64*tFl~XFwQyd@^~vH|NpnQ*
zgMUdAhu2EdGf;w-)D(=aoWnp^M@YM~Ow4EF+!fwq_c{MNDT|%yl>D)*C-LwXig|L%
z=R(Jzos+Ujsg39y=YHBK9dci?%u!n@r$nPBR1p!f#n-zHJJ(zmS~=6BDe`^RXUlz;
zMp+n4UmC6O@m!3nl*nk-n_`^lsw^9|QKH#UAJhKTa$Jh34!05U!KTG^JUiNW?cXHR
ztmF$&`jqnd&=ISMplqNjC}-v)$EX;=)c&N$C6Uec7{bnQ(oQJJpUz2({io$`u0SF9
zI|J2JuQ0#a1?AIeX;76Mh6GY{-9gG+(}K=uqyU*@i-rDm&>OO?d)>+{2}tV84-M7n
zfP(2`v;H1gGnF>EU_a4Ui<LJ-2PTT~qz_TUJ?nH#HTfeB<hlOIjp{)Rqf}u2v!fO@
z+~LDgRta%b%8Q9a9Sj<RR(+;(weUQbx(1XbAJH!sSf{c&D+vyqGGhb(wn67YXPshN
zy+a`@;V980@*$$a9-DQ^bo;8RMslyQHVIP7F&B$MVal^-*%Tp!{>C({ZsYbBY>Np>
zD|hI7OCou1@f;@{#li}Ht+_js?PY4p(4z0raP8@Zr#p~8;u#qh2s6BYXp%53=ti)M
z={*s+ErZr$-x~0YLkH$96CiL_x8O$?inu@!7yrVaMnO3IcK<z$4xIn?>lyk&ge8f4
z^>OJE`o$f(3NvhU?V|cVq>$NrFqK8j%ZX%HZ)N?gdzpo)E!dTlU*m~H{9h3dKC^#?
z{oZ5s-(c3_*zBYGUc7@tF7FXCz7#SuYsR`3ci_(<&#LIX%fvO#2O<0fH@%luL6}cP
z=1W@jTU@z2Glb8u+K6-1w|ciz{>@_)Y)7!@{$W&0QFb9WYr>X=D=(fEF$0`kt_{Ve
zq}EM7!W5ItNR-G7k+FVcuTUt>-0|pzDJH5OE_C~?K@9<@1-j27R(>pS$OL`@6{-kd
zW?VBJwO9hrN8tN)SS&sl2tjPl&LSM)h{l`M?<55>H(S+c9SZ2GZ%{sH@dKSPPHk?(
zb4s9d=T`nQln)@MEHgJf#Szd3-^gEG;~8jS88NtmUZln)?$|*cjPov<B!8NW6G^#p
z6G=&f^Ecd_a$p7hn4xBezdJEOyiN1uNAmF(B>R%%1fxWSmMmsET<6;;?S$-@uk4ys
z@?38mPdo6L50TB8sQ<k{ZV*w_W>UhYH3ZP0IF>vz=Q#fQ3(G{7z>}m}fn3F~$lWet
z9cujA${K39g@^a&RL9Q2*{QN*MyEqatf*^iz3yN)cLD}{g(nM9Tsic2TK(|NTKjam
zFA5mX)SEa8b=Z(Tq$TI|xW#x-3e@B?C3u)uh1W|xbF%i$lt{JV?22oH7>@s@m6<;I
z{&}SiadnslU7~-X!tx1Pc(?RBW=0`uj>cCBolw@?R%~MaevDt`h_NwPmCRmuR=@(|
zt_dB1Jg11^D6lI^J|QFZ2_F`*3~O)rZ0kkL?mWDSJiWSuFp=np-KP(wnkwXxp|&ZQ
zp<hh2B%5hqZGeUj+)^#{Yi7y`<=C}qYRbY9I%!Am5H#xWmAfGm6U$YB`L^5&>Kxun
z+o%gP{lsxnjyHiXSAh&sS_g^lG@xOJ0&L-+#-(u~`@S~7k~s-^k$c=!Pb~ZUtP%@2
zf1)29uTVEGrDxOGx&JZ&Aw<VXaM*OW{;#*Ruj#tB)kbbj)Q&52zKawuPa@ROej(~Y
zJN5|~NKFO1xvp2ov&d>H{uQzb2h@^)xnw-?m}&N>zQ-4EDkMDB?*^<>mYIu)6Sszv
zj4@C08DTo=6t!OOJVA2PIWg_?#`62l!f_|HBRnf|l{YBXBgEZx;KiWtiR5SHw=egu
zvT?Fsrxb72E$%5qSpCKin#-54wMr7VqxX*g%o?<AD6Lh?x8_W%l7+yW*~1s?nX;=^
zI#U{*gfLf;<7aSg`Sl;U<;O~7#FLf~3F6G;oz7s{Q0r&fiwY;--nb#kL){*{+YL(g
zI51sF$F2Y6Ll%V5_peoWq=zhiLavVIr5#VM|9C(jlgq+(sKR<n1QG$es{`X?nI4nY
zWhzG$fL2{jk>0RrN&_J~)SsQYbn)GoY-w|?Qmj}>R39^R(R6fkh1XtGgTi)xTI(xh
z)}aX&7sl!d_uFIkfeMVJZE6VnnA!P;)$@Wa=I3wg)oOFEH(L>hS)PDZ@ESL-v$Soy
z3JOr0jO)S$&~`ZSuaJ?2>_9d1lJcJ@hq@FH<ncX{I5}I1C%Z~G722VqS_Tiiwy9ke
zh|n)@fqY87F{tKWzcV}+d;5Eeo^;*!T|{HM2St0j7Rl(vLqjSi+N-I*WUw041q)A9
zl@nk2(j(Q|v<^j>4<S{K#h1D_WwPl~P>5!_c)uM*DZ5<!{^PBwqeFEh14=)AX+NHm
z*$@R0UprNz%^vN(W0aB7g%D<riNVrfSi9J2fzA%yGb(uUgvpGJqsjzr<RFGfL!LDI
z+fAuoigl1U{sdbTqAVR@DZvbeQ%?K(RkS0n!3ML$R8ovxR_+g@?O)@m4d=0|T3dp9
zok;X`{}L~wz-O)xVW#N~Sw(e^0&>l^lC8C)?$(!7zHUNOu71_)YxQLsOH?L7UBwBM
ztgLX2dqe_0TSdo(Fa$OkMwpH@>~yF3`wmWv#j)o~<*aF`rGTten@Q(`n^qDD2+3j8
z8!4hq$%&{F6bQAb-q+yY?QeI+eOIzreWNE#ac)nYJ3@{>)%d;ZWAOQY`lFboD(l<1
z+s0*j*1sX03M7&)4Ymx8TCj3fY+5$L$Ea4(3O;BNB(A<Aiwh%yduuvZ7T4FD-J!(K
zZg)ePW(Sa@ec18JI`B#z#RS`o4#gApL69R6J|u{0$23ozOeHt{1^alQZc2?)16F!V
z1B&V!F0!{9ZDs@0lEMolE;t5%iG=DMCiaCojN^5YJcNEW&}=ULQwzXj1L5ZBBH+{v
zNajOWQy{68>gnyJTyLb!JHb0YfD5KgcHr$zOv_wt+ma`-F%{40Vg!agNY<yWR~HS;
z>m=_i8MwiRvn2WleM)ng;yUT>*(8bT;K}QtDE;c4w81UKsHefV)S@BL^AB99=Q3Sw
zle<*3Yt#a~i$vtrMmp9`yew>J%FHM$HH)iHlwA7ITkO7T`>26O)%9jKX-;<!?(i+N
z&-UH;_rApV(oKo8qdqD2IM_jDmAkWu8Skvmfj+nk+ez82g<Zl!wlPk{4Cgq4Pn#qA
zVPMO{oUksiS#~)F>NM{bPh(#eW40q7-na$FW*e+NQKWTBtb*xRsawOhiMiWU78d01
z%31FDHSTtcLxy}WWc=rDBh34}>`zfpk9IqVHwNl>a+Z)<4t5{oNZHt$NB<Gd!!2qS
zKa^7L^P?als#;SgJ<8U}my}fI+-aQdn#S#mrcc7?R?<ATOWt(_G2P_Gue+>kAkRFM
z?JJ6jXEk%K!k34Wi!E_tF-<Gn??DA#169BM(Sg6wBbw5Hm!K9?u8+J#k$%pmZxMyA
zYk^od6?6{^O{h(AdmdvVtzWs@bFv|r=qzan;S)zO+$tIG5!V|@x{4b(G^(GApe{qd
z1Kl<V0`smI$1{%qI;Ov;>yB3`muYu1Oc`IcU&`(X4$|$<Q?$N_H^|pyX^k$tTwo<i
zpqNi^eGP`%hsy5EBX!-|X6tr%mQ&k{udFxeeNgy$Yv=#0i|a)Ac6WCY^k6edZcckm
zR}fxS`1RLBZnid#QY0-CecAjhbo?{i9EIXY)%t<d;6k4GfBW!k?Ejk&uk2wDm?}}1
zGqf}@aweo#axrxNAMZ%o7?=S*P_r-wOq;N=G7!>Bm{^#ZI}<Xoa1zoBSvWh%n>Y&F
z+1T6Jn%FuMauCuB+gaN=D%l$t0ZtP!akVfqQ4|*fe4~tklNBKcC*U^~|IafmY;3HI
ztpB&~|NqRiTxLVMqaJNP_q=4i;Nm!*l6Bd-Hgvt{&U@Jfvw#GUk^(fVA{0>(VOJ39
z8R0ybNIbUZO&|>MKSCK)1*FZJoNVE32^4VWfVP^p)UoT!%gG10nHxX;r=Kr=sbmU;
z0+9%0vj6{s;xwIXZZ}s)N6VdV-W1gTdBXfog={ugTU*=QnKNA)^)D!O8uS=3Cs)_&
z)jE<C4*IG}Hr{tUsTr4VwjRFe55P(96OZo~H3Kxvm~aSTB=wRMsDab&Lwv5Q*3DsI
zkgGDqOHpb}Zl|+*O)I(|FFikBJwG2kKc6T3dOlA}$3CZi-zR>f*{M6k{5IcE?|PZH
zqK-d$KkqBlHJ$gznRdjhg~)GqX|knOTkWil=x;)%D2`#1gMOa}1jefSa-D2`;2#{(
zlZN-0_Z{wJ1c>)M?vCq?7D_5AqGU{DQgS*v9j@TXlBIuy-hA)gyniP_D}KDb%2PgT
z!wy>0>1V!4pO<CzyeS_dXw<T&-M%U$pC}uA|3O1Rc_^1L4&*P?wtZX2%_mCxv03o@
zejuPCq2Rq8vsfxeOF@Vuv9q)LrI9Z`dgM5Zh5z@55U^``SR=a<xW~J@0%Thc#N_1U
z^YeU9#P5=~Ux=BzyFf@(aXC`6qVF0-<-Ep>9$4;QX)>)mRB;puGh!FZRb|evIumfy
z<2H$Fq$yH&vF9y{6-$P7n8)pJ3_#cdrE})W*y~l5s?@#WLT`5T6$FsWt+Mk@XU<$y
zANWdm+^(b%)`)^DH>Qjl9<lnDG3?uU_tz#R%qlha<{qA&iK!aPLLdQzWMsw7&8`x#
zON)!j%E~|s`fd*iqzeI#;^N{|j8`z@H9g;+PKCD*ul#8=14x>URx7dNvykO7nQRgS
zA8{S{2@}TDYz#6koQ0=<%8GGuarMJ8yX+erAOb5YD#Er(hlht78(C_#p0y}bq<$w!
z7MJpdUmiYWp|2ngu&jkkHYq0b|FG6mzJX7e)KdqhrD3EDp=Rqj66u;Wti80F&U1kF
z0o{E$S65U7MW|LR(Ox+<Y;J9pDSzRBRvkkeVZ8P{NXJ?Tyothe;RjfR_p^weq^Y{U
z-Z)H64P!w;gop7;<hgo!zJEPqC7(RJDU%)Q(dV827gw-wN08q=Vz~F?fC%HsMot)_
zcRhXTm59cz5d9)3J^fTHb2^z8%D?sH0`2g>0gZ8Lu(Y(qH_rt&=gO8XzS1jM(eul`
z{Pn@~1L?O1dz%(BffzmbZ+aWBe*Cxh_iepW+5`!I9YA+uxNVKk%+z$-E_aNXgPU7v
z?4xq2-Q~ioB-(xKCo_rJ)ENMap#!Fr9@;II`CuuS$Z7go9v+@+NNR7ExRHtRacn+s
z_pO_<=4RExx$vR=k1OW1iL2Q?7pjEZ1VckZd4c@&g2a6T)pr26yteJYS_DM(;%eu^
z2mjgr1FVO%$#SU}MQ6LDf}CQrUKZ!$w$q!Bpplk>!k_h9hl`zs1q>W~kNUrh73lEr
zY<l|TW`81Y@$LKsfOOAfN|s;v8~}#n5AJtd+1Dl)qPL}3F~J2TUx01WPCK!|>)PnJ
z<DOrX4P-F02&C2-JgV24*C71mKM@cRprD{IH#gT4WxCr@Tg$+~fjKfDn51sUnFJkk
zypU4nwhQid&+7ab@9d1x>`(-<MUG}t@+|`V?ai2x#S%K+F@mA7BEy(b@ZePBs?&+P
z`rFMUGKywf=I{Q&&dkK>h|b19f(cNDkB`sT8pl`&EgS@B9<C7z=2nd!zxsjrY+H9X
z{)o7U$nNXv>S}!auU}Zv!_?Db#!MbO)$B?luoU;rk54Smt_fuBSYB=8$XQ@Scsgs!
zI>Q8jh2#{0xzlM<RjE>|P~VeAmy`q)pMG8jUyyQ>HCry7%hdoawMThrVKjr11BpB{
zJ>9V!`bYD9e{t5q&tG~(b{^402$CKKjRF?Mqe6jdU|>LK{iu^L3GPQ=kQEgf3FLq)
z<{L8DzGjUM6JpJlIb;a@?oF3Tpg}a2BMr1ZJ0X*DUM|r`U0q5|o`<*D2#eSKJ+o!%
zbdq~)n?-*sK?gG>XDe~_38U3wE8{}y-tX9Lv(c=~CGYI3*?+Lu=42Mu;aEQg_yg9~
zu((?<*|*&f7|nFdx7rIy;t1;-SYnZQW;e2K%5EedsGV1g>k2GczW5(2IM3N1!RQ;q
zXABap1VIYL3X!yJL#>%D1-BCuk>y*z1^Ykl?(Y2EAMHYvM8(7cUw?FL6)A3ypwQZ+
zReW|<5I3#yJLQ(#z}bCU?M?|j?aSok8RKOTPfzIN&c|8!Od|u1Yr3AXg)m5wkR$&v
zlIpBf>x`+%e~4zZVYrDeneux;>z^#;D34r{V|-`IM+nTljV*Qicp;C|(b3tyj4dst
z6g6~oaDRi@0-cIK3ff%sIwOmEo_jI`LbO*b0GX&B{!)0MY)^zQ8SlE!5*?1*+OO1z
z<h!AK@0{32o<&_BJaPnHIEADH0F93BI>3226BH7`+9K-)xKXzBz@Fn!4yu*LRv@-<
zMd)Go%U3uKVljKL!nRk;_EU?lUPR4<iEecw{vdzK5n0UnF_CXlYJf1PZB+c%+Q0y)
zWbmVh`uQG-J~sIM1HSY@YMIbS2z`s5#}OLUyXU=74z{W2e^*qk(_wnwzJ6$kB54eD
zljJges7;`s6&HuNX_(1o15{?}`&F9cWY)eciFs{udS!BXRW0SDRMKA7F5PhQ3eO1#
zr5TF548ICSY&4?iBeCYz6vrYz`a{n?_Rvx}+r7hwh{>?~y=>N*l%5Xd&3_ry*Vku%
zbbHJeMrT~N+S%&)5HGyG|8$=fvOoDZ65#lXUJRZsxPY=9$a&5J$qC1{aL$nD^;Gc8
zN&cD}Qrjtz_5XzoF9~*<WI>D`b%(OOy*<rFG<PfYn`FM>?;=z{&d#}Q`K&M2xJ~)g
zzQqxf6h0Lm1s8t=kj8GIbQ&G0yudI0^@W9niHRuLQlnC=arcx}k?#mRKtV=mviitn
z!^6`ls@qQO#&5@G=>hT7XXI{I{o5m?Q!ACwq5QW)Igt}E`FW`4g*&XF^q8+Vg)|XQ
zGT$gUm72;j?b-Eh;2hfD-{0<b!)LkpuSL^*FdS)VX=xXbznLlP9XsB_f<f?gtoKW0
z`;X3~TY5#bEMqapGY-2jHmEk@h}?l%-^Pqi`#^P4w!DR@y$7gxpVknpAN=fV@VHsX
zrJXY3$3TD=!j1hcuwz08^eYmI^jR!&$I9(QSZB6`KW_YfGN~GKBfsBf{MN7jsh8)!
zQ9^Jk69eFg42)9L7SxLoA|zOyUO8vps7b3H!452gJoN%~BRNRiPCpS)4b<N`*;YNS
zOs>#KXcAcM9^`QgEIp;5G4qz;Yh{z_!3<P2sYf@RQFiu|U+{u#Q3LLrqeu#vq8$@E
z30sm)>MwmbnfZC+)9LbT0%Qcl3S@HXma+VW9}MVP%*s3=mA;Sje%9gpY(O|u1E$mv
zR?W*ettoZ{-Osn9jJu8>ki|*1?N?0ON0E7+x_DgH5H<xcu+t<^%3cM_gnF@wAQk0G
z7xvI1B1l!JQjB{X>WM+pOG>_dEku!SO(^!tw%1X!b498&n*vinY5d{lTP&|pUIXL&
zr@dNf(4p16BVadxh`qF6Hd!^U&z*N_VO$sGC6~D~3S5GTs`yC9GHfU|O!2*EmD|L%
z8gtSZ<`-tA_1bCO7r{lJ|IlrytqSqr$KkGjGK7SM<#IX=zZ3R}4Dj&>PKy=R70Y4V
zx8a(|>mQu7$^vOFTM<#0i)HYtCG*K-@QSZH@kRG^H%jI-sH@bDtJlF2Fa|<`%x;%x
z(<|kG(vL>j?cRYYM*JfyL>eX>1`1EGeJLf?9cu>d{60gt_3Qcvb5;7`x5Ed?t)zQ%
zrW*G@Qj+9iv2@yTsOc;BlrY%Nh+$pE83`Kf8c#wRX)?O}VaeFz-HqG>^&wfz9Q-O2
zg>?Oixiniivl)n!eO`T*sJqIehg?pBgkJkl&2f0a4DwblEmD%-{@V+xy~dBYG#>%E
zpGcFdh=_=;?lyp=<4c@OWwJwpgY#Z?+#bB58i+m2L<9?MCf8L`TCU5bW{?y#mR+7|
z8oW>fs?Yj<T=Y=~T##$vbf7H?yon6MihpaFV-2w#C37c??Yd~rBvjQc)8ivj#w$p`
zA1aF+L2!0Q{tU@JnV_vAA6HjbU%C=K->aq-3KssYpAT1e73GE@B{?}NBnqw{izd6B
zoI@u4K4T`}*swXQN~EYE9(;wgiG6$C*WZXfa=E?8h&rxnwhVa?O4gODRRoPPTXr#|
zR;k_AnFh12GIL><t6YE#)4l`w?Ls~<tP`8%J-~FBS7xavFkp7ick>3bNWAx>!aA_J
zkBu*Q(%ah`fkLiHP+-^n1_9+8{1h7<9nAg{qhj{qC?wrTh9jseY=J_@<a|ly^dxAq
z?YQcv(r2l%lQN&<<O0fm&|SoP0<A<2yc5pVIn~yTzokHTPs%P6TEVg$MSJkjst7&D
zIJr!d=OOYBlrHJH&1$9E^*hh?dh@R%Fnwo5!SD$0(|)#QYh=NL663Ugh*174k0Mu9
z=qZpG3J3H6?_(Da%irll4tk~P7P8!24$WjXctX&GJ+9(4rzaDc`^}d@tc4~~f;|%V
zliNGBfnh!9+}o-;!!*CGvQZ0u>JJ)IMjz2BSyfVyY+)fG{JGO;Ie>L!?bAv{O)V@e
zOrX#fv+)U^_lBFw#pV7u6dtj{`n3WP_sL}Z2o8k3g0<?7UA?1|DGG=;@;I~-8Ke>!
z@Uo*e=~{;aar2Lwf$VuuX>DWZ>PS=!6F10`?V%(DfeYt#@!}%clZ%U}@bEt#+&B1Q
z`EUOfDxW760JTt4?g5FMmLqnALeuDxDqSG3$YRBAN~BKq8a>cP-8vNn?Skv_0f`*^
z%iL!ucuwjt3}+F?sn0@h0r&hSX++Aqt$<OD@b8Dk{Z&JLhUqZATpz8InGh-vGVYbw
z@LaCmw$R;LOqv`ZF!8UvJ6tZFzg~6#XpTasg@J-{|12>}@Sj1q0Kt)kWX<Vw361GM
z=DK#A4R;C^<W>g1Il$AX0dEh`f?{#YrMYQPUvmRzo9DB7_X~%o^f$sJDi+acAzo8S
zxh6wMlqindZBVU988e=G#yVcK%~dQXx(|tSm4}iCUR9#sDErp`FLg~Vt;fd3DwaLw
zUlnP_Xw$tQY4OB#b$xxf9YJ`VdjSHd;-g08MXQ32^xP}J$mH7*5NRIM#GbbT>@MwX
zmU^eBuz~~pi2r^>Y!2$%-htfl_`WlIcaLZ1Fb2rCc27}4cw8sG_2V$l;J_gvSC*DE
z2+Ekvrg^MKnb#iSUbcA7|Dt9dRa~?T*Bqzs%o+s@%=8`yO+XO|iTM=c)FP!LSbp-<
zek*<XPZdPBfR~}5`RHfuC?3_wAJdXSNR{?X$zb-yVjbsS&23BXkw!YMJUOQfI|fTt
zi9Llcn>nhpoL=tk>bIDe&@3sXb;$?`$(TOTjJy5^YW|2(OB<m6fq{d7uJp!1ElUKs
z(|$x}6|?ryB;410gg|?GB?4J(5*{XdIDrZnwT8klN~g5(`V+?kY!x3h7{%eJ`I2<1
z7cf9j>aR))a5_&8%`)z&mm^w4@`0@cVow9oUoE)f-qu^6zx3~<w^KR;K|w(rc3W?j
zY`dNA?95EBALE=UX9QT-!iBTL<Qz9{+Xj(2G{X&fdYK=WRtbAfu>2vd+zBswA!j6c
zH>%pbhP*v9&~v`NEsspc0DXF4gsGZ%e03p2>|Q9W!N|(HyPFXxyu<1-`^3RkDn;B7
z1^ho{f2i7eQgP18<tZ@XAS`lqw6!mt%+XXvd7?<0D$0?|srk@u<?GqGS}d0$tzW?M
z#`Y$6jid!OpopCwOgTYDhP8V~WtOtB7{Qto&}VbeSWzWFGCAtt6DKu359<Pxp*sf^
z0NMAC%ExXAn#aTt7W19}=sCX%Ml}qM7ql~i?***Iv$ma6j^CDo%eEmx)S4qQvNnY8
zqrug*0-Mj*)5`MBLNGK62q-9lm|41Nm+p*Vou;boLaUUgL>rUc-#Tx#SuL}Ok`qP@
zhr9OxE&ZP_tGEkD>;BUE9M~e><JZ?0$OHfYF^Nq0fQR^iMWc5WU6rfSI58;wi}K!H
zoK}oIwcgt;yqrrU+jvCydd|p7R4W{STlG2<2i7j_0h`g^d0Q5FA13tBK(YG2%KCq{
z!UzZZj4QZeZ+oJD=mv`P(e$5`B$l~}Wa$U$Tb4un192!Na@cm6g14uGL3pzfHU^Tg
zSFn`UvV|;>IoomNfjpbr2>|~n4}3A<jDinx1=>6hRLS_Z`nr&{e6LzUuVGJS>+=nb
z{+3E*iS~0VVy*^uMu394eRvoE=ZT-b?F@o|ujdPm#pMhPy5WcB{7Zg~01PxXa;(9N
z^#>i-XRzpIO66m8+4!eD&ZO)3B9eX7i#GX$M)8nt$R}VcTnudMxBqk^euD|`3CZd&
z*UVEC^+3!LSn}SDEJSr?lJc%e4?taTvq{ZlTM+zpJfT*u0>kv&2k*#%o9@Q6Hui2~
z9IW$%T9A>MJLmVt-}Q`z=Fdz2e!~Bw0q~lNK}ApatDdhmk_!K2nB}8J4FRDuil%Fh
zA(dr;8Q5mf)Rv$O>=Mi9gQfLR%?CmBlLM$0UI-G)kC3V|Jhvw%y#-5j11LS1t-{|P
zTz>$wr}&d?OtAx;wWpLds9eYW@9xiGGQl0?pT=%$9C0if)X$b17T*<PCh#>d-@PU>
ztuY_WZ7gO}9)#=~oep$-eBU0(kz9_po*w=zR)&}l8~l$+6UIeOCz8@{T*fm9x_Q4<
z$&F-6&U-%T=#&mx;n-JrUR!)`AoEaGbhAlg(dAW;l!(djK9(M#oms){RBuCFTOYZd
zctH`W5I9GbBImra_}|;UA+qye8KeSqX=g5)^#yq*^+_@_ac##ddW6}s;w~tjVn0V$
zrJE&OXz~fPWFJKjICh<TXJ?YK$y+>vjW#w==sL!(N@lGT?Z9%)&kGWAKJ)+qXWgBD
z1oVryc2H#ve$zk{5zVBNb~=Mn%5MXpNjlh2=rVJeWc8MVU!F{~Qhe0>^)64>cUz`K
zW!sVXzO`OY)3I7;mmqn&I`5q9Ym6m;v?12PJojf(wIM2F#7Ou8-M*gyCQjNkBqBao
zw!y%ZDYNTnQv8HjZ_B>*A!1*7P-5-XpafkB;zkX|DK6ih)*vDPY|dmj08m>mgwVc#
zW?RMHC5<?sh!G;}4YGNj)@KhfUS~LqNl=nE=VacZYc~+OAage(aQkzRS0|*)O~xNs
zT*d@#En47*ZtF&3BFOa`G+h5p!{Nj3OZ;-aFFHTceRaeU1OWt9q2d?P#Ir)d=_f~^
zl2OB2&rmSfuD!8k*ChD#J_m*e)0866dOdNupzf-$sXN*y6TL+;sTF#<G$I8=$sSgO
z3lVckkP^d6&M+>3XnTWiD*>4BtW6+?QG331X8FyBfl)U!o=cx{8@Fw>N)Fff$OOFh
z@tr$E+?08!fSfz~<>2t}zK=Z?9S0?489+`DUXtf}-*-4y*VK$89NwX45|fbBA)Y1;
zx4nhwuR6f{7a~q&lF1`2l%m$W;CQkeU+Dq#<ZWs3ad^3X6X^w>?4Rtkx;R=S2ojAf
z5ruApqwc+w+lkjCb%KR>P-3V3FcK4gBROF^QVbG%pzp!slJjfQh&yQPU1GUMQ+rb6
zgrbmNBx`alSJT<tN~z}no<hmOPogl8MajjF#>I>ojSed-xl7V9uQFX<RGlS&p+O8K
zjVa(n7fr-%DWPVYSHTqWZ#)(Y!VU^U@^Le^Y`}oUNiW^1)xuLAdiK)hd?1EkuWZH?
z41PAn`j^?{Z3XYOnv=3tBVXP%3Lv2rmi+W9K)}9g1H>N`Fh1LDOFy?%F(7jbprD}i
ztr=$)6cm)m<$jy9=yiK@va+JFjNDPO?P&qAv}K7mIQ$l66ZZ1(aNxkDiz1S;4$pX&
zUR=y2A>UEPsAf|mAYD$Q(VU5_ZmPOqL`Z<(_WYMNg#b`w21cPO0qe2-P->!)cIspb
z1HV)b@-OB>On^t$6b8({zjn(UiYkcPU^jMTQjSSu?+Tnq_jiESec$gCpB(9VewC$0
znW+^iJ;?$Hj<#Lr9mKB0vNekNF?6;^fbBWlZXt^B3>57Nd0dc5vx_ufl2{WOuCCB>
zO@XyLAB|91P1hbHHMiq8%JJMr@=3{`t3eizzfL}%A8%mj1Ik>|h5%A4<<)t>e`$Pr
z8Z~^clQ{V|H6ti7y3Cqw&Lm^Z7$lZ9#!31E7Y}c1(nOKa#nkjT28SI$=>g(NOhlyd
z6CEh#A$`h}pTDQ~BX|cS<ITI<f5g<v=+lxvHtQ4gw35m_H=U|;g!^{-eme>Pp8dzO
z$x%ypV^95v2A1uSo}M&H3}9&9bW8=ZKiR<@=q0kK&Y3{O3!g9ma(V_lw0+E@yDu?|
zr6)28z@H2(g)8#N`>ir^U2WUhn(VuE3&7ZBnx^+|lgWQuJ8~R}#Vd2pJN*n6(hzLs
z9^i3m^+q&Fj3N%fWJcpl@2w-tJyzQlW1%}~C<tXDDc6VQtaGUy0Mz|KsS%<ee<C@Z
z=CEeSsd<dwn(<8d-$<RI5yQvTvNwaX5eqx1##1oZ-yWyw(ttWxeQ&IA{HL8OcE$u6
zObF9V-~pds8patSD(Oba*!Xa{N{fJTwXL<4gOBe!$1Y&nnC$JVC&aB++|gD}Pw#hU
zPvr@Z#f(R*p2;`NhWBo+<H3&OLJ<j{L&ERfWv(}Dyw`j{Ix5i9HfCi^uo!oDSmxdg
zPKcfr1P}sM<*GQp1VRs7>H_RDAoZX*_uMoF<TA*etLm{^mdBF2MdSpap4ePlKwrot
zOYag!?n7fTyWoKtc?mKHAcE!q#`*sFIaS)ks&;<b?<wJx1xefiIcyRLUeG8`2Y5il
zk}II{d%*3%fJOr8S?9dx#0Ek+Hr7!1VAJHMQO4P7DI=w>u@^<B+iIrnR!?2_KVQI6
z+e{{y&Ol>5rF5<cA^J2^7;&*1fGOpn!T>}nP{~t(FB>5tkh83BJiz#dK*axK;3D95
zrJ<ra|F-Mn3kwTtxkO&Bcht(4w{O$dMvNOUhI~voqA7TL5#P<nX&vghPHuZA<ZWG)
zM7|(|a#1Sfh9OB{%3Af*LUSV!_f=PuB}4`r#xfu7;+8%DqeR&&S(_Um{mTptpgPPT
zI0eOoB{}J@jT<6bkBPueG6z=SCLG{_%{Ci~M{Ix_NP$<KbaIp|r03qsP61}TlcTdS
zqek^Srl-6GKnl}4G&Y;_@_W!2011!l-|fP8$+Q7-Ad7T>ojE73IU!G87+VsS0{jOY
z#eizZc#`(vm~M&pE(FWXAOp};b)>PJuq=kGfLZEaOve<Cah>enQ#n?Zo5JibrVW@N
znWK!;FAHmib?JY86Fvp_hnDjBeSd<U@i^=co}QjU`ckv8saC1-$4^tGz|d{Git-TT
zBx0#lOlnO_V_y|Ws6OjhRIbLv7j~s>;%%=%=+PV583MI-lh09atY+e=3`AHZ8&4sd
zbjCd1u{=YR{Y4U9oonh0Mv|aI)IkKWgUV5&yas@tUmZNNBS9~CLqm>=U;x8KZ*aL-
z_WJ;R>!(41E?mDCtNm$tS*j!8>R|w0<@NBCEBT>?zBoD(b%+6k)!nqsdZUB!Uw}oV
z!wJ0k+QRa+41()VNm7Kmt3RS&3g_34<7wU?V59KvY^q*d`(Z40E8y_%T!6>s&0tNE
zciD^hM7ZoobvtX)^pM=I&n0~zc=DW6U%e?%V5Zyx9ocE}SW6|>EYy3FNKN?Q-U+G7
znq-tqAa)WS9{%=lvJ-cFu~Ji7T+Ca`$3sa8eRJ{j4%y6&0|x<UU{&|^88YgSDr^QT
z0Nhfa0jh9DL*TECcIXWc3EYScxHAEzwlc*tK{EfEI6r>UsLmB)TuP~2%L`<-WF$@C
z`T<pxSGJQm&?t=rRteyp+VP+HHv5?J8yI?}Tsmt%L}7Y%88I`n8up#n5Arg1-SMlR
zHfn7Kz<n-Q+&|whJ^$(Qbd1YJ%cb(8Tb9cMyv`p$i^hmRj_>)$0Hs@^K9gs!p+lcr
zvYma$`+*at_j<<U5YO?~TR3i~*)#rNi~9qh7ud_}|9^3G4sMmTZ4|F*vTbvkY-5_H
znmpOIZB3qR+qP|=Y}>ZK{eJxcU0t2MpXXj{{nmX9s-t$V={+B_IQVDjK715ytN@)a
zTcPfB+{@;!_TrA`-ZB~o2p1tlrylbutS#TD17M#2n)H{Ksn-`{W1BhxJQ5h8Z{mLS
zNxy!@s53Y%;6DtV*@OxW$sAvy^IX@;*oetF9fiL*%JZtoK;OEclXX^hJUz<0IQy2%
z>M4N)4Ah}NZ@L-ue&xlSEsHvLbc(^k!UB=A?sZhx{7QobIXO8o%pFCV?FzKIlc1oW
zy(T89z~o5Wd}!fOf4%FG1A*u?VLPUk7GyKLC1HDlVH6`^J8|jTC}c7j<&wDZaauPQ
zF{}<AHJN?9qYm2c2D{u%SCf_5F)`23HiEYw46>VtEl9GM)Eb;zA?*-k`H}%SMIBvD
z2mIls7v8zgpP_@lct5URF@~~xZlVi#;C$>;c8fG(SEKA2Z;1vbxIU7kLrO9i_UJ=4
zRvRpWhoId2SULM%0NWen9~}G_PxP>e%teWFwR(*}yH{NAghGIMbJ11l<2Nwgsv4U;
z@NGeKPOip}k>DwX36Z($!loUOa^O%+tD)^yj_fT_-5L(f_2IbVy89jY#;28fm?RF!
z<6O{%S{#0}I2{JN^u!@Z`5}s+Su3p%TeE!j--B$NXer=LlCfRVb3rMY$i~iY)$;ue
z8ykD5zn>3bVhKD$BhtnCTkU#Ed=;YzmC98BnhRc89DZE{JhrAYnke9MxM{qgi42ZE
zhX}P+)9HQ}w8LQhpwG{VW9nV~Lvkh%h{_To%Qp3Bk0JUJg!E(T%XdU`f6#b{xo}gc
zPz&H;QZ(0I{sxTZMu`^IXjCLedZTn=D5DwFPb0yV8Uryj6j^QGRizfI$rgmm(625I
z4zFY07^G@ru4hKYF1d!Ej?M}c{&bZ$7~Z3*U|$MV6S{sx^I=^9Cv_N0VIQgk`mffq
znULq6A5hO<6P|*U0RiF}nrfwAU|>JT>KqbU*e3hMrMn75$J)7*bG}UzekP>gvi-yh
zL#Aa$kf&|IhUdsb7&k&O$guqN&!!+)N`Bq1ywHdUtHP7_bI|hga+CFXYmxD*Ke7N=
z!(|mR^`!qJJyzJVCGc+@Y3WSsgNhM1W!a9sH&4h7gKqlHSpS`K$&VN+$Q^ZJlv2A;
z6EMMmL6^=Wlf?&;`HDL?l~8L4hCu{j*9dewW#W-`+aAs>)WYP%g7kn`)sKq;^X(?_
zoBif4Ww;z>B?rPvC9SF`i27hxWolaN!-YS>S(Q;NamlA`T|{w8ll|rJqE}`Hol`$}
zp4{+mzEC=?>l#naeT&&6RyX_v`4@2U46C6GMpil{_5%-gc!CE7zcEDZ=vqN(=XX<Y
zNTfHpW<fa|RXenSr9s&QbgyK@j&7Z^VRz7BKXNd1{JlCDG5wet%+6N8jW#?O6`(Vi
zzu<YxWo#@+FU3?^91bRhhyo`eZNgs=1$d}Z!m?$p*R!{BuL4Amgo~3;(>W^nFFsqN
z!+?N*U^kN~+^+j-S0M_tYgfLTmFhnI8e58^@x6a59^Y3zM(xNMYQF+OlZ<AbtS;)d
z=WUW?t+Ac{sg*)J0{V(n4WJoR+-lQGdH=)$Ipu;+xm+|)D5@sh>BPUved=B@S%m}o
zUCPKgjEdPLt3Kc|oDDojzT=hLQE&RdFQh!4%JJ_=ST(#a)Y3v*c6}FPwi0yFv-v<8
zf!Ta{w14&{4I3#u8JaXAFjJ9Sl5>|iUK^Yq(u(s#%w~?iWKujsE0m6TNQTgc`7@{o
zjTM!~^f!wo67|nU0^ydZRHytIt?Ui#k&VOOmj3o-<~~~rO;$#|{4hFHxkN)y?0q`*
zq}op$D+068+qp*|opicWcGAVsJpPd<KiN>y@&)&+=awz4kB^Uk4_LB5I4CA2=Cb2)
z<K3~-?bcXXdEvdIH&qr56BF|DquY9wCFRVz$*R_VRi)z3zx?0O+P2h%%YN9s{|I^!
z%I>C&zhsM}f2IDrXZHi&c{>I*1*T0_Pjn%K)tIc^PQm4I?omHT1y$>ypIQ5-*aZQ$
zTWsN{`g=G1kJUN7Ena3Z(<m2PS}qIYCtbAtr-3f-PAiXHl7C|THqNhtkH92jVcqt&
z{<#athoJZT(vF;E>pC1Pu$H&v_wEw&|1J;k2@ufQ>(s-vHLp#F6spVd<UK+t{W8&i
z^9QqSzXRL<r!L|d6SMd4Bt#-`+4N>8eel9+IACz-#><M$$l$A2>yBn}A`>)2_%IYj
zbc&$@Oe0h<F1S1%*Ka(oWju9nrgWtK7qj7l_9Mg4cqW2rM8YOQ54%x<-diE1t*xyd
zshJv?6dW86dK<WNR4Sg<og9pemzABVGE$N%+DsE&SxF_x%wZvw&?I4CztXuC;yExC
z3!ssF)eZ)tq76Y5RQ9lR!f<V+f2<;I?tx)TOrBx}+K22I!ASaLhRR^TTCM7t3*kWx
zh>$eeg=arSBo-{4x{xikKya;0(u}MbSxMgruiD=k^3IDox+{$@JtK|Vf1*Z6dO1dL
zvctlP9#SiMs3eB5MILNsp=v^1QvJ?i0J%dyq#U6o@+nW7*y-+?c^xZTP4k#N5I9@%
zGjpp-5l+vT51U<nyKmv$#;pFLKsS+DbOfjbqjrj0Lhl={H;KqSz!yOjZz-82XeM?2
z4`oQCd>abAHO3}Rjf#r!{<K=%x8*{%{6b}dY$0)&=o`ci8X0*$F3JM(B&9Otbyj?d
zu{a6}3chh(iw1R`j12<5t(6ttm-_qYhR67pve|CU%_;lM>feCL4df|bl{~u2zNu56
zHlSePU|{7->Hbn|`=+q%W7i(c0N)eZ^sULisdp?Lz83v{_1kU)X4{>Sbgef0a3Q!^
z)N-mfakq5YO%l0KEdn>)m|6%5%%&4sKxevB=A?nJ*u$f@?y;31?CVKc)r!RKfCG=k
zP#+odSKS9z34wycc+kZthO8F{Qx1bB{qHPeoO+S}CwvL0d?=<c#h(<}7x|rzVXd2Z
z4)@^B<4jxv*+ZO$(P_0BD%R2dB;(Ta>(U8u3(Xd+*Cn}=f1IT{Dtg>0HS20G;b585
z_F>Y&lc})kZ@yn~9HJ|&#V##Go;hprZ%K&$AmvV^B_Z+E9BnBG9w^gf=o;c}%4dDE
zadNuvJAmVLzdOo1ri&Xvfa>G-eRzC?&KR*%3L&<QL1L5OmF516lB@;{C(3hj6fpeN
zKV#Z{DmBB63`F=yDm6oCHa3t%eX-px<Yko>0Dt#0_AINbnpVTR?1sN@-1A7WN1Y$`
zU<?w&0yOO|-SKrKV5odnM-3d$vIkm32LS_ZVd;C~iO5cP3IHZ#`R0NIQo9a%Z0+r9
z<LqBsGADZ`CeAa38<uLa?9)=q&;GAW$X{I*22*lxXJf-C5{VRdN{<SMwcwzsf2S2P
zt{T&AQi>BD=1kmFy>sa<jr@N~-JHrR{hMGrTJ6z0Hv$A*z!4HhQHU-Az$1{KAyMt`
za6&Y`60<h+Fh=;ybE}-?U;2~o(;3Iq%7?w8aG94lvLxtLcj+Q5&WBHqdUaoBM9D~T
z;6Q&vvBC;)C$jT@<d6*U4h;<f`eb{18y=g%6DPxFtr<8Kr`PV*ZV<Y?ZV@)<3E7<I
z%$>x=#Wgk42lOBA@6XRhEe{Ls*zRg89_3|j8T$v?X#WJPQ~fQCEm!OQPF;z*9UQ>a
zW#J}~SFn65bQLLpCo3Ymg~Kd`?F>c|X43-$u@fFGymQwSxDJFnKE?FUFeMmw3(MJq
zlkXO?;vVBYDf0oL8>NlXFuR2#zKD-<CiM>aU5b?_1FASpEwV9MPX0~kAtXj;En+<e
zClj2TKf{F^e=Bq}0Jc=<3RlgM-vDsUWiR}3P~_BQm11bGYt>DtAF@mqa0BU&Z9(|m
zXT>@;&Jh4F#8*DICF3RA#QxS{^zE;_%m`X;q<7hV)ssJ8Yw2PL*)hTo`WJf8+|%r9
zkY6A;fw-51SI~_>(*aw?H;ivtDdoiMY`5@yyL-<3F7fDSxVd>LBm^emw<y_+B?rVr
zPEHPh0zILFUi=<#rF>NL92{m@BU`dS-Tc|aM(EDNISYI3hQIPi1&ET+(Q#0wa>68e
zWg#yz4i+cJgu;W#7OJTxY(QjBI{ie=!ZMovaa_OWy#AGLd99jW*8s@I&#*c&+-VtJ
z7g)9VEgBeVDn#-F4+%6QDbD!IDdZKXLi{n8|02uc>u{fQJ0JtLK4G(dgE0)qo%9g*
z(Obdr91~frgFme@^=(Aw=@QqnxFU#+i(Q_iT$Z@$6mMB;b+RyrLIbWF#VCL`W#BkH
zG#YJ~BVa`q$fn;Q8B@!rU#0;igTF%6K|w<cnwG9v(5x07cP$v=8$;f8-+}X7q)Ig#
z`xc_l&czgh|85Tr?R|04Q`3z|a2peL17sw?RB+|c`cO_m-0foe^itf!4OWu{+wKNI
zoH9*sB>vFNpsF=>94sET6%1Q8ozfVmiui@73ILM;2K6Ny%75GYHGJ{z?heq9K#&QK
zMQ57Hd`7g3fEDK$85-&*JH!a}q!AB?D=?D=5kM<RjVw%E2x&;MU@e0u;y5mo(v3h|
ze$s}v)_;!Dq}&*B<g)GsC-9@*-!}k%F@)lVZI2|`PKcKIj_{aAOfR5~sw1vyn&Z~Q
zrC)IGW(o+Mq8kAWAoYW|^raGPZMzfP3$TX@OT8Mc7|{^jeoB#Jv1M1_Us9Hs<@0n3
zoX!jqRr;yx#-NEF8av&{&q6@ErrmPcMa*R0%6~wh)TRg8ntL};{}9vpO7(Z+g)DoU
zmFY)HlhpL9mTt%y5tK8|n2YZ;D7N{Nt9GR|wA)l@Spq|Je6{3s+w2Tctc0v^7pLT#
zb`@cDOctNAN-a2yzgN&uzgstps-$(VI3R(1VH*kOZefY`x{+NDalHnh@MH(}#t;v{
zKD)EGwzjILqKk`%qqnxQySpz+)>{iG3?zFjn&JZjo=Qt;Pst%o2=US5Qh$5KErkqZ
zD4?OEN&OO0H2ld0B0@vky0p2zJq+91@FbMRJuf&1G^l-I(UiX=Fsu!?@IbCp=>py=
znKa<&*MJX|c3$E`pQAWug%jEx(B&7k>au?~r^yblBKbHTMLYq_sb1J4K+uk}IX$pX
z`$P940SHapefp3j9->La-5lbYW%(N6-NIpx0b|K)h3HFK8S;v~Nf!u<{dt}s?p}S%
zYSclg(DA{$Q6YXwB%??xg3tZhkb2Bn0UU?&ALujKxD^CAxag_~EWKdn{}w}83ddgQ
z{MK0!H;7qY^oNV9z%8=5P5Ym>Vsz|xg8f+v|E+toi**}GM7g|KM7<*OL%ATZE)51Q
zCScD5?n56{lUb~>_v3|>AE~>6egOIWd-i$I4PohDC133F0yYZ{p*(u^C~1{;zhW@z
zU%Ko7Ai!7ubpepeUVq0l=W!WsthEmS5-T$^)e+e&9Dj4CC-(IQ)3H&`2xjgz5dZCm
z)CZ}zA1PT)zK!=s47pe&5x?_U|IkomR(IEup#2j{%lT%P&%}xoOUoY{rsI34W`hV&
zwxWd^Uk$d#$$&if%ulWMfDE(RO3N-JfmRthA2IoQf0AC&%%8W6cfe@vLI2}yn@qMJ
z@n5T~(tgz(hPQ5AYng6h=n<tuVUA@RnaejpH@=HVOzy8lxTel3MGbZbJaA-ywAeh`
zF-VJVi!MHt&7vAf4P+%L*$zX!{mS8u#RC=knP4=m2PDdRyOlvvT|o*ZmJS}Yw<@X7
z#MO$#e*;}&DBVqGvW{P3AEx+T^vffRnrh?bhcc7o72|36EqGjR_F8CT*W*k2ud=)?
zFx@2S*@GhzJi1Amd#kS(|B88wx*mG}muZ4e0HhVfPJkCkqwu0>@Q)V?*6JUJ{;C5S
z&#X2c|75mR>7|jy02SiT58xzJ-jh=oLZvR0iD~<)6Jt0@+8Ug43uL$e4D;IqM9!%O
z`eJv0pdt!q8t9J-v81~$L<fbsxRo>Zm(oS02=FU2`fEV7KdIHwIH2XF@LkkAQY%?`
zv&s9Gro~n_B$q>TT(+T|9%6MJc2ujmYY`$X<I0%kwD8sgbB?#@FUKET(1y4qb~F>#
z`4Kw!!`&ixni<-esLj6c=Z9mmhNp4;%VNn)-u>PfiK;mYoKr`)l~(t~^8lL$<UH}u
z-)&>ArU9W@n|Z7EY_#lPw0!F+GHROG**ZfH4A85;C(xrc`{~SMadhLGf5@6)TMwsc
z9TB!Gfnrrr_^Pj^wyx?z{zmt5Spg6Q>C%V>7T259@DY5s-PF}&-=Ov&mcTZiK0e_4
zt@};@Kqml!+!biVm;$sQ!%i}GTKL?Hz$9Lnj|YPSVKI%oqjlw&y@E3Y3y{ozG*J6O
zC~aaUk_H2$knX!7ci0>N&vZC(SgsjOE!!N`bO%k^N?mCj+6wjW;WPc0Gl-fgV0tC`
ziUD`Gjf1vF>dM4wQy6cVe-r>?!onl8_nJBHL-x`2B}lrD`!Nq8fubCs*gStD%$Yxp
z`z#HCb(EMBgN`}xihb6HKKf*~CVX4t0Qzp`jC-_we~*OG^^eFh$8I)ZGi~+c!nQer
zbGJ0$wP~QAx2^S0FZHL?E!9K-)X?0!U_Vf?qKv}qUOlMtbxS&=<3J#jDWFxAo~tpL
zWRB-Rb+CCzHYi^)C|$9j*<_f=j41mnIdJ)R?x5#&+tAMKWrjYM=yL1DN(uZD;Amir
zdG`g+@;9T&o5g>rLJArYbQTjgW4cU85*%BB0>>`I7n>ZC>Tjc4USL*uS%KhQz;6IB
z?}0nPU(8F?x7jvA=lnWEV$%Q~&pR>1=RW{-4PlpH$rex0q*4IJ=bTm+7*#1o+@W>8
zuBpkhYb||PmtLG}a!+1x=O4Z4pU!uAzqCN}8$a0M#?!pNJ33>xl+3I(sC{(|B2v2Z
z#`a}hO2(J_EUD4Q-hzobQ}Otr(>Nj5sW6X?!$ZjZ46XWugxCCJavQzMLH~4JY27U}
zZ=XA?{r$!1dhfU%FH~_M(C9=+q3>ix*#f|2^Sd$4p)J2y_<HBcH|S%xvmPO+T&O%U
zPv@$)aGE}0y9Q!H<_npOMf+FpWp}llfN>QJYXN75Rq0w0ezVey0YlfSq1y-Ty)Tv0
zwX)2Oq3{cx{3K=g(sX!o%v(8RY&-i*JA3chKzK2ab<?-IV)xCZnBFtdnNExR=Q2vS
zGS%xmu5WS6p}Q&929zz|69zOe-Z{U8HQlft*tf~X9aL^nNy|NGS(Niwmh<UW3Y61t
zd?k6ACkV6)QlR<)PgePaF=9lwX@zg?3>T_Pw^=f*QF_)Eo9xHxM8@l$TqN*%k((0q
zz!s6myPmyR)AAD$LJqL>5ryi=Eau)9(9U1`1e@jvge`re!8mZik-cTKsz(>lZ8-HY
zwML)-KhiMK=t{WG>WQ@%Hn%~2u}=c$gYI8rBq8q#M8-Fj`(6)|uev}NgeL(zUfZ7M
z^HisbS5lwgNv}J&MxWD^=CVO<_7lPl)C*ipK@lAYyI9QEZPD^NW}&cESO5_6vX~K`
ze5cjle_A-o-$M=(m43y20v-{VQ^e2Vf2HSJlu5(!zsPIs(KCunFY}I%@(Q0HZ((Uk
z6PY53D+#Mnf8qR^NA7IC!0^oKbhH3eGA?+7*3Y53!C7H)5cp1M@BE72>eKp0;$g)!
z!cxvfmzvhsTF%g!qT0M5xV-NPCFEb=jm0o0C1^9&^W42ZW5yuRZ{0EH*`%~Mj6q_%
zhZR}vy6blXrzZ5z#l`ImRoSi=ZDm-MYwfxNL6BpDy{EYeX?TO6qnKHjyL>f=BmluN
z+?^A}r=nkC()$RqgKx|~P=-7}IN`8Xl3K5g{wL}UfsOk@KX<P|b;+E8$*maA+^6XU
zvnY%688kUAf<$`!1z$kqF8AV5Sa%oKY6QwwQ)dAfUd5Xx(hu;}BZUeg%DO`f6iO5u
zF=pCc|7GNd!Z<lXaM<l`{(YXu@G>o5^Vc1*Ioi%{5}>9=gt$P3BxTes1~AGW&OfE!
zB?N85TCX5x<^W8bEMG4m9&mLVU9kD?#R`BdMltOYo{{z`=jVvb(HcnWNO(L{OJWZH
zF}>=WI^*Rn5vLu(Ssjp)9g&Zw!-^12*#f*F=MIqrZVu)LVpfp!6Nmi@eU~0Jy+~AW
zWJm8Mhs1&hETbC%zoDF1&~h5R$~R27Ks2<lfpg8|^EFB^bmj+cdl+UX9C;f)4TW~3
z%DC$Bh9P0%_Pk*Y>+&@#n$1XQomL5*))-vvLYeOprCCY*PEOl96aEVqxV3IRyP0|Z
zSd7H{?stCguU#)UV`WKQ&)3116<r_t9q+drTC1`??&g=7TK!orPOl5H9?qBTp2s^a
zjG#=x#{nB-1YY~2I0fR*r`-vE=!YiHPnC4@$H?2PEbTUz$K15ds@e^0=hs8gO)uWV
zUX?mf2E9%PhvODgR}cQXMWgvE<NTdQ34V2tT&Kv2g@9B?>1Anm(Cp9mhbe=daa<Tt
zA(Q;-IHsAQE*<KikNe0>1T3A3Kco?gv<0eU=X>;zOm~Tfle<(A35WD5SmmK^LwLNa
zc$|xPJQt;AS0ByX%In<~;$_*WhY-P8N_$mP7+!`Y0+y8mBvKicIQ8&8WV>|iVtVy4
zIWoN=4@7S!*93o_Tx*Pw5uxK<3V%+w_y-TmV+_EQDehIpA2KN2X?NdTio|_=sU9GR
zi5)>@;`%{MwHZu)zg^<>)0i=HK~nc-55FCA`tPvc+ZJCU$A${(#l$SrPpBZFT~U*1
z980XQfzIy?!Mg_+(+ko^MrRQbIJc7H5;I_yYUW6B7Fgs=BnMw2hzVX6Izy(1@~AT-
zJzmTD)sk|ly)p%ToOU1R%gS9AR4eux7tAx9KGrJ9gMC_@ELIz>x;|dMkIZKIe6*C5
z9Iv%FEd7^XJ)SL#bVj`+;&W(rxMRM$T&ykE8jVE>euTUO6fO`49-l4O?db(05dxn*
z<H&m$Hsem-e;Dwq3ZqtPt;yu?GN-`#e~a_Vr$-`U<}O*u_SecY@#|(|6=sq?(wUE1
zZ{arjR9Va_<+m(`-|UnA7J=Vq`Aq}kh69CKooWm_(F&c3#9Eq`R~uKW{nBUBPjN=5
zIBAiYnx6<UnWXn(?)Fma0eX-#&A*AISyt0QEEF?FJ+7N##gE&xW>NzF->;3}mpoy^
z81EAxpHCDZubuv_JtAOwJe9-|iZX+8i;G_C|EqqTgBaV-40D;xFOYsSa%b?eI%;_}
zAZW3Oq3@4jI&cvPM3L_~i2m@iLPx72TP=%bM7yfqR<P7;tZE8|&)OY3to_Q@810%R
z)(!{Rc+QD-^_;=q9N8<LTEHmiR$lyy;UW#ieD3@Kmn-A3rED;u-Peu$1OG9f;f2{C
zklmn#-7AINDYW6W-b&K@!)~R;!2pR^+h}+9Rr!<`o3V1G5iuK*r)L3Z!stSGuLC7D
z0MSUDxQyuwuGH*w{0OUqemJ)>`=?hQMa=8IUB>u$wsL<fEBKMs^|8O;oW5ZGeitX}
ze6}=ge;yv?e4grXPwVZzpRszs>b!6vaR0!58g5fjaEdf2^Ga)#((wAn;sws)=Iv4h
z&p9wk;O}cKB9xAP`C1;MRV*<6E;!1hofr+1CO#kSjqckAB1eOL6Q#(b4~(hyM4DY3
zin}-`6(Vy```Niv>pL9wX$<%LkI^1dlOt+O&+dt?B*o(7LOiK0ByXjKe<BoReQoW`
zFMLjX$%K~oI8szvLi(68-=Rn*LsL_;y`uvcAt<Q{Icf;=>QmBi+dz^6t>?Buvs7ZK
zy>QN}S8PL|(`tPH0QII)4pfdx#S<U336@-`07cwx{!bu0<wkTS{gf1u*YBYI5r$_-
zhKuERc2aqTz+yL&@RmUbDjVDLas3Su-}>{-ft3fDn8)#DR<_CV?da4&$J1l~Ch2f8
z>v?BwgjMH^%dtJ*)3z^PO_KA>)#bf$;4~GhYO~RHy-}3-e3Y?*){3Cc&}$ZY^}%Y*
z!S!%E3OR%CeFR&l^4)YcUm|XyO5^i<zO2c5qkf}P`lDZ#(|UZR&SZeGYtC|Q#r7%e
zcvHPdyH=*dGUMi6b|fzKPtqHlP!ZjefbMXRHN1_Z&tqbxE2i}*pKDj?<!_*U0uZDC
zHWFLz(k|ttiSezHh;6!Ww$$kNmmPtAbuAwuvYjzL(kSw++hn@L;G`zTxMRS<+uFL?
zlZ^yATI?GsxvZG}^0{{y$8SKFmxq^k)|8DZE>vHKHi6bkzfKkT+AeB`rNxr0U%$pR
zn>T5!7WQ&WRv^nqu>AVihlhtZ%s`li(Se}`=jWu`!HEdCQ_=W*a6|!dKK(o9Pb9Oo
zG^SHDFhb7VKhG<K-PbH-PA&7TPr<||X{m`oU)W@|k+|#gV^PN&Tu3&Pr{VHwx`EfF
z|9+g9Y4?4!Kx1rEP_^@;t~TOuG?l@4e1(AD(`>MJ7c#ADY}J<EXZQY1dK1&~nZfSl
zAl?k$gKmS<cG87%wTs@YRlWB9rn_+LeIausy}kHyJmq#W-OXdMas;0fWM!`2;&K@z
z68w<C!*2J{US5M<=8}<YdwmGI`Ns47@BaLBA%(L%K|{8`!;9!AQt@&;-zASV_=aG<
zUQ~{!iSF)k_I8&4G(hkGtWN$A!GG~I6GFErZG$0me-rGt&}3-0pM@09)B#>B`=4?$
zK<)H51zN&y%A~Pt>JfK72%z-*?jS2)rd)Bh_t_g(^EJq8R0a=;hDX7D`*^L<O2gGK
z{*8oXSr)$PrY&~F@F_lEpigxTZImLU#{c%|`58de<HF>C@^^)D82T4ziZF@(=-y{U
zD2U1&2<fr#@*scZ1Nud*3Ki^1Sc-!SZ9>lTvigFBi%ZmmK4GU-h0XVZq4J8c%Qbbq
zyxU%lCakq)X8p^J%)Veo^_Cm|_tzvst}D>R`mMl+>wUslAE(u6m%Hr8NnAsgs)dZI
zMWjY>M(u-^Q|V&c5beXJgqMfK@`ldXgF3X{s)@9ZiOJ@Ck*(B0`NVv%^*lkHKK@Hh
zgX8PG_gI>-&Zqa&hd+F!=5zN#XZy~g&SsmF#W7=-6px+$Xs+n(A>)$s)>bM%G}ajx
z6W{}^+^^z4X5qQrc%BurC-!K>9eBbs^xg0N@-?%XJ^}7~@QAE$FLVG)M50{rJ8du=
z?{{4A>$6%k_RlcBB*|Y~;@|aKq+(!}g`M|+0DyWima6c6;WVa{>xM{5@XNyRmvnS!
z&clYC%IYY_9ZRNOAF{!^tbOtlmTlZ)y)oI(drG2APFh}?j$%h=TU((T<k(SSGK|Ha
z8mw7SX!uopY2w6U?oGzM_Ex<m<HoGTsAy<Ea#IwmrT5ZMTqF8`R2#&I6_%HmhmM9O
zL@B0?awLUJ`(KW!bU{8yhb*95xg=yevEZK8xQ55F6Vl@Nai!IEkj~HR_4#3j&&jY-
zTX1|d?Ucr;!ok77<MWZn>%yU0kyf!_1@^dHw~vn`-EKIcQCs2p=t5x*zv!LFZez0#
zZm~~lQAIiAu=$zm+0|e<d0vmZ3cu*dv6t*LEN}7gnuSpz$cZsT+=Nvjs8MS^wuaA%
zy;&pk`ZC(wCniwMZd+B`A;atGdYj3=!0YmGvHp6yK6u32<lX7~Jaf!yg5>RKFK4Bt
zy4-AC7dar!o5lBO(vOpJsqqeFQ#(g6kL8np);7bkSDi!}Z2V1kfP)1yr-peNXs3!3
z>tc{1?qKzE8{!2m&9UgEZBi%IQ82HwD5Tz&aliT+=W}OA`D^^rf866G`^uB?5$ry=
z16vPpNnjT6EA$!Puw_GMSPSZkil~!ct8Y@+Kr}Ocq0NA&zO<B_oLo$!KVNipbp^DV
z(a?RrcoL`~dJD!A=NN1E!KO+TK0P)2X6sVx;m~J?5Nbf{>M3AWViih95`3DS&DBls
zB2!EA#&THbyd4?wzdrD~9gHsmk%!Cq%6Xg68qhup{H|im$!^u`V3IfAI%58Q6--R4
zY(waE@$ghtjyP0#r`--;+2Y#Q67J%xOi-<S*0j^$xF~yE=bUg3nd*Y!<%mRADPUP4
zpijT)RX3Dw)j+KA+UZknJ_`zLvSw^H%J@WY<YmX_ta#eTYQK$5<1AuhC+up2+6N%0
zOSPyQ0J7zEIct8JSGjeuOuL>KK!uLnXIxqH8r)bKp(dwJS3anU-lzML_a5vuE=jOo
zGWNiBP*Xr<&n{)JG&&?ZX)Fb?@5>mAL5lKhY-&n#1AqAQffVJ30GUV)@`u0+c(>Tq
zZqz141o?T<L{jbKr1-<ET59cNyfz=tt~ivxFjg1x_Y`@@K11+Mnh^u<vZYO05Y?7Y
z)uN!lZ*>+(BxWX(Zbqg#HN3-03FyZ0MzbGoBj|<_knbwQwC_i9eKA!-na|$6oJ~X4
z%Pyv$omCe|)@1klNjevRl8Bslf0~swA?E(F+dg1U*y-l6Uh`e(wz;Z2u1ZyslVC!}
zj?HVZuvQEB9)%5p>hsswibZJ4woE=Vt)ng6j!j<8*7uX{3KhvFt?%m}Rs)&%YIRNb
z$t{iElbw`cHS;eQRSN@MXuJk8o-Cg?`;aP?<Lm9uwN`SNB|6^EMz;qOO=)fa-tT4s
z#*<OAS*YB}>0~Z`G6czTtYSe_a1H12n%d~}0gmjI27ciQnO6cN{kv|jjb@hX3oIFF
zHZJpImbes(TEQPuPcn$o#8yG{>wo?+V>iL@!q`Z*PEKCwhoz;ZquxhS4FZDX%Sxqx
zm~jfQLqS~OecODk92#UX-AX;S>rn^6umv4Gg@W*=5sF4Y103Law`r_qExO9SCkyH>
zkIH9hr#a8_9OpXSP+F|l*l=8IxEwy5(`z)Y)kLI_HMpHFjB(n0=r(*<HdF!w<$lv9
z==JUPgfL60(~{8Pp-2z9i>Us&?ykgwE2rZNln)A<qruixVfPB5U*7}N+y!PqWTK%A
zB9S*=^@T1<<BNX}BD8<dIV#oS_u5~SvTK<vdR_~w7Vw)Er7P~`AKk|*T4Fa;vf4Ff
z9rq8ikL6r$d&4n+B8W98vkTnov^t4hb>Og4w>?dw<mjxbRv#h~RPJ<-AaM&|@hQ+b
zw-Xs)u#wbZbO4tvXE|~*Jnbfy{uS^io8JE18%vOE5k=wp`ArNgwS52K>dJNeaH{g3
zH6tTqsVz{EU9PlGN%ahx2@^g9J~a<__gpsb01H4Y(M0?Qt93=&sLkO=3<?S-MsJZs
zs_vHlcG(IpfnNkdl9REA!f>_Dt5=)v9A}V8)ZSJZR}oJAt%QOfwh-;SZZ~@=tR~_Q
zJz|&Ko;dk$8|EEm4Ij3z3wI4=9x@qeU1vfocZE|qW$&epPxAzccberFk2x(b;!6RS
zKZdjr34Q1d&y19|q5UGVEF?LC!aIE*`=~ULvmD7?_ZCaNiJum|av9f&I6i(|9hLc!
zcfy)-5l>wO7Q95WxS$GF3V1CABRVwFznXsb^bJ+&v{^QwB|aY@a{^TsXtqt(M!c7^
z@sv<%0d4csL5`2T68T`{s;7IIyK2gr^hXM{%Va3xt^T<-GZy}4p15fgG_;$;DFQsa
z7cmD!@Z<q~KjFj?!zk0;tNVK|0%$86ZKy?mkNnZ#qGDhpL0{!jcvy}xRnFa)DO^)1
zPh143a%Q!Y*oY{UjifathlxXCOjy&n5e!e2sN)QJV|0c5%W+s>+ro4bdkQqM+@r>9
z^Wzm>S%dLyV@#Gew9q1PN3+qTPtk0<)2Xt80=ijNTBFV9K`^m*y}D10I-^!stL5&-
zl=gjlJnV;47UWzOB9R@b<(ZPzv>1?{0mDEsyb9oKU;t&qqhy<EVx}|2y}dLbYnkwW
z43>Rk@SJf01Y|Lsa$xuit^KI9)wLLD9z#G>7Z$T$a@WcqFFA#J2E#j1h53(c`J^{7
zt!oRuOTj%k&+~<u5?(5LYx5Nc$LHFD%PLLpk2kkdFZE`p(_CJcLZ5f1cM#!T|F;C<
zb46^cIm9KpW9BqWwOyL<yLizOrQfxJ54)V32GFKZv!9xk14Bcv&(E-dKll(_5r#n<
zOEJp<LXl@%6Z`f)O^7Q<hapcxf*->7C?5ap0UDqQb-$UY^k3SHgR}|o(`;;iRSxVL
zfYXln1yP2NiaOol@wn09fsKWgn3QB<WaNj$?4u0!b<`n4#-u)!;B_{Z{IQ#oghZ%9
zxsPe4<pz{aVs{_~bRZ4(d$#@shS_X@4H#=R%gs9)-YA`hG)U*$Q914OhfWH95@2Bt
zrX*vX^{ULOuu(4BET|AEVp_2wtmu_45%Rp?TLQg-R=+fIDsAK}+jXFxO@f5)gK7Fe
z?98G+^XO}{lq}h=P84UY3b@U8I`ARr77|jOdZjDSbqG?>QCza+?}`oBLt-+|)2jcK
zq!6Nk+NFl)ch8w9zq~5*??}Hr=nn!CFqI;f>RcX!PZuwj8czF1;)-Om1Y9l%DsSUt
z1qnMWT;4%il0MHyEft9F8%-+jPz)T(giMGVlc5#>zvX5&8{;S*4XrF(AMQ5neQ8KW
zKGmBQwTWOS>%9#Q0}qd{tKjYJ9Yw^ChK6SQqe2<}2x2swmHQ{W-FhA|>ub63&uM8V
zV)xc#vtUS)O#XNX#S1z2i=#CFi{wa?j&y>|@H>c&js5rUAE`(XLPuF@D&EcH+1Z&O
zv?<1YH}euWFBTzT1mTmZeh#HS_>^Boqbl8mfz$Q34N_ieAc1zT#Uu(6s4mOBGfR!F
z82HZd=~5e`bE#;vCgAvq&5KObG8A#KTxW7vdHyPsj#FnUsW^e<@Q$5>?#+~Vd57;#
z=h@2+5I%ss`f?%wRCh^}el*rR(QZ%=GS3r0=7|h>L!c9KY^QcQmX+E@0PAzFU^l<)
zEZH&%eAxk5i6baW1Y8GF`xDuIiNdw7QWNvg<kX+O{{BHe)-<b;V^DW??S+Pta_8F#
zX%B0S8!e8zgGnlto_5zeG3eBdmn#i!w+9z)Zhf3KnV;(}W0n<@%{?1+i^XA0WC$n)
z(oTtkZ5Ax|RG-DgG!%HlFp`oB_WlXtob3`+<SF@2Y<zs_cVps*f{Qr958cFUrYI?|
zu2$YSKuaK;nu41fS61eBjwZ@_q#>@S8u@!i$={ulVQZ7=pG#AP()hIgH2h+Y*1$$5
z?tOB#!Q5&{9t#T#V2=^>s^YswjxE4Ufunwy%FBX;X3i`BSpAlekof)kOO-IV{a385
zFEr#oz{-0tG^NCC<y359Y<@PB*hg_06jthfl!sw&G~tdH0N&R^jKP3Gk>=A$8Nj!6
zGiLF*-CS+=0sXUdZZC6jb9uve@3E^Yks%#wQ{Dr7fT9Jm<etx<@Rs6X4XShI)kg0f
zXQdA~01p#xg4G@*NY(KJbfNsq>Q4AEU`4*=GXn+%Ko<k`QUZS!fLE(MYkAwV>!ze`
z)aKRAOHpmwQ3UGKVD3VFMJurpANNJ|1E+{at6-+lJTK>~!hMrCqb>o`t+-6Ci!X$Y
z&m9#DS(Zu{Qw#pko68*@_b>PO59et!b8Nd~j7b}v{P$rFL9z2UYXzh2MxA;INV%ph
zq+e^`X1<ODDw2ICi<TQk%BSzn0G5yFb5Jw3Kr~4z`@iKd=sm&a3!)%kHF<(TZDP7=
zsjIKp>vIM;gFv7t*O@bCGHFem>KnxaI<FC<NjoI_WtjuKf}h8BNB`Leu?mhHvn{}j
zo6qe|l0jOHKg8GR|4SbEZ$MB`ELW$_1RJT}7wy+o45PS#$*PPtxfz4T67wn=3-7(H
z>pD<Y8gSW;cX*iCEHQb^my$*xZruQPFDR4$bpTr@o!8|eElfuxB6(yYxr(rv-SBRa
zWAe2ExF~DvWr1igXSed`cN0KK0Kno|vff$%YI}bSKs_!QF5kMsl0Sdjg>g2S-0x@q
z+XXOht`Y`czwH7`aR65SHg1rmk9OG(@hq7K0KwWPWr@Nf&=<QAs`7zGu7VQ)OJTFU
zS&k!EjNkFbC|LiJZVSXAsorK?x{bD5^oO=t_1-$FYK%0`zU_$;)b>1E5-4-J24w&~
z&)as}B@%(4fX8g%agj`blJsGL6rlY2*&X$tZ|8ERV^Dpm<ly}gq?)Ao&-m!vgsr%a
zj5*)io&(2wN6Jb|FN7T=c|@KtiU3e>;$a!U#q?xTwGd0)lRI*GQ2Dtt&bgCDj7`n?
z2n!A)L3?izG@=@b-H#X`5z8;xx3{<cfB$Y(5988j+opj<il>P;=0-!OUYk8$tV=AW
z-{5?)7?uzKM^E{)I=xu_eYtpCXB(fR*<5wKWVWbEv&H`Q^-^My8we&YyIw1e8Xig}
z1s=b@PqEy&0_pEbFY;S(Mo98qN296M)i5KloaQx1rG@s-Ry^gh1Fl}ec(J;>?Kf&`
zE?c^hbV^XNFfyL6i|3aYh&5oRO`{dof8n*3Br<M)q!o(Mc(==d+U1l8G|<^(Cbz)R
zQwpDgvwjoTzH#+T!vDldOD6VctnoB?da{sL@aerr*BYbJ?91xe?Fz%G-JcqfwJp)n
zY-VzKeb7=qB`R||-yhR(zU*wfnm?~%)IM6T3W?)_5{6`0l_xkH{3H4cB@}^eXkdV!
zZpBUa%0GF?l7pOlxWr(1*?`QFjFsW6QnwG>bdOyv$IHic;ms0(kcW+*zr$$%A&=K)
z^>atq@K{>O@zX4)xaQNXb+|P4zoOBz=*Y*h)Pihovgmd=*`v!b0^ppNyIgKSLPByC
zmY0_o7Z+DoqYMXC{CC(ZD=Vw3<24J(;4Oq72!1uqtz4b#MJsKCBLf2Nz!V3x--j1V
zjSVTii&a`_bvBzb-&G!np1Yh6SDU<Pi9Tb@mdp$WWK<VJLN|{k(}3+z0pth$9ImM^
zuBrCVC2_(gGD||8E%*^5kH$A@WbAQ_1~>V1nE?Zs%${DybeX&n5_Iwr5tKlIbDh7#
zOo}vu-YM@k<p-i<zQGlSwqqo+&hB^Q2PM&23}!lm)FiqYg|uU<Z+DFt%@%d-2;8>&
zxp)7yc;ADF$XD7<1#9Mfh<IJ^b9Si{FIHOKAFN1HzTWlndsJ$7N^J}unt;yMJ6M_W
z)RWnV6kDQj1)A%c`X_%lKK_e}@<}(eOgRvj<ucWM|EzN%(v>z^-Rzw7eDo<=PvISy
zmOrM?S3whpf&F&3fr?TN14G*Z8!{s*iVE&Q+!onEE*ib@Q*?kK(f}Ls^m$<<bHoU2
zI%W0Oxcf~2z*^daHmF~<y0jMlnz-`o^#%a@O8(=Qs+o%WfnpJ08RX|v6zjWUAL}bl
z`%mF&1C5OAHOk8$hmS@NbZ>@W%s#KHn~D`f#HW8<qs-OVaVlyb|AHKO_}SPTudJ;o
zGxRupZ~UQ=U!;cI-(T*>k{Gl2Js$496|W9)TCe%?3ztNkn_PY|)xb}af3c+B{(9TY
zvKz81$w6?|d^Nd0Uck=Id%4Da^KySMGo#Y=YllZs-6UojiDj9ua?WX)HEE~l;=2w6
z@D&{C^#gGYu9X-gs*S_Kx>N}v6Lf}XRB=6eb>UoRJ_1(>A_QJu{Ir2md4E2hv*oAZ
z|N038O6}m;O@V$9@C7%Zrb*<*J`HN7%&qnJ2k0BhFx`@?iYaKa-sR`u4I1><PzsrB
zmD81|$IA5}3-;G%M%Q>;Mn+0fA`{Q)tMUgCUY_v$QG->nI|OCGiGq%Hq{W5};9TPt
z{S=9X6LIa_fUQ$q*b~JL_<OeibVMZJxdiNZT3>&Ed%oS!(}-_07>ay)dRp$VyaHLl
z!^0b~#P=M(0lagk7&n0u&r8(uE9NfVv8j?y)96RK<FArDAufNIA1B9JM+hER7#O;j
zx{TWGs*RSoV6B2IP(=vsr^m<F865tmSu`nXl&PKXYtNB@TR*?Buyf6n#iypD7P@0x
zWn<Mbzg!|W>+;SLwTjK>Jba2FN^GaKvfKCusVlF+<tIQTn`B^RJPJ*wq4xJn!j&r@
zSxj+NRh05gZo)cE9+o9VjQAWUPQ)*N!tAvvv8kdhT#|ykJct2PV`JlQN!!U-sc}+g
z4Q|%0GbP#C6?KfQ7|z`Eoye25`FO(pF*TyW_Ue^|1m(lutJJg>ZO=d%6ZcLkRoakj
z5JK0gl!SykCHUb~HefY^>E>g*VS^FyPg<H#)I-3&viO_h*`m#Y&Xv_wz(7n`akG6+
zkBI@_`0M8fuE~qht1+w6;Swf`nkOt)#`9xy`eS_dbK@km1WUW*OjguHOG_(yl4gX2
z38P`q-^D^?mJk!u>F!AWEaQACM`S8I&wAeFtD|pMS75=uco4tIo4t4cv~T|XWY%^!
zu-(REVOC*truKP-U5FLTeewO3@*nDI?gS;4LlrMQ(3JHN!|7bKdX`K(3;TL1{#TlA
zgbW9=sW(nRp~P3E*|r<*3d0VIfjFeRs0dnvjg1Wf5mEkP;KMY?q&qq>5$g;=OlUTD
z3&vK6i_S6t`^s)fp9%x-He3pliib}HjID=ig{rO3ZN$Knc`>{LDeOXxBFUIdQ^TEj
ztm}iUY=1Dl_JBKhtWSRj2*Jw7lTkD41}JWtC9V#(F)SbwmnIhv*QXJnoOm1##s$Df
zp=vTSi5wx?6F|230l~o{Ik}FBa_%-64{skqK|*Awx08Y@>|2+AlrN9X$bX5@m(j`_
zeLC>?gQq6+^awoyZC`{XlE)keef^yb9Wn9EX`wd1zaq0K8<11r;2#8;zPd@nX|;ZY
z^<M8$1xH-K;yhlTi?xT;R0BA+yrIU{6;>+~!dVMKc;dVfmEMr|tG~>~ojF$Negkai
z3NZzN1lh1xkU?nfxYep+dLfOb<lWfu>>XL)p;MW*G)su`M2)8A=Dn;dFi=B&=U{ex
zL?H>3vZlgLlPuK;i%G$1M(ONfT)Jr<0|J+cP9UeKXJ|MdcX?O%D6GyiQY@Epqdy;s
zfRN<iWasyK!Ki%Hl_-#(Qns@Ut)wty6WzJk<{rK~6DWW)WuT>X|9E@8EG?<fXzFZm
zz|X;jj2Fr2+q%~farb^$Hn}vLt$1V4xt}iWoDx3sB1RHQR;I62SpG~00e`a_9Im+d
z*P{<B`%CL{z_aY~e#d*waDII_2dYOdgh3DY({n{NgY>&lRxH;#w)ZQH%uH=<ZDC0a
zTJuZIz-*GyWOI3Rw&s0ntmhVL<LI})W`PvuZLjV*S#SiqHEz1|JJqDIkt=#2Nlk4O
zSBeUyjx!0h9K1ik51|+_&|qpzmmCa6Vpo6{qDt@c1KRIgzebgT-9+5eRe@u=<G?)S
z1s~d~e&o1%^3-v>i#lr_#UWeon7{l9iR1RC#hxE)U22<4g0j;_yBaV6FDxwFKjnz^
zVjVmw(S7#n1s&LbfA#gQYj}M$j>2Uf92<kDiF)^j5Ts>cv5y9w*Voqrdp@$SGLIUC
zB}9V}@gElO?M?SKL=m5ch<)DHT{cVbQb-T{?;rYor%PL)X<v1wag%c`4Pr<k*G+^G
z;V=_2Gc$pgytlNx9N|@DVRlwQPR_R$7D8akl9EYBXLBPPytA2$7nySlo*dckugjBE
z#{9Bd9Z=EC&(H6yw055<AO9B@rlEn&&UwhrPH$};xzZl7lkz=FiBX~)MU)VAh=7`#
z!cUo!rvPV4T9_R|;3W^Y5g|nJr;=+^mort|2*eCQBnL^h-{n=JCIfZ~teFh#pm5r~
zQO??k#T5+=>w~z7vYLrjtt@Zt(=QkE`DffVe0v7n6zuBy4t9Row(d_e8yi6-Cl-#1
z?kRRL0tZVr8-#UnE{5)|CP@g<;87H~+11rm;Dme28i}Lk<ly+I;c>s-5p{OvnB!7x
zV)>(K!$v;8_S)D(-78OXX)l7rJRgc@%cL|@w!smftMT2VU8zS(@j<7MwLTFnW8(%F
z98dj;8_qwG8~9;hKw2@Z*M*~ztI$k&4YnNkhY1+F{?0@~#Ds*1Kj<hZ40{7$Z4I*<
z^?T}B!>sa2uQ(AmC2fC4{gdo;J~uZpIgF#0B@%E&X}C6(5jhnQ?XTAQSC1N}oQd))
zi3i)M1w+t|!QC>=LrIJ;<s?&-iI|ZY+t2+x%6RJ_Z3@9A=UMALVn9DeoWjcL92`8N
zwfU9C;>jG?Pw{9IR{pKhj4jGCaZ`V2;PDLm!;raXfbOPxWR8%bByhX!^gz3}cBSU*
zR`tig#LWs6<@Y2EIx238klj&J?b~0bv6fbrQ~&H`W!sZ@nV6X9=;+AEz69z{OiTo-
z11f0^AhVu+aUl`l(i#DqmgR&)Lqmr}Ahi3ez$#^kqmV9$Y3g0NsDn4Eq)O@(A!c>|
zn{R>)>B*(a2H6!KHgi~UQ1X#8dcZ`0CqH}G_Y1Iaj^=k`M@NT$uh5(IM!Q=#CAUoj
z;Lv39IF&w<m77dvIypJ5x4B#b<|{UXCfFS@m#r^(v}r12?GSxZaOFy!3u9^@f^&T-
zL(Pl!*~f_$W2TkL*EMbT^HeM4IEw3drv=EQT!zJh^#fP7YO``*fF{>4V1Ra#d)%&S
zZFSFz2&t|{-GV2A|0lyul8qYj`e@UK#iobQaRFh9Zcxb~J%0CHH(ZS*k{(>A*K9-t
zn3-zh-Ag6^2u*ihy;nG>XO4G-7^M##T$Z>#6HKCeJ)AAG7HI%`%>H=6KkK!1&glDR
z(|`MBW@aWV!3GR2FoCx>i|j_a$V7am%7$Ri5YERll-<m8T&1SPb{2s-`5W02uf@i5
z6%)pFV&OQ~k&k;Q3nazhpGSF7iil4rGDrmki*krkWoLrYCSY)>^Rq+1N~`)2n!O~J
znY^@KUGn;+nwlCFAO-C>7xetuV*D)qaQ5tgH<^BtK^^i9er(P2c4fb@k%M5LQD^U*
zH^mJ#jNjh-y2Tj<)W%F_sSrBRo@HszFcMp)l_=mu>pJ|vNU=g4n_+?~fkEY;Q&be}
zo{UPvj>)0R$kq;T2^q@*%?`%)my4ifQg{od+lxz}z3lCet8|VW4niw<jJYnZ@MC81
z*x?&5C@^ZjBqH}HkAs^KeUbAQaTiwew&7Cb%5uabnyhB4>;B1kmUpRQx$j1MRr;Pl
z6sz98>Bo~3J_n+<`xd*BlG3~OChS^1uo!WtX5#!j0kfjq#StGS@JI8Tz$0vG<GY8=
z80lx*s}G!qOOYsL*5;7<$EnU|DV%0r?I%G7V8>X1+4%@T_zQ`A!#@##T9x0~0v@kV
z2Dtd7Bthl=Z353Mn|o>Q-9$T)N6Te}FGoVrt&Y$3z1ytI*fjv?9~w;S-CynWc*$j$
z>8O6w&-__Qb5ML*OQiOPy^lKTC|RH8&baP~<ZkPSZ`5eYJWQ&o08XlAM4?`u(zq%8
zd|E(VEa)d$j1*O0U1jf-p2uTiA9w=>ln&z1&(qu#1iX3D#Df&c>VMHx8MRV2=%U2V
zy$bs44U)X^D{m|fm|@>K2<uDCZ;%skB^NVF`9gnQ-=8j0D5%s8RViXawzxyLJN?!M
zP6k=Q_q&S(dq6e-PRHM&z}P4RQyOSf;_dLGHxDtb&pq%ICB&z$fseNujn9Zo9K{-1
z{F2d+b4MtM=DS5uH%W~uf>rHm8X}&9>5RgkKfv4h%z)wgaF%3C=C9#7V<~OM^<B5b
z@ud*pwR#dyPELl9iO=ZDq^&(;=Ogzu`^g^J%Y9sJ4r1`i%Ex3RiGZWc52it9%8bpf
zU^-u^wN?(TZz-hjSI&>A)Di(g5Q)9~7Hab!O2;k8cF2oS64Rh4u%Q^izJ2@c?_fu^
zdjhAWa)R7fslN+8&~5_;NZ`|@dY%0~v%e!^ftF6&_fV&nOSU@q&j1Ee#kffy?C7pe
z^)6J#&Gu%4-($L!wtBcF&5adFp5J~BA)<<Q)DY$M`7le~rmt=3<NdXxy}cXV0e>#L
z5*h<!>$K_pT2Xp<tA%&vlCmIRcm@n`pF*BYM)W>+(_;U>Ss!Pf8UIZm3f$oOe$kK<
zd?1P72r)HX8VbjCxZ3Jz`U@uKvE3WkgwkFlM&e@pm9EJ35W29^<Nk(QJz`!?Bg3nh
z^c5}90X%KhuYf}!#i_oZw`-=PdbOo`te|<7>NQubAH49AY<FuS)sh#b6lNmZHUU_|
zI5`@0J{q?FVQ%_|;H=x-A3S`^e{B<w2PSRemw?dcg*8Ua*LFy_j6Nh)bF^BeMbCeB
zS5&u8e^%GVK$1i&d)klciY$k09ItFx8Xva-O8c|O>r1tl;`<|Tiu$jSind=k@q;a&
zf9tgC-99uqTj!}}+i2}_csy9*53tn<m*T90Z3W#Q8wkUf1R)bwnz#BS#edkw>UsH&
z#l&YO(}n>0O3U$MCyWt{$Ez?HTrM8A2+E|su8x)MMomr41eDGV{4QX(CGakrnH_KU
z1p_4Df6GFERw_N<ve44PH6llb16O1u$ymVlYp8XL#-j>NlKzBN^#3@zrl?HYww-Oe
z-fY{pZP$~hCR<ZYo@`G|wrjF&+qO0T{rwxQHri>`bDcPJ^0>QtPt--LMXXVvX2%N8
zIT0uHqHg}DP?kikHJ)fs36JeluUc-vKgy?-Hla1p%Uw8s`WYF?-?(dZB?^O!qL1f6
z@yww*S?m7`HKgQ9#0<&ZhvIeOw?<hrI_Gp-WgWD&TExYX3G66K98=xoU+W|f_O@z)
zx0gv529=4*Q)Qc;yXPFA&pRmqCDzh9jJ4OvDdd^+M|7*g+~^!eHV%&`6)~<g1^?Aj
zT57XD6brOWs9*rJV-xjobaYfuKt%g@GFt@j1<lRPp$gWTk2|4q#p=!?K1KM<pkLWj
z-<nW8dX(|6A+AbMcg;ms^tEH<D=xOhLABqyKbxP==xqXaO%{B*5%YV&W{!@I{tEQ(
zQqU^>C&(a@o0bM!eyL<Br!=m~frv4j%4U=pM=GFQQTx{3zLTD@uT2ue=OJAE@HSE2
zQTB++RG6A6MdYsQMTiG#*$KDOuc)T{Dd5dCPZK``wKxHung27#{{#svOyQg<;~kH{
zllu}>O5q9o;r_#UPMe;=Vvx?$dvgPy^aU4j699d%A5+C=Ba*I9{bh{fo62inWU_bN
zL3JNRX3-$1JFsT71Kq2%+@S3NGk8T=^&N85pEpLfjCP4UDv5$Y*v9($`uuz{%7|UP
zj8zT2uOR&uTAb)@{em%{i1?nSXMDGC^AFtXY1HyS-=1+;cfQ7929d9qeN>zod6kPK
z6=49FSG$$93QV6eEs~Cfp#M*_D1{v(5(>KY#4)t-xVC$G9my0O4XxJq`D*JtD0y?O
z<U1(w-Ee5dysJOCJn69K7TE4RMNZ{cG;tc6{(C1+bwL?-7`E2ls>^kiC(_u{ycsFG
zNf@6aaoK<`^zU@Jv5wNBjSil^nA6bxCrfc7$ZHn72>@>%pA%bKxzooI=zmW`Z8`&b
zpl8R@x3K83QJD!kGY(r;*#9D*AfW<dM9As8?p1=A!P$OAyLFQ(oszEC=%xhCc>r<!
zbw&z*)>Q&@nqbIbm(%6imtmLBBmIdLtueq$TO}uc!6dHs_V%u=y(wvY8N#x+GF#0k
zw@U0peLcRr+6<(qyL)yw0r5LER)_khfNEA+iYn!kjX$)kqM~S5pA1WW5$X=Vs?3|A
z$czU-OH@>*TOa@&fbhV;&CM-uLk>B1k^)HzcoQ+V_H*|<hP?k2G-I2e5NC(^LtD6p
z(>9^_K{><JH8oA6ZR85oP3iL?ODKd&hEmUChl;vo(}@^JH~FhJ+8%j_6QXjhJz_Uw
zrmR22|C3i1%I`(_cD5lc1D3JzCy>y%XaK*pl6!_`EVS$@19*Xjw1R1=7GK0_3ue))
zOx(6K6ex|*)hwu6YstEsLJjy?|17AU2vo*^^$ECbY?bkkaj6ws^4Z&hChSR1s>`e&
zhE}UItp!JT2A|1{<k!@vJ$`P1=nlOqGlwXL;m-EIFVB?bHx)Ne)J0<mE$n|%DGlmY
z{|IstLbyMg=wL8{_c*w?faGay>3)=%i~b?8A6avzqzkads;bV{TSAf<jWOAQCs|oi
zq-;RGP}!rRmX?(YDoL`%WY=7J`PB~%6-XZK5a%fy-_vZ3y<h)&{=24WDD5W`tYV+4
z|3uWYz|}CCe}Jb%0HCzlo(?c-a(Qy*BCC;5I8d_W+|eXVSClRz$!`myTjHu>aem@3
z>d<d8bk(xqFmKY_z@9WZ)zFhuo8_s6*QJ6ke;Da*{oc5F3Ct!di3~|gDPdQ_n00?o
zk0z(a#*%WPsxKb){vNPFKd>Rf3w97L2WGSx>FM#gpK0EBBqykC>)!#N*iTT%V~w6>
zfiq!#mKgeYG8FFJ^<?QwFymkEVNyoBo`{uBlbINL(Lmn6RP^uDv#+FUf=n3kgZT_J
z^pM&?y!p}F3o#6*NFVhrErGcF&sQh8tY&?^p|E0LQ4WHQzs-isIyNnqyS=vNXJ=gw
z$5NJA^8Q?xsn`vv8`RQeomp#d#<H9@W*c309@Hu9S^yws$Lq2Z=wesdc$}m~G3{9V
zKg<+(QgA9-BP@bfsq+HU_aFzU`(8>6crq_%QOJQ+ho>wwR7q(@i2_;GUzp^#%MFF<
z$aIQD@C?vq9S^@XGK3L{aXw?zMb;;~4|4(qD~QJhBZ~9<vHdG89A<;{GTK_4KH0|%
z)z!ukS5fjl$9nu8*W7_rn6Zfos3{*GpRa26v{B1YPphIjTj4y^E__2oLq;rO!34U-
zU4ChL$48TySyR;uiKD`<k2?h|8o7V^+De2Pv#*}3vQd7G2@xn65z^lh#4wSOxo!|2
ze|E`9N+hA6J>TE^da8qIcXoDeY;IgQI4<fw0gs?DNNW_4rwGLWEgV;XIqX_EBHVnR
zzC*luLDA5_p5vJyQ{77BUkU9|`s|NG>fS6{<F-s6N-Jw?QKrNR+Mb{6LvW%Ep2yte
zTxSyz%b`X1v-s8Maq%c~zDhWm&B)<hh`DM<Hu<jCg0qqmkmVK-*|h~qNbB8jc?~(P
z_|=29qZaFYH0tN1e_D3?&uty2X}-}Crv1{^(0V->#iacefqJzXQG-30a<2D-aL#+n
zI9K&ZxO-SN4SELO-LSKa#0)mGFPR4{Yw-6Rp1$ESy5*|L*WF^^p0U=aX@ti)a)_GN
zIK6bmg-OBvawZbfu2&|AS!$=}w}wy6lN1wo>G)Qr0$uhjrP<*?EnD|mhpI(oULJFB
zB^3fE%9glCp6|y2i!Irja!cj5CwbS=>xDLB+pd=S<~@9cLU972!}n@kP)>{dw}$L_
zaTg_yS|pIyG*4q3B_2i!nxxwmDJNlb5!bjZMqhU=YGQv$Nl9o7FT>>67=kUxs)pqw
zdM^>5fZ*ws-wTmD<oiFG+)+tL3*RUMm=pi4WMmY?K#XCcq=7Vn!|jGqhiHC3KA>H=
zolPAy4JX-MTc#QdZU<CXR$km4j;oT<MSI>^6Is78e`v>}g6`tte$|uDB(p9c&-rfX
z(Pdm-UhcO4vb+kj?o1w8nwOnwkZB<LhxvSlYvNa+KJ)DjJMiOnSfZ%LfNRneO}b*v
zwCv<Vn@|kx%s`PRw_Xg({f7gk8C&Tm#dolARW~}DGxVy7$w_cW@dC75@R^`f;TVzu
zQyx#x_-pqQ#qL4sc@IQC>+KFtQY%-d(kc=G*MOO)A@~ZF7!z;7n}{Ug*ACmOQ=En@
ztI?)Ay~m}DV_GNMAZxRifT%@H_Jn*(GBRlum20I8MzFM?|9A~}T@@4*D5Ibuf;H*=
z?WdaoM8J?a0vFit$v7q_CSLr6iL^~jx&yP>R!~3?0T7ouTBw2U4-ccvu2A!Vmatfm
zIYv<(yW$h%DCsnsFqNf>wB5p160g~*fB1-X{Z1c;nq65;x+-gGzS)#+2R4SBnn39;
zbE8_jwgv=rS!_+Jl0s*T7_w)x`h-^mDFl^EDr^myvt^UqfqaTG#K}ch?@B)Md+(Qa
zyaPTT11bF)?GT=Ks-BPY0)?Tf4<+DwKO^j4YXUzj?2KEYrb1U(g}SOK2Pkcl1e+4B
z5|n<s>P>8OFFSB&GQre<4~z1U5!f9&V#Cy8MoQLhj*gxh;awh|Q>5zo+nZ9SKT+^F
z#Gn6YZE&mtaEs#NVyj}B4UFT#>=(MWD>=LI^O$?xnI(_;Q(p&HX7E|Vs9fHo?E-Jh
zc_CND5+@$hs8Xys5rJQkl31k;BKhExqFPj8RlXT@b!@zZq51gvFgkxelD`YEq8!|3
zCE9?s2F|xBQJ%p0@$IKIKt}zRTVx(o+0_T;Fa^*O>t+#km2jp%xCG=3-K3B`YHDHb
z4}fj|e1CGuCJ$4SS;f?jo%;_+C+$K^6A}{I)!dyZD4AL9<Iu#!f^Y~P2fv2M(Ir(u
zno<0tMvD{tQERkQWs5j$fA`s?P4gLXr3>~;UJP~1C-C(!jb0S>QeWHMt>WZ^S_33t
zhxd-BqzjJfVwf?SjBLpt@o6AR0Xd|ns|%L$7Ok`b5hD@f;pRq^Ju7U{8R_84q?Ru?
zf|F?Q)VfnTmBlmTa5<05B6IaEqx~=2C6X|MgYvrl@`eRBpUxyX>fZj8l+oR)k^|t&
zm}1-(lpt^OY~coT5!DXi{sH>!_l}p%*NPuh4t~ht0F~X_;U84}Q7K6-<~j|dXdk9s
z`nlHTD2w%~y4ipyj0m(D7riy)sqo8<@xpWrwUzc(BiKTe=HGb%3GUzuBJqbL<udS8
zw-C0r_0Jh&xt(fQt`3yMG>kpZN<tVv$7@wcnV-AJlL?c@5BuZ_s=u~|^3{-9oA-Vs
zrJjOz%uMmCsgUrh2j8YQ8%0U6^S8^3w2U?+)-{PY>G@G@2!^)wAp-rQPwNi|yzP<(
zv$exP)R9AGb0hv)PB4(NJYF=cton>NkR`-KM8D$BCD4FWJQ+|Lx@1Be<xj#^W3<MA
z`MAxoMuHBNr9aKWh?^et)fdNnjw=%ys8#9;*@$_v7%KWDO>NAVZ17tXIi0D(*1=sf
z%c434tG-AuZX~Khg<R&}3OO7i8p4RdK#AJ~UIg-1jzYhB4Q%jydRVB?;8YE}AKH`T
zGNDpu2No@lHQS&4{!pRrL=CVF6X9&ujR~eLyZJ00j?Z?;?~PZ3O#1dvnvJ^r4wv1g
ztKQ>uFpV0N#G48Dr^x4Hhu&VFNKYcZOOH=)4_FlkyF$N{{n_Igf?`iXQp$$ncoKsj
z2l41n*#5UhV}tO~wP^k!b8W!H#Dtv>`_Jfq%lE%(s^fnQ8d7xL_Dq36!DCu(BpaMc
zom|jgF>HwWO5E`*kfMkbC}SFD8QMu@Smc~*e>+;iImtR;^5hn?X36H!!~o+DI4N1h
z<)y<Z(bqEn5M%dX;Y*Tog1ysrhihTw#bbp|jY-$YrZ?8We)duI==r#qa0F?mUCPph
z<_$2=rLvlL10(8GYjKFJh`LZca;Y>Fq<=mV7{Oo%OdEfGOqjo_$VZTyv{9FnlM`Qe
zQ=98Ca2AbTrWp2vw-dUUE#FioIv*JsIi)>UmXcklK~cQrJX2*i#>6_ueG32pY)2`w
zHl|)6xnNlxj_-4`HQW<?m8DX)XdN9o0v_ev-J8fWM%?vYw4YwEIj>*)-T9p5Rm*LR
zu#?Np(YA@1RcW#M?#T?br$l5zvHtotWN>1nFrg_Lo8>xVkc6CeylS)<l;pvfi02$X
zR>6FBJ3p6~ymUP-e|C*mObfp&9)|zbl0=Ptg`t06KFz*M>n^R!X;g$%1YHDMuiAVd
z1gt3MbJ*~|r0r%Znq`U4ZvvD2$oo>+lx#)esD({7fsldzhzvenu;#I{z6C}Ar_(Gm
zn5u=rI*<X}P9l`;U(0t+o|{Z>Zes9IXAjO$URlqdjc_G^D_oyG&zRmEecAz6Tg`%8
z@@o`K3@A;FU3gZ`S0XZq?)KgAH7A<A(JQSk7XMT;fS-|&kc8TRuxyExysku0l?)6F
z@Qb1H)+X~4t&(^#wefYs`n7bevU^VrZ+e)o19mdUgjVcnfBwa5{yJWVk6^Gk#$M9S
zd!vG090y^799d(lz@`rjLehuO*K`fU%*HI)MVYb%Gq1vYs<!p%N+ExzS(b!ami+u1
z8_L~U<wvDdi8Y)HhB|38XLtYdz1)|2gV4mUrsnM5N76@6Z$z-k_Pa0qP4vtE(Ay@#
z#!c3g&yL1>xJ1oua{^I4;SRmrHMl}K6@tF3dh~&u@i`EVZ52|*C?`V3<lCR0FeJfS
zehi4JBJ|EsXh&yf2;ZkSHzx-)IkRaGV@#&-TVgOfT_@ux!1CW!^0afx+x;$m)K60`
z5Otrt8IJ{V8tXRx+b{Lsn&w``1)IEEfZ=6J)L;ko$wyhS_Duck2UkLPkpLHb9|v*n
z5+O67ii38p0skM&1Y1H%3RQya+ceerX4_`oRFH_3t{Sfd@S!y&kC<f7zx|0adJ$v}
zbg~0F_ahgYpef}rl^pgsCqUgMgA_VUg2W)aEo{3~3A7s_zZbNep)tqdSP)Svb<P}n
z7VBEj@y<@r7^X2CRETF8oF*>tKE`&fy{y?wHRDv#TF2_Zvh#4{%%Nvi1-l@wpiU<=
zc7)Y7=Kz|>1Zn=IX0v~Wxh(VQKb9KJ`vJ|r8}$6$(XrMmn3sF61p+ru-qzTjd-NWn
z=s^-#t`qbXo5e3>8fIy?FXR&R(Q*_L9dRYjMc&%iBhbI5f!iNmFjO+IMkNl72?B7O
zseyH)zpbip`Hf9i6p-wNH`}d;a}VBR@IkqM7PexAd_2T&QNo8E)E?BlMnGIEQ-2k|
zE1Jg}*kQCs4>Z7h3+~?q0~gC6_^|7dT$ql*i&d8>{MzCvz~4fKhxZ2X*dRA%78clL
zb2NU)yL{1~lH|#e?9}jMdS!wP&V<hVuDp~Ovpwt^USrZvD^%6vxxs$i(SirHC!qDp
zrNU@pvg{TSU+(LJ(Wc8w0~Qk}2fK1!?>hiQb0(Oq0znn51JBgzjgiQjjGddyQ)32y
zvNu462Y&R!WM;zh7aTuSbrtVXtNW7FIA4{vtz#elJMC1%jxsCqHcO144X0UgNo0IA
z<;gg@{=Y*3J*t{5ec|hO$!)-ztGfu8-t@y^l(bMt8FL~&4_PQ|m!Pt2dLK5A;IXq^
z2KUk}`fmK=UOg|q*F;&wM;D^>B38tB+z)(@bH!R!%~LK$lcO{xM}rOP^*i@5+kp_B
z*_>QdwBrb|b2P<fR;b>Ga`MRSYCS)66}{gFO*0#}T?hzKJIF|F?1pCypEVG$Sn`-0
z#!k14wjIRy!o>%fW2LP=u-%LKs@o`{WC5R3S|x0Rl`wVgP)$yDc7h*M<Ow#5<yzmy
zrs2*OCnuHK4(xht*KNwUzzD43z$E(GL~`)mGI5r}4qJX4>X>%t&%4{ke}-iB{R~a~
zy|ytFEwa&(WTamIx1j;4-(&0V`odP+^s19RLrrPx^Sw#{-TG-j*v{g~=%2FF+1_av
z4-c5#4`#z~jl=s_FED3FNM0?1!CzxsAa5&l2lKE5*opn|0T-GbaG#7l_K2h0|KQeb
zG3vue+l}T|WK=NNLu{V5yEDWDO7*W>p1yZ4r|n!_RcF#aE$Nd$D4c0NMKm?ECsd@e
z28Gs?>eI=U{+cewRa}Q&a5nUt_8x`o@YX4gOd(r+DOmqaw>XrfeWKBdRVZvDDU}N4
zck^5o1H+$=pp-*&DOXaxJ%1^d3&5ta;sHpJ)2FRI#q}jey0*5i1_o>Ohxl9|ZFe&U
zQuB&D)6=HoVvcc9R*1BmRKa>h0VVh8{C2B3qJ4a~PTyvcoZ$G(%*>$wu#goq$mUj7
zS_6-p?>6}P`D=L$fZX?>oOM~xP(hsl^fOtR)v~wwj95>qd3Zlip#+A-1cbd*_YZ|z
z%;P@huf|X2NWSQEJ5+a1(&1VQyylh;-)~k_{J|YjZqs73dNH>ZR&tA~O56BRQs%C_
zuywkEvUQMYy)~?CY+}38w{3QA1Y#y><XM(Uo4e_OCRpy0cwO60TkfUy5g!7oBdvnZ
zMwAL=P1*D28B66f$ZU;qByy=ADpw#{X{~dimur7G#uRAusxx~0fgTwxtF|}O(@HCT
z?ePT@9Rl(kJ==seIHm?NhltM*Lgnad0c^qA(Jp`(G=s$gS7kir^r`B<I5PenEp&<{
zfrEHGC>0ZqOZIDfz{XG*IF!|u5$9FUKdk{2B)2uEsO6BJpFhTRsyDnJPN2ixg?yWj
zFET=s7Vp=;)p!L}`NU8WRV5X3Lr_T0$^aV@75pLNwqJe&vqqm88_aKxbb!B$YPWr;
zD6n0$@oS#er|+>hH6`ONO-V}&?UGJg0uA6R)v%vg36h{vrCdpXuy&LBtrPi<IBQ<t
z?re{4i=7yZ`Ei9|YWg|fS|?u)9#gP=Nkm}yNwvUyE^qf)a)o>!?Z-xTGRiFDiSAc0
z_tN(>OTgPTyUKb7&U8rks|=+TN@OW9*8NizT;lCr;*GAfoQEnKSZ*XdS0=m7x^HMA
zq&$JGW?yA<D5n#L^dwIUWQtwDg7`TldLz<VA@X{FX_K#6^*qC^CDSsd=|&PX<!8ba
zEHchQnkPxzDVOOs=C&69<lW-%o&N{$a>!A|A)QiOq%$=gQ(Tm>{0CfSnU<|oR;Oky
zB_M5X)y8WUW@RV;lV1O_^8!<h)J?OfV#q0H$aUZ}gJ!Ww(tC*;FoG3u+VF}s6Fs))
z`Q{|zTJrYVFd;#uz~7<cX@=fm&p0==v--;-PH#`55q9-UK^7(Q4T@a7^2Qyk3;fh!
zVOn&Ne5BCU)CL|SXnk|@Z&lT$lFk3@9A;+!$$ldvZY9OR!I4LgmCktgLnK0jTk)O2
zj}51Q4WG0=B}B6&S*84gMVSZeFr(^c7k8v#jEo)c^RYR_#pf)BVElPKwWP{<(q-Ho
zdBS?Gm6Xp9Qf4RmODp7R2WI8IL#%~A%qVM{>orpw630_hQ*EAfuA5ePGxWseF4Sz<
ztD}Rbnq6yTF&Kz(Hk;Gu$>G+axvqe=VZ7sW|HA(3SasIfJe+%$zfi9K(d*y6@B9C@
z-l*Ac^#$k-@=8idi(2fm>(%Nrq~_Dy!%`ZoOO%!kejXIL87_PD?azV%Cmt>rE@)G6
z(N6onr})2-=p5BI!V);k`L!KpQ?FL)Cq4pW=HPFQ6wKe~MS!6vQ;sTWbSDfS&v|A*
zGwGTq(MjREB8;f41_6_Kpsvq*?_A>DAE8aKlOqmLbDqF*keLi{dCmXepnSOIk5hHy
zSQpZ^Q;qDo;&Zk_V?50oO?uQJ5jSF0;oQQ)4_w-U2K7>czr8`)QLg9^ojYNY<lrL^
zCE!cI5zR9Zwu`Y8hWQdD2{Xaq@g1+5SHB+4&Y+@Rt0(OfdHzuCOC=C+Jr>x}lN|`Y
zpnDaXGv!^zXIP(LP624N2N=P-^6fV|HH1`1%xa7mTN*3ND>wcfj4quovzT;$-M!ws
z%f#bub#_#rhxwnOxNOKxy&Nj$b*drBIObMFM%Jm?ARLE=L{Kmn%{~Px8-n(tOORH+
zKebws*t0GdPg8PK9PD5m7^hI{tRK6J!;h=m{Q!2KLx{9nvy*fSpq0)I$QK%QmW-zV
zbK^|hUL5~p_ZRkK7d7LdH}K#4wifZ`xAxb6qW9JH&zV#G$dhgZ!DQKoWyk<<lW(<@
z6z&WW<?_f`G3B(y<)?nT!EkhY$SkFz>Sm+&J*U7CX^|pgkA3_Xll6>krEzpRS`oN9
zZDUhXQc_12Vgl0=8WdV(QS4C7FUnl3LG3r@wZE^}sextpHcCue!G=1XC&zjAtd%JJ
z%{d}@F6HiHs7>5$Jm94HtHAKZ>9v)8%dip<9Y$UzYgu9_TMPtH8VEz;Z%cw2iIAkI
zPle)Q%zsf+Q*U+o8$=!f@>ZZ#kcT|O8$k#Zl`?*ILr9Hi&f&*~_IH<d5Sm0<|3~3`
zV7crT{w(OsM|6M956A;s4{p=FCq1t}?k+t8zq~=--e!)Mks5zIaw_RtpYFp+9Oe-n
z40oB$+}aq6mv~Pi65I%h<yeO&mL>-Q;sl%-yfr3K{-4Jo7p9$f+VMb6r!lKz+OXY(
z<10f8b`6tymSPBxLQ?q!&PHukAaS{FrkgyLYc{}-!P%Wo&@Xf=xydeY1F9>YqA$(o
zHhCG;mvvHkd*`l~Jwz?HOmJ}I$)Dv!!bJS>telVw^&dC}Q?}5?wUAL>Gm<pfjZ+S|
zQlwltT?TkAJ?0)9vh-w4yiZUc3~>VB{|I!Y!0KZqnoCb-Q~2V1k8L3ey@T31>e`0R
z1X(1`ZGhuUFRiCB{?@ZJg-@EFHxLH@kPKAST;Hd1FdrX_0qIO1E0`NaI8#MMgMu+l
zR0Uf2HAlRF8OL*dff3Pq_s>HP6Pxhi|1>broOHk%)rCW~iGNJlB;E(v7wzN&^01VY
z6yOHo#OP>npco5kD-BqCV4lY{o6})iFCy^e-De;Qf^1Gc<6E}rLs;KnC@iG_WF0|M
zQ>&9a0^dG0TDk-L>q~qt8u7i4nS)UquHa#;Qv3=>Irn}s7Ly>DsBmk=?w44x56z9i
zMByP>RbFA%DdKi5PHfxk5%{%t=)^G|ChHyiH;?Ztr3O$RwpeL=)(Nn$1?#)uguQ00
zjgjbz+f3<CMXY`-tG1~?Q*Cx8;!Fwb>-d~k5z%66pPv5LFfB|dufyrE!m4HB8Mv&e
z1xJdOgZ$E-Xlj!&SVz@k<1*obrq_q>dwH}CkSWAJu;J`P;Tz|qle7W&VU^_rK2NOi
zh8vC{TqRfUApu>}-UuQ^k@^wd3VzzDn%R$|EWECz%a1@%Qq4;ic|TvN?`Crq8N?&p
zza%o4x_!OnRan{B)P&JbpDVKMyFWCwwB%S!HHB&Zx*L&=BLe1rs=hJ#FwRpOWeJ%R
zW(`oe@_%C<%HrZs=Kh#Mmg}?fl4@wo^UUHueuwl1l(xGsOW(NB%AMT-poXlzr5g5|
zn4H~9a+snxyE_#>aD$P>)8Gq2kbI_mYH*YLwCa*rC&jr7;5y5auoapwbGj-x8=Bo6
zm+zHW>cTFU2NDE|M<0P_3akpVN#Qm(!)NRRaVSqX-}wlh|0!wPOZ3V5Q-mm0(jMW>
zWS9~U<}^z4-zWQN|Fv4P-Aaz#kgK4A#|Bo)U6a>Jm{K2qJz6+dw5cBZ@Mk~w>Oa*s
zb8zl^c~7}O`ck|Wc|}VzLAk8DxA~Z~amxHze?db#dE%cq8w^n2cgiR+75F}n|M-DH
z6p&Ef^m}|CVu2Qh$d{TOc5toX#}*8_%&&?G<?&Z(uyTDhq(IRh`*G4^T~~_5o?=yL
zdm;5W0uCEwgA1GAB5MzDk4*8at!Ku#Gv52I4i>1~EhTgwUmr%XvAzuSb_4_gUH<QY
zj9yB>4p>FOGkLKE-I)E5_LL1cBhS)*hR38VNciX}DIw!k71Z*?!u7ZHf?x|c<hKQX
zc4>RK^cQSNf_O=RuEC4XTfAklCko=uVoJ-@ptN{A;Naa)ZDYLjFnWSGL2_=KXxBB~
ztR%W-W?(oEczcl4Fwpk(ls_y!5i0gR=3nv@YRGGf#qA_UCFbaWZ)6(%PcD|nx-~O5
zE*Bg!PxNa#Cc{)*L&F^+>ma=kzguzKtx~Y57YlkuyB3d@QN<BLG&I6A$(;m`NujLB
z_wYMS^^k_`p9SE12w|ux=Ibi!qsVK{zeOFB|4C_K{NU^v@)O4}HNf9O3mXXLQ*8)&
z60IkJe72lUY1T^_=p0XdP!23OG#t)wO3)5hH2(k!EwFV0KVGuwlXtm*2WtV^#;mXz
z<z(vMo@_COaGnZz!U-q6EC?fmUq0*i)IIgeKo4v|4i7>12K|1F){tJ(%qEE*q4t)4
z=Skn=s+CrQyz&%%MVx%j{MyEXulmy4-#?YO8@&`F!$h$BS;Hx322(R!x2!z43dM6<
zZrX9e6j)Ctn#C8cfZIAW3HRXM03PKU0-TRXIgk@T-+%9=P)z-W9oVfD70@`uM6N6<
zXf#R_>QB;UxkLxbYnER3IV*{72Y|viH$N>h_|a)Ox}INT+P~(ke8l12${482E5~NE
zB=C9A{WL)oOtMZ0)}mbXUUIHKOe><JXtaf5C(8`yXWNmdIx^=nqVMI0$)*JcA<<f`
zBR=Qvq67*UDt8ji`o(#{>Rf;*G|9i0`aFcPiwH{y*lPd+yVMZSVJcyoK#t>IkAI?4
zmFNd>9fXPo{0}czQ9La4BXc8>Yv!y?FJS@<((IxcJl@2)*5WwwE=rJ-Rv7KEq|N<q
zLnI9+6ku38Tsv8~t+_{h%QMvaImcymc<)u!Y-u3I6j@1Yapq}BiHMs^*+~}Z6FcBF
z|Afp4h+h2)tIhKEYqQeNqO1!M3TWbF6>KOm`#M!nsNU3je$e_VHuyu3rJtEKoy}MD
z`#v~PKY9Wst&b+PAq%Ly0Gj}{u8Z6pnt&cBWm<<Z<FEQt-2#c%1hLksxK>0FgvEmt
zYySL4gcoK+;zYy<xfGeEhr>L{5)<I4%dYf*{Q@G>C`}l4I~23C!E3*_B%B|s9dq?i
zAuTb!iLpKS(5UzBn|a$M>g1n+`C-Q7smgjlE!4{|zir-LVAR<RRFSB|79Z+<PFa2!
z%HXmYL&ik;2`*$jiqfUE!pI#kuxRjevR_IeJpHh}v-R>K_9;u&2w%Db;2j$k01hDv
zY6TB^zf0yXnI>1l6cM^je!hyh0X!jvw_~Uc=#bxI3ApTROejA1^P@y8>8cP0KL49}
zt+-xjg8Cw>QT%hALa9kR6cz|J2sPa|0@-8l)tl2u@K94tDI2^0L@dUeXw4C2_AIn$
zmmYk>;JzlZK)VLTFY28Pe(YQS5S{Qluyht2hyO?ej3~bee8y)EZT_?a_-%W^q0L0}
zDILZIJpJbp{krDgNEmAl`-&@GO6>t*l)XKJ`Fz=$w6|TueWDB?+FCzw1JhWD*@%oA
zwd@KtjQZ0hZp-3>D)cP<pM&HbKJ;A(ZvOeKWDW2yk@NuYPm*|8Mr{%zqyu^aC+*xs
z;7r0?Aeds&H{57NcQ56a_HS}%CmK79SiHi>zH7n?G<nVQOn-e*+dD`=MAhSv{3e4q
zM!wjcWlQXba3XGm-^P5LlPpJ@C?hC&QRZ3lOfex*-E)_h^|Q^O8WJqaz{j(tK`~3k
z0Tti)vngk}^2BT7IBTPVqQF$-`n_U;t$20VK2b{@iCWPo_qa1#F*;e$)`9|naW3df
zHi}61IrobkY@YP!E*lkeh1TdrT9Qm6KYjx2gD+cvj_g|mvW^Ac`KWm89|tI1xIT}W
zzenF`s@O7H5lX=gL>(jdl~p%6%A8K!rrX}`YIiA5ZQ{;tQulfIH)<A6i||nTM{%}K
z;6+RPs|LSv!=t&3Y=5RY6S3y(cFgq9&-IwMbJk%DWCHZhkqZ?G1E~+LCLnz9{M2wT
zLe-w}dg}drhm)H`=#Ma|F>c5++HX8y16#Ab^=}vn<=>N|jL6Vj=BySqx3cXuW&LFg
zgJs81`P>J3!3~0&`K?M@yE!w1mb+_0l@DvlIX22mmNP4}!&IOOC&R6Ymmn)Cak+!~
zaOEIVH2t@uY0$i`I4paTV-e#8y^=gA2E|*1?`Ye7qI<vgrAxzSvi(tnb75IWQxgc$
zJ5GySXUWA);hyY>Iy4EyZY-FqI>|g9>*)I_1T<zzwFvLvb2=y69c`B(CM&OL*e`X%
zc{j<VeD&}otL=G*ZZ5Qd3j2?o0zQXVbIM(N$0^zHzp1uhi%g*cwp9g=L>H(vdI8YD
zh@=$4jjyC@M%1at>WKSo4CdGL4bBF3zx%4m=YBMh5c>A{Y2<f1i2F~)j8qx&3mm3K
zHBLT|x2ZC!n)}D;$8+~T|CgIxOANJ~eZ|^1^EYO!VWK~6ATdlwc6UP7HNgU+2V%};
z$q30ri~0gb#_G-&3o*Z7X6<9|_}{DL_}($*Yk@_MoD)44giLYGt5A*&w*Uu)oSR0n
zC6tSDDd>BFfo4_z+YWFF0c^EYLdPff-y8jB%f0eMccba$C)07eET8uF<jmWL(|IMO
zSd|VEUn?y>Y>#zIBTm58BJ!_&v(p|J4MQ9m)kvUflg2G=<nc^#t-P<4KnpEMV9@-K
zqO(Nb=~LvXlh6a-!3dZ@FuN1G&dINAzOMLH#UY+*()64x#KdTV!CPZS_&=bKMT@}b
zv{O?d3VxWP!wdoq97&M0BBo&~3>84oYDX>^*YvjnnwvU0-T|2mU}OWtkbsvntWejp
z&}VB=M@&Ssow@CzoA`e9Mw007HvN6JyLE^A5MQQDb4eu20+z2~U4ar~FF(No+YK&v
z-0yX;WEu6U#g%G1>8pOEA4RLonI&Horyi=yjN5}3bz$Kiw=bd%DkPH>5jKQ>sBVVm
zed0_|++bdjT9=blmf0-{PR;`!7j-D4W4yKl$$*oeqY*_7!V@Kg0WqYS@hx`2i@$!~
zgYx{jfsxwJ3N4~aItkKoRGCgOPpXqn1Q^>8O0Lb{aS<p`oN20;&|%Ny{+!8)=d;1q
z+u+5|9=zKke02ZO4^(L)7<7XQ=8iI}nR?XZ_&$%LM6N4?af!Lhlxori&}zUML^MF&
zG^(4xu}0pcm+4GvDsa?|?8<Nzv*l~|iCFnE;&QQAv6a&fxB*Iq+}T_6#S%AcNOaB&
zUsfDnU7u5GCiWG2<z`x9SP%5|$E*Nthdu(Nu$`UVQ%S@3Qd~1lTT#0nhO}|5Li|BI
z;7anGZVw9W%1=#tsUYdZ)r}3J;-WKG(HG5YG3R34@Ov>3wD|hmT;GJ^h;vMl&qP=s
zROi}(;SYNkI-d*Gn-zw?551oqfKQ?A`(e~~DJ?52D+&M2q0b1KN}*kCFZ?L`3mk`w
zBb`JqKPdbY2NIp3tj^Taq+x=N^owLgmy9X~cyK!8D7+p7t$YiTm#W-R+V)M^%tyF-
zq7T~6tDsFk5u<yumEenXq$xaJWq{*Eax|CbAjTI)1X{dkWWQJ!je!ju#DlCNrUi-N
zNS^EBJ!t~~)_aYhDhBRw`0moY>VzRXDpnla&lGXwzn<50!z|$u5)!)Tz6~k!Cz|fi
z%`)d+@!^uhu-)5|=$#qHt1-&h+=wO9`6$t#(#S2glIUAwc5z?@%Anb%$GpawTAQ*#
z>iY`up*W#Ukxj~?aLC2U0ZeGrzA}LQ;lX~IX1U7Yb&)pm{d9?V{S!RJYu{{E9X<U!
z+u^OQTz~YCMkV)u73@!50RMNJ?hIErk&jx9UVqVs^}95uag!dztKV8~HHJcZcsYYj
zzZB#w*<+<8<Wha(KgcBuwR&Q<3G`i|@)GHglupE9#WXr^9O4m-kfuc)O<N)(;f0Bf
z3VPoS@_N=p?UG+1kOl<&w5ic&4=`gSnk9CD9XvZacZKC+RPYMQi}-FL&Jp&;O%t%w
z0=CTS%fgLrPab}@QC_dX^rq+FsTT_dY9(Iti~#9W&3<4QX0MvG*`o&peI@gr4MR9K
zc)k_)ou|2csxLbl2{{@C{Gu|Dl?C6s7mw}JUqW~ug(mg4&0z!^C#lQtJ551eJuW?t
z`R$-gKt32P2v?m<FpS+vhA^K#ynp8<2;7w^fI2}f5YSw73-y2A^px;wdq9(=EH88=
zeFsudgSmpN2mR=p!40Hq6;wIC!^UJneG{AZim~cq>GvfWbVr#L7K`17c~dapMAv^z
z;3OozAFV~d9>$4_7&1Y3pu}21Mu;P=4?FFU1#0;`iwTiVl!{OkOUo25C}qnUqf?H^
zb>VNT>xS|KDeQxNUaZhu`eFrlWrx?UxP`_3?fx(z_Q}f36qPb=U#PR5B#3A=vUb7V
zJ1M%;Lk5a3AUxqe+Ai^(4vpwy{4ts~Pb(QjC$9&sdx0!3N=e{z1)$$G{h|ricvL3R
zs#lIS=G<ublzc=Hx7O#M#kCwZ-pPL;r=kWezw5e{tSs8h=_lnY`P7$jyRrbL&;!jI
zQ7rcr>)p*B@8R}5;i2OocTXNp;B4uAWcwRFVW3-j;1F}i5OvX<|EW7I&$wQ3;tvGX
zNmVlKj7vZ|2{s94qc)+|Q<uw#rOCQA{$CCPN=%nvj45UnOP;dpG}PMOtQ2K01-?f=
z^5BkruYG=7s%_A1GIA%~Iks<j#t#Q{X%aA82s=X+0&!41dd{vgK9DN*i)iwP9YNw|
zfioXaHzUi<Bw@O!cr7GQ=jlluXlL<f1nm}R7q&ufTcx4e>#7+Yu;zntTT!s8S@*y?
z!HyD%lG-qJTxhCG+Cf2aFlJKsV~c|kxt~Fs`Qr~vC1RndF+iD)<r{G%Lz&6A?3QG2
zNaP4;8(pZ%b{S`!S2h4OuUmTlv{1k^&Tx`SBR=9<wbougTv)f(ma`^77-j7ActqGa
z)vEcrbUmKhYz=5CIjuDENs{QbSodDGY&aXJ8@3!>l*ZHvQU<-^a`@q5#xQ2M(W8iq
z{ud2>*XHTP{Hq#r4)DOqMimCQY&ob())LM<DFLBPMRa@eX<%IsxDmmV*6cO}Fi0U<
zl*uH?7`^o3sfPA|U9{|`j`9?11Z0JV(dscXHD*aPT&_GRvYDX0ex?0G!h51!=E;@t
zqQ0Wor3266Xk7iQs%Ljpa2U{!1K>058?Bcl^?wHV!rnnT;nU(0>%24LK0w7{7bIL8
zIQvNh*#XbZ#yL&I^rh=us%w!pBUpdyZGMHnlPH!$uydU8Uz9M!uo}nR40c>tpM)pE
zSj{c45&M6=9!ospJ^)vHkm_IzsVd*H*t<4nO05<TF8`A_1_TgDGnHj!Wrhe)jx1rl
zvsf?o@I&lFjI7bTv~ZiYHiXeU`y?5np!$2pnZcQK359A%_EmF!k^nu<*%@Y(86-uT
zj^A^Ck0r8D7g!Our|-s(Bb<gU;-rWRd?yb9$6%4<Q0mPer+w&Zp4xQl1iuA~Q?5LB
z6<}^bSPAxK?k3UrZu|8$_WpeJq#eh_#RcS^gyH(e=hVv!Txu>{8a@oPl>!<O2|anH
zK`!Jx_apX*vkV2aR!GX=h#tuvT3MYPc2#u=2}n~RxE$~z8?yy6w{Xwkwr^+i(m4lY
z^s@u3$+POabUUlrwwefX7^XIPBUzofeAeKt4A#4_jK__c41FntN}EE$?6|&<K*XNc
z&)O>|FP8&LEoF)Kb`r+FZ`lUB*nTtK3^=uz_nR;2yD8Ex8L=#Z3YAd&61@Nn!Kq%U
zF(g#zGN?Kda_ak1AmBC_%)ckPK~)>K5c0-F9D%Ar8~(euioFL{et7opcjQWPq{==$
z3q)DvQOS3xmo(Dm&^>;c$q}PyD@dK&bN{|j1}zN@LWuePEkL~k;n?@f<o#n9$?51n
z%ns?pP-J9&R6=f~u#OL!oHj%1b=jQD-wg1D?IqFVs7W~*yJj}H(+_%ZWI@*1CqXU8
z`I!Ze;oSdxGr2C1;%6Ojr(UeUAkYk{HCos{aR>qFnj5zYEnvRF{Ba4Dybn23ARwlp
zfI-h_%@xwc5xsKRc7Y@O(1}l~;67O)XWy9ySeG|8t)8#H>eEjS$J2n3t{Y)=+1@i%
z-_qEP-Uq_QgV~QUiVtRb??pWbZp6FLq8?{BSCRbt3S$plKK=oR{?(6=Z_jyz;XMcf
zorpQRHUQEGP)MY4p*XtWMcDpp(hk%)ipMkANo~P|H!gnHuO;)(JGPRA@qA1HJL!ST
z%-5j#G6oJEMiF)I`qu~B#{A0S;=|)Sr<~HV@iLK?azulByz$EJpCvzf`#mMnICoh+
z=Lqh9Ajs{>(K@4bmT)}%%2AJ@u7=PJm9ZphOb~L)AiI#sl)>@bvwJbl=UynPru2f+
zvo<A$4#4%&_EBASI(@riCFAPlM{pA=^-b(oEOqlL@wUcvC|rmUY@+icNt=(Dnpx{!
zMy%Z0+>zh*q(cT(pS==o1kGM|&^*n`ke2z?{wE4T6Q(8MYDoiMkWLzeS(*{10fh1e
z;1SxyqLn9SND&?+@e<|-;Vg9t;W1?6$c`4KJW&etjc1D6QZMC1*U)4zuVw=9+USbi
zeow9D&HpfV;QDD$Uzp>B%m6tRdzDhI`ZxYhnehki-bP0f1p&xrvj79ume*m|smB8k
zQ(KVFcaOA9jmoFXi;K#ts?R^a0v-jwZK_3QpED0`rs9JvN4WqQMjNv)vr@I$3*&*3
z-gHK-ya8*t<Z->;3oDzv8GN|5Ab1UU{$KD|4BG<D5ij^oC4uLus{*2C={kv7Nhj77
z2qMtTsUl3uNQx^&R9Hg9Np73vuqCiXIv)XMNPt}pvVNEHtg=p_$}8RU-!*@a#>HN)
zxh^ZG=W{PFEHqzzL-uwAgmxAT*A^9|W~-Ar8wf11>yh3yZ46N>RKphG4hba-JJIDg
z<U*4K9g+sTxx}h~-Vdi-6t6Ci%^KW7k)%})z%e+V6;}X*PV89%vKIWKlv{ckO%{I$
zB}Z+xSYvd*L}eB%uqm2#gxB%Ffd+RTH90#BB1Ac~<AP{U1brTCfFT9-$JAfn?5%AW
zqD&{P^%v$`yzl*6^ZMGO$Im%aJ5+s2r3+JnVY~N9rwa)_j5e{0ZCCtzqxrcbadZTN
z-;k5o$ZBqevdzs*BU>&hhb_st1oGja*iF-TtRP8d`zWG$tcqr#{LE1Npw_$v$1NS;
zWd&0M3xJLnHIVRL#W6LSXa_YP&^Tt-l0FX+1Ot890Z-U*vCyedb8Os$mfzr+veoU|
zvcETI(BkxXbrSG-=lwQ@$?fgAJHi}hRIQK)h#@@;Zs*uPIbXm4EZcUg4UDd^E;hHB
z8h{l)mL9OBb?`nE$P<FvX#4OodlzNI80PhRb9*Q_f5h6LxP`~oLg%K~NaqISR&z(n
zE)+(JO+oS$C_XS*rw`{i(*>LXZbTsHMu0hhR;5gy&e8BS|2QJNL;=qOy`P`;!a-&F
zW3l4X^-f)7>A1<Wy7QmirUqlnW%%)&J;}HLAW8i-?bJNgZc5aNFOPcSz}5jEgG_#i
zUNV4lRu%WgU6qA)3r<*zaSx1@;Dfdugfh>Qci4hg>WI;aZ^*(l?_znZXPLIDT18>Y
z9q!?>lV0Xxh;7SK*fcp52%;qzb)K1}aK=I(-xU_ypArSpHlh}|`i}gp6u}X{32l0I
z3jvhS*YDG+t5wM13#53gVf|+K>`?p|hPdbQ5~4y0*@-fBM0CuyGr1BPsIlDDKOv(;
zwdemH%?Sb<{IRleKjPZYeFFmV;Za>nK;3ebJ9uXVB@k^M7rCzxyzCWzG7<`A1MoYy
zE7WEkmF(+)HwKV=Vk>X&wjW5}G`k|!Sllz4xw7oV^Eqzn3B+BmcRXL^2%wS*1|1vZ
z<N*05mnSEsH5D%DiIW-ill=RY=9Aan>)9yp&^%Uj?#{wkEu5m=E;l%k_qQ-m67*pU
zSyRp#WO=hqMLp&=F?eLrUs7z+$qeOaGfj<+s%Wg7zmPk6aiHww11T^Xk@3S*=kdE<
zovE{36oUO#6Tvh9Gd-ZyhES)9)WT^nD16p*@`+^ndvEElv}uJTPy?iMEiL_}QD(I5
z*52?I`P9_tE2y7nZ_Ykq;C7A6l)7wD1i(pel4^n<tpAUA2$BL4N`RLyg>x~aa|mWB
z_C7o@P_%ewDQGX|@W%zOe6%d!oOTE}K|z~Um3P*be3#5zHY`6US`l|soZ~`;Fz%n?
zFO1AD*$u1qChKyV8X$DX%69_iL<>6k_<Y_DtnIF60m#U|(J19-8()8)0{=Y+Uh6Pb
zP~&x8y8$M2-m@hjB{NhT7b-!NjFp1}z|$S7Bja9^fnMlGUlaJ>h?!$UWRKNUs}<Ba
zZJ`ADDi!vchajVtC|lz(<uGNrFAT|r8vkO?_wd<)l>9j8c@H(iW3a$ojoLqY2Vy1D
z;$B0`?~<g@#ISo)0EQjisJ&Ch*(Ob7GW4LgBlJyTqWA6R=Hl$T-5$ykO$w(BY3Ute
zmRhUD@Y2{A)8gX#*|}$P;|E|j?o~*e)2VTmGo+8IsF!t2q|#ErWB}AK>VDh{>PB*`
z0pYo>rxrMP`r)Spzvw9Yg@a)F3e`@(V7`HSt8F1X`1A<v8qtcXPImLLhdId3(BKI;
z(^M7sPV=B_S#r8y*QyX{kgGU?@n=bq1YyGPBU)b|ZW?~|H!Pob)LD7vRhJ1u13-yi
ze$(T^={%^~ByamWdc4ycmciqVR_|e!1v?H}C52M|Gr(Xm%}?uEHD-lhLyJTZtCs$J
zP=`=PcL>I?eq{kS0o2+!@+5g)A{MBgeexk%zcUjA7KM`bGGWLwFOnhV(uJJpG|^Vs
zGOkjhhbdi@i(4oD<D)z^PzHuXWnd{_eOf|$Id{4IQR3Gs1bFH4reWj5M9Ikb`|s0$
zc;Wy4E$!TGQxI2SXa`ORzh5*wOUYEs$#((7dCQE0qbLVwgLxeuSLetJwV-2eba*7o
zh-LV-c!)cvD1NKxb1v*9ALlz%7HydGkc?XQA3Z`{slwU`&Ui*(G!k`P!EEH{K_6fj
zjQQj6yR`%o1xt=|Dad08_b&Y{6csEFf(`_rc_X4zKeQO1OOq4JQ#?s)I{%qJ7A@O%
z7~^^fuW4v_R#so#G}$}iEq$C9T^)`u#>UFz3d8H^nR7gsoX;Ht-UXxRj?Gk|H0s<&
zz$cDTp8^!Mkr5Idq_0q%)gLypU~crMF?-+6yH|t8X?+LDV60vUM-{2h2ouXZ)Iw~1
zRpECAvY(0+)Nsib`QHm&cP)lLI(lrvRSJ|WhuW+Eh;;S@O^si>*U(n9#+gM-XVX>H
z1csaanaaspUaPx;7bbP^e?C9?tHpSY<YL$ckkC@9X-cYzZ#eS+-f~-MrWPT_F2Ups
zxhw{K3$eUww2G+{Qe=+N8F`d|tZMDW5(N7ZO?gjS654D-?xl*9CnyzAVGFo!R9Ph7
zy>{3KnUoU31xVuU`kpE3epriZ91=q?X9JkUhTu-NC_Hd1!n53D1HeVzj9CfLAiF)a
z{6E<wG&;5s|JIaOJ||BBAunu&9B`z-!Ct~iQ)_YXV5UV@{0fR(BzAl>uy(wY%Ji8m
z@N8)E^Pkkc(k0j8Ha!Jl!^1g{%M9XT)i4VXf#BQW&20H;W)!txq_x!JjbZz8Pmy3Q
z?K0f`3_2RDsTce*M84uEdpyO)KvY-GM<}O{ft}ubIEf$0!WuJU_rAyBE22=1329mR
z+wBDY&UeUgF7QF84c*Np?QpZ<Tu&KCLF<&QE5iq@6<9+U?9=u}SW~7rKA0dJ#q~(o
zXp$tEW~5%<woyq=Q6vYcNaAftV=u%gxWRWH%q}xn&T}LbIdY6fI7NU{p7*S~O(xx8
zx6NI>TadoN-?9A7Q33d~J?UGo8xQK-kNAcckU2b(2Z$?2D(#K^|Baqi>oAO|q<R_(
z8f(mlnW?Z|$Vymjg=L+$*Kx(?DjC=1k+$a$w$7S+AQITXNLAYINemoo9{<X*7JJ1c
z#^CbWf=S&KHE@P+P10sKf}!kfnEjFRF_Y6{lWGFq?<=ecZd|}UOICNf)uyUJG4dT~
zwPYD$+V*$;$j&N^%RS#=NX(=;Z8^sy`?;=MlhQLYL%u<7MMq8f>ko=}pMQc=557DT
zyl~S#oac3>YH>Yw&Ep@$CnTtu@Mpo>GFos|r-656dt-=Z9OYIya;wGeL-arso$oP#
z{zyQgoW<DSf;erlO$mOp0SDT+Jm3eZCTE1jmIY4)BwXEd)66fm$XUfza1g#LF8d*H
zzV~1?2ma@5qC!#<4=UZD%v&_gRW&1OHRf+va$8INNlYWkTJ0aj<kUUuc1RDQt!J_S
zG$ak?h8SK|QMT%RJmmrix#2GcD>n~QT{DfM^2R}n23E}-So?^Do2lKk^4}oK;PCdK
zuz=DDNViWlTHE<EX@B8HsfWNMr@W@6K+Xh_Xv&`$Eu_Q_<*12%OLl`aDJ`xjDyon{
zi>2Z#hAX*-IYk(m$9OR)&IxyQHohAvt{oTb+Ay1S93G5kwAO<DR+gm~Kvi_;^cB0@
zU|&5zQmRN-6V0p(fx8lGhKh~b(NI}<qrH5$j@s$kk{M2_&1kUY1i*E<fWmx)`}RM+
z3C_&MaE<`zN&D(t&I8}_;$ST(P4MJ~lE`bg#0TGUY|XqrrKqr+;O=m4HqD#-8$Mre
z=kgmd2+dN7#4g5|qXY?;3`pf=PbhSGwn!SDU7xeDzJ)T3;2@Tds$L;TtB?~x02E(j
zQMRR{qYFG*GQhsR-CoqX19|iV9M(qI*LLc3$9%>#Y5wjZfs6Dk<KyAHj4ET3BILtw
zUM-o1Y&xgY7ekwM-S%U{LDoY?uOS_^*xm?9G2-~&>%lwZx>9P{AV_La*$}&MrWbqD
zI?f+^*$A!6DNx`iVykpV^2H|xKYG!FOolYL_cG#=?4Zm<lnt9flS!cH%&8_%58<I&
zSiC2!e!J=1O?`v)hquSGw>OOJaow^#7mWTp`m;-pvk7x$KnyW}i$2|Lo8&r5{Ra|7
zwl)gt4@4;-0l<-LMA;;Ye|@!=Wh01{ro>Piz~BlK8_ln?21CpI4BLXh>$NxKL|Ba#
z4`)Bsgpd$*oi=@8WY26J<2`}R|6{twtoDE|Sf?{bCkbj8k$Cd<tOOB9^SgQUx-%Z#
ztWF#Wyz<lz=B(_k>%KfC;@`r{(9;vFuh$59&4ik*SCoHW<G;}0Mxclv*EsVEJ_oFc
z*0VBL39Hni_z-E>!qb|fwfxhVV>t4J^oa54laC9mx3ewjB^D&5z0ZV`9{9@1JMb)&
zy@2lZ#nShlPL+bJJ)`Rtu(#8OQhF(QX`_4NEpl!4(0_*-G^gx0&bh9!7{SVTY?Jvv
zj;<*@tZoa&RvX(+W81bG+qP||abq>M?Z$S}*tUJof8Y9)ereC%d)CaX86ZsXpwSjR
z`wdqM@8gJ<13f!jmPbJHtp;wLJHq@ypyZComIBx`SBC_7_%_w!rUSQt@Efqn7XFU2
z61R(z(zhg<gi-$n$hY}$wcyIY`05yHN?>Av3jb$``2l*{Dw^VyDoDlhx^tVf9S$AJ
z+pmHFYtRL2(h+yi4O^(`Q+#<}n@<JNFQD1;7ob{3uw{6sl6C9gX_H8CzI?6nAh(uQ
zyIyQD9IZRYMCuB9nR{JzF(P!i|DpktmVgk;FuH<-Z~ki`8$%vBguzB*bjA0G)e-q4
zlA><%=~C12GKBfN7?z_z&!`FB<9Yqk=Ghm2h4k&Rgn|~E*T#_b)76&e>-`a>f}hv>
z^XA?#5FuTv+2(XR6sy10`gsxx4<Daf%I9P=HJfy&kb}+mhH5o}a*{V0wfiF~@kb82
zcb7!=eWnNO8Cy0l<{4s&4m^Jb4%bvDrffdbDjl~r!(UC-zBI#Ig}qJvFw+c6R-7Zq
zr5ypB6ye0-+gAn*q<1jo$V_g+80YOq+<&LW-FwsFliq*uJ!SP8XU!+D9j8ubZnSBi
z^m6Zm?dHgJcQU|H<+m!@sx$p`e|9|_4^_DiPvw4{ObaNyKkv0XJb1?uR#lcIm`x{R
zUd@EldcRmkdss$;lo4zO>nv#g-KL=w;}%@aAq4%Viizu_fGNb_PTj15wNX>Tg3LUn
znFDto77{wyizY(fN>!J8PcdS{mH&(9&o&%P6j4vWh(U$s%{+Bq%CC}<0Pthtyn4fM
z#^17@q`ZX(Y>wQSia<{I1nP`O%f%9+EAWTs<Q%h1(_-@&rkQkEieOkP1oKcG`2yw(
zh`Qf*DMjsGNPj}(L%wHXb0A9F_fj$Cp-nJjG1)vo_%&vF<#oRkZquDX<D$OHWSNXE
zA@pos&Rssn5KMJ$LYUeY)VE}oRumZ%Qq~9aY|O^)Zy*2NbNb6%1pU)`7_sGGSmSi9
z*wd2m=8d7hxJCN+_{~+1&ElAGi97+PlgU@U$Thdu`AXtoG_<NJX7wh!&-AJibWI%J
z&E2!g6Q!<)hLV5v^;wr8YV)&P_c#78$73CvW(AM3@Wt@yH}Hs2sd7kfdRiT5{xoc!
z$iIt>$7oaVf`!HVhg#`M$hJ_yfvpopa~qMY=8g!p1n<o3Dw_Km54G8bP4U|^r?-58
z^Y)<=ekkluN{W-ZZgV=8J-153S8|_996NaI8g%Z1ab;<rznu<FMjkr41-vhZ9<0L|
zbUgBEmZ0`J?=xfX1BBhiBn*ZiPofQ{AKxV<l=RI_i}@&55-3VwS&dQ%Q~JX2$WS8b
zt1mIs*}KlAe)aQhkn!n0Z`hP=oLgv!?$b0LNeJ;5khkH7Xygdzg|1o|R-%i|ho8#b
z(?azi=STZNCh;QMg0?!*iSRYpjw;tlt~?`waz_suEzlyAI6EuTq?lQ#I&!ygLr<d2
zcpye{mBd8KKREC_oXii8jYR<~si1SJtF)+aV$Cl0=AHru^yGiw;M_Q-1Mf>6CswGD
z3OwM0)-`jWn&BA{aY4%JP8F}fc=Sx{-KCHz<rz5Y*fyp!z8MIqy0wsVCvmz|Co?1~
zQkh^Rfi9uy$wCcGPcyJna5IzW>|Gp9-8g+8)KvD>SpX+*(=Q))s~C4<LOrr%HnXc<
zGv~z}MFuhwJDd*YtGvbWySVO}s}FG=Gh9W|daU05IQz?TKUo`Hs9z~j3Lo+!q~BG1
zC0C3QYcp9)eehZQlQTS2wmStApCEI*{MlcLj#h&}lRawK0MZdL6-Hf;s3Sf(WAU3s
zy=3x_iX{TFV@T#tehsthal4#Vkfz?U<$)ss*z7y)<8%FkU=Bmn*riVl#tZYW*^Wf(
z<pW<=s<Y@<V(lq8-Y10PtdGIe$TUB~p61J8!aCP1_lr&6%aUA>Nx(*;c{eET<<Z3G
z-qdas-|oXfYF2<qJED}a+Z*pf?5rj&ChB3~qDt2J?9<)%99%YBeEVHm2Nq4nJdAvF
zJ{8|+^16kL26U-wUDVftKb5R0x_cvcn<ZlVXG~S-PP!Z{HqusWSpy8e*kcfuvLY94
zei)t((kAvUIkSZ2x^@R>HZfQTV-27MKa+l8%%UjHb$Pq<>T|Q_-FWUyym+$eHh}s;
zPbSHh%x9LZV|3JYoj)@v8qJdoDT+vNMxoxvOyxuT4158jm4mXfcG{M~RcC|(OT^(c
z8jh|#2+>O93!MG!?q>2(OAhzU8(V@C2d;;#1|JMUrB%gFM>jD(E|dSOLQT4<WGI`y
zo+9G4R`_@pRc+w3u6qppXJo_A361;7Jplp|?p#hsFm1HGlO)6k@37_UEI<6hy9F74
z4BrO(f9M;Hgsv>ryyI`T0SXVmIz*3UMf__FKjbF%F+$DEY7Z5ZJs2Zs9_akKA7Ro*
zboP1~9h6$3x~Erpkur7CXh<vM!>4f7M6Xbha{i(hEF>>xdoqre7#udJz%T`<j4l;U
zD}qjg)5zm?Mu*|u=XI`&m|8oj&8G2H-4RKN4T0~y+3KTn%G+fu-(91{z)4{4a>}AU
z{`E`hzsU{+=g6y-j;GDLqh-G~edY&ew;u9|?uK4p@ZF837vJO0l`(x@fqaeAy(%>-
zG@2B+f{g2Hhss550YqvmzVPWBqNcgg?77JV#Lc!6RLsgcL>3fD+v#X9)C&>?Qs|$G
z1LDVf)w0O8-5nI++p$gj1{FbL%EV(vlv6?XeZ}xeWKNzzLoLw4ETVJ)X04%solFZ%
zw1FGjhu5ZJeEnbEO`IWxQ~AbppGyr0OisRqtAj(p9QyHmD}?%pscg=IR0i;wkwPLA
zip7Oar;?$!{wUZ&ahO*u4;}#ij*z)cAvE*p_#vwjE<+H<JVD?Ea)(wY%xK|b=+8hm
zxflu*)+Iv8bJGS_+-BJpFPwdTdxM66fPjLcL3a`m(1-fDKAvvMAiYnv7*(%cIt~Pk
zEFHzXy#H&m?ydhX=<a5iFv40wJ`1nO@hX{C0ocr8aTy=QUUr?YF{;sQaUp%wL&9!R
zQ?Bi^e;3PT+NzK_tIN?U;^7^Cs(qcx3yZzKrVc3$4s_Ke5KbO;P9|1P*C|x@vL@EM
z4ieqI@zB%5v9B5$2kJ1vlO~HS=2|~6W=V#X2E^-7&1j-};IlsjuO<nhWM7r|Ex?y>
zZQmhlPLudapAZ1VdW!77v<_0z+4OpRWjuX6l)D_u|6y))ak}5xhdGt)`+85!?)W@k
z5-y+3>wBSPG)Uq%rKQ&sPySK?nV$RarhtxGtAI@Q;sAakbcINefsl|;br1LqSH*)G
zLt(=Ew;w4+Y`KV0@Ld~9wC6v1A$g+S{@sueyVVoX5Si2$zqMP|O9c*487{`Omi@v(
zfl?Hvc^Eg`Z}UP5(HI|OS<^*95YjScjTR+g#Y4U%1jO)nPye!iU8G|IG<XhOU4z|q
zA~7*Be^~|*tedyIue#FWtRUWOYA``?DhohooVxxf!zT4Pd=J=(3j0pbiB4Ubwfz&J
zI-P(m=lJJt@#vtIF#5;2XTqy#ITSbTD>$Uh3URLmv>_gRCYA9I_9Cfs6I{1aT3zW9
zV^kGf3oxZXttu!IaL4zDAmm>BX9{MLejxcYyZ_>(-#xJic1{5O$Y*p)<1&}WeYpC=
z+wFq2`ncQg;q)Gn$4T9`=AXCceVw+NcXjfzK3@Hc>dxuL$6?j(;J%dePmA)n*H|7x
z)(^Ll32%ek#e)u!qxW)r&#X-D+(mb8feTWt1rNW&qskR2*(;a8eoXlRl_Iq3w(o6;
zKOV@(3$sI%LV6AV_BtV){o}RVJk7J@AjRZ=%5{)ZsblcDEUf*vC2Y2-gs{;krHjUU
zdbaacZ4~!eZ+9Cytf{{c%TpM<TzZw8X3JY=agbpwfdu>tc0kPUa2$Sz*MZZdM09Yp
z06K<G1y)Q|i7YIksnkt9W#k{Lt(on-kr5b4Dt&$ZFHGUzVJlEg!G*83k_u1~20O$Q
z&}t~)Vcg@&z0BZoKcuW^%a?@EuE=EE?#9R=Q#_xMZm;PLD5rJ|1||>l@-<eBOg_o~
z1`wm@8y40DOrcdPQVF(%<X5nVXn>f89P<0Vx?HYf`)_WwzuNjVM<L~o2L#aTwr_LI
znT$1hHw^*pL(g-QKstY3j935#3XB((Odq3gdOVH>egvMjohV^Dg9mMpeS)7;BCpjm
zmEWro)yhur4~})Q+1zby1Q%oSiTz@Fhxz?hDi#k*Lc?I^a6(WAW70UFB#NS37v>(b
zI~_56+o=u2$e$F{k(ZCJPg>#_9s3)7Ld6d}$!l%1EgZjpBv$7xRx>mAS$)HC1l*3h
z&-o(ZNT*H4Z_S{=QF|1=$sLD^qwYsz?pxI>-X*Wc1LYr2a6Mcaoi8@q#}}WdHDObq
z=*M_E&JtWMJXV-t?yjo@^0zwIvw~K3r^1{l?0bz=W3lw)rOK8&q*BBK8@^EUg~Rj4
z6u(4hvdJR(3+2V>eZtgIroT|>JbU*m0uE=|;@^uu@wolc6x=L7%kh(@1i;`0E+jUd
z=X)@!`S^DVs&7Z^3%kn&+;?{KKi;lR(jp-~eBRVYYiZnTbj*x?kl~P<c-SWn^~(ts
zI|ma<N5%kMJn)E!p*}ZX9Yj^V+qmZ}&=t{nnr-k|2bOd1_Bv|dMDZcu{J_0iX9#tl
zZ#DV~db0A%8K#K0fhajB#&}3>aH5_j7>zL*PbX6DS56lgGY)%gzUS<{N=+dfM;`-t
z#?Gc-@YD<nOXpzQL6FC^&pFE`)#}b!%Z-ql8Lm(|fs!doay5{A10~4A!y}>;Lz9({
zhXM192Dh5AQ;}<qYaIs>^34aj!xPEmuWV31;h3}81lcpeIT--#aaGY(?HD335OpqD
z4!r@K<vW~1Zb+*%rH<ws9otTlfUx*406ieOj}3Q!Zx0*|gcTYZ8ob!o*N23ZRP1{?
zqLefeVIkcr4a4RsfntIZ)79=^w8g^_fJXTIKi~K3j60C<tF*lHUhn5`FkSYhPKUDR
zQ<ijTt2%fe%I)qWp{4Gct;4V2GaLmhP7zn){e}{^Y7=IYYN*0($X}6(Aq_|)w~atG
zAbN=D%LD)BF}q(0_hOTJ1#x;8Enf#m?aFPEDp1uHrC)_kzu|ADkZ*-fzVSI6GISTl
zBfPr2VG8JAHN5RV&%58MFQqJJXFJMqoQ^02x9HGpc$$8;+ah0X20Cx{_zP%NzwX%g
zs7^n};HlE%H!5jeBlh`Q0E1X$Ap$HYQMf_c)fp}U0a6)6pvqoj&<a$jvcDKFT``{&
z)s*g!QSA#lnMKv}CmcVlU2>^UD-OBlE~3<90W^A*tyO+ZT+6}uOOe_3a^||wwuYbm
zup>}a;S)|u#XPjkc<bpt1kHIYctN8k1w|q#8A>Ven`&>lV?J8mj@<4ZZwN5G){+@A
z>2Ch7S}VshP1Zk+1-(5yU?9LY;cq<g%*$8gQO@YDD5pN8v);&A?4rWmy9>#_@&5`Q
zTY+U;VV(N6LWA|ESGjiOb&*V9Q2eu)0hAuNkxnTOa?YzIi{?wiFw9px!S`%_$x?tb
zS~h95h5*6afB*hX-rv8jW#st25r#`VrUgrp!Llv$^&a26a3O|-hSH_q19d-%yKKE=
zL%ZSWd+xar0^crlCa-$Y!FE%Cq^ClE7G~Fh#a<`RXVYRNiQgP%Y?<BYP-VVKA3?&J
z^49w;eCl%$c=2n>gUgB6(jDi|JraWM2fK}@L#AHXtP+rT>&;g$uV3QpC=Lr^^PT+$
z{4{*~-TRS8S(@p%UCq)iU&?VgXwIgmI?FQs9yC{*Rm7~H4yraY-)>%k5SZf}{y-Lp
z<(Sk8e*aHl=@$~=6_1_XgT&67i$N)MJDs95e_>%>9abF=A0J+_1fYI%g<kU4kV}|7
zQ3@5>nEqq>AXy!H!MR#NpuUF(X8$Ubd>8S^vubjaj(TI@cV$?iWU?cfmpnuiI>T`<
zq*bIg@%Wq(>EEDS<fHoFSJxtyE$p<vrc6#oD|-;1ePAkVY;6&%u_f+kVZKkw#-Pg|
zx#v6K5~?neUH$vOewUz}X>&FQ-uFC~Q!*&}$gcvOu&?=aJPAWQb|sBxD$5;i%RVis
z&7X~>b?T~>(F}Z6c$A*itp13}k2xuCUE+99E>;qFpuL{z=iu}(%mT)P3YZ5zQc6P&
z*sQ^hP?fAhv4K<Dz{Z}m2Qk7ttS|ZZ$P}I}sq|3Ku0HKhaQ<A*uQLf`|3yCkZtG=w
z(NpvIFP$D^h1J(W<0*mZ*8;!y^x-7`nOK20n}wTj^bUH=>ea$Y!pZ2GWDL7|Z1-=S
z&esz>27zn<1rtWtclc253-fhPw^c~LRanLwz(()?2R`?U(x-;%Ur0vP*D1Jqy>4;`
zx6OH~l=hRY@4H2~e2Yp+#cGrL{p^}Sy1CBm<fjye&(}f3c>&;W5orDTsrL9(qOjHK
zW;*}Krw4xIIG5IdYF3wc_&`F%SGs;3F|z;XtHZ=)vDX5~tTb>*8ijlgvw$nWS8bCk
zVkrd#2=3U{GEfzJ2>Ff^zEd<x%QzxuYJm&m+M&vBmVl{K!;Xcb2bAx<0>tD%xR%AY
z!a+^iR%&>n!fdpYOB)6Lv=4tFoZNUXi$b11fFk^C{Hcz)@%-zEqXc+@e@h{RJx$ot
zIa_#ngre5g)(Wa52e1gUt@%d|xE)#OCWG5>5esTt_HlGJ#$(T-)($fCO9W;@eFi0{
zrd>SZIEwPzwo+M#)_X)hjqCwwgEy%WkP1zt3IWJP4k>u-sMzKp(n&+uAR(`j<dl?@
ztVPMmCY2n?l9obCzl0^~ldQ^anqUVT+?JLJ`JA4CrVW3<b#|2}fs9PR&1%wD-kMG;
zb2fEUwH{a{-L`XGTQyAS^QCeSr^3RZL;rPYu)i<&?{vT~$qr(=v!;xDn{AiM;HS2)
ztX}J;`NUIx6Y>Y8t~&A>cHOn*et$`T{T7qCLGTP0GPpV4eYOuvtw*T8)nh&U(bMq}
z+D7p_Xx{MRMeglDLxy!b_0OEx6*r^BrR)y(%T8cP5Lk*kJv+1CYIk+`%8|NLhhs^V
zjP5obQgBu@OM2N<PRBEf|I<^rssR_;KRsPDigx$-$ll1$Vu7aaK~4rwS*EVnD5c&?
za7;J#l1RXXR1bbD$>hO8n=hW#0B1zG5|+tKc&nP+l?Y%*G|q`Q+?~n^2esl^Y+2(y
zSt1mWu~J+te${4URqtKTz+nTArB1P<nL-I<AZ0hNAvzjfzozG$CHSD4AVBWoIAuI0
z6O6&|&^&RyM#Hzny8O{Pj4&15_V2pp>v!L^1E?Jk8-98Np}#@qhR*)+8qZm2$}+^L
zTnsTRAVpG?FdzUe13Ot^ztwF}YQNsf@82iP5x2xw3#W}so1c1`nU<i|>iF>wpU3Td
zv+E6LYBLA6qgw6l7h59-b885lPfIj8>aeToo-*6Zvr8~H-fWBHTkdxX9ibH{raqVK
zCx}f~`87w827ht)Vwn9pddFct^BXybj!L}hKl=E_cu;O-p}W#K?8ia?QpiGSuYI@C
z2mba{&F|`T69^tHpSALMu9+?0^714x(1KUmR)p)jTS9$yI+exa|2FhcC=pMx{62hz
z%;DQ1MDSY}ruwkhPc}Y#X)q8o>6hEX>7p2+KOO<W=DJ#NFcKVGP-_v_8yy;4f-jIt
zeu)=*)e3)FuN$aR{SwCq-$sPfIw6%oGWIIWx$R8U)M)VIx9JF-_&J*0m`?3<%yA^Z
zLJ3~;vl1fre0NWcdg|vC!5cJSR9rDLH4PQms3`*qgEPUgZ=BwAOO=P#vT?`p{<-MA
z;I+G+L4`U1&?7r5D^EjwNvij#y<uVM_H(5&{!^r_48E7h_%6i2nr!k3+kMIW9NWVa
zQQ!|=ee9NhLNs%B603xYCkR5>^zZq=dH?U^3dVCI7P8fnWL8t06{_Xi+Xewj&RJ10
zF<C&Fp1#`OEq_9k*=6;0g~YE|GXEdc9?y9<<JCHCMp%}R1h2a<cR-$3ZcAOY*A=AA
z-1m<PpR;`Op3j>X<BfVsZi9RC_}rMyCJsh>55b<Dz=KTR;q0dzmwEGl8@-x6p;I3N
z4Ywc$@_O~UJUd$3r2_}fEf@R_=Kx>Y_xVJ?(|k8eDtLA3d}Ow~Sk>eIHGZ}nZpX#$
zG(5Vv=KgVmlF4(hQfKUbxh{0n;5iJcwgu<bBUT^eN)V{lOm;HCb(hOG5XSp;J%9|P
z_QO&@MjG>$j&W}I`1pL*Pt#Fk*TCg?PTTjBOZrDJ^`3jjbHj3CnAfNxuTf8aL6Ok*
zvT~J9NTo)(&P{0%Gw9UNSCxdtIXq0%ABEMe;DNGVVG+j8)dM$RW8+`z25m85B+2KR
z!G4;$0>il3ghYf`n#S9mTkhc_dJj&7P;)=~Y3<Zx`0VwJmJmz}fh~jeN={TqF&=_*
z@T&?dd)ViS@37J5n!>^Swm>eibMX_E-l`(rlIF2O+C5PW)P@3owoU3U(H;T_vm6!U
zZz$UGqpN}P@^8LWR$4xY0byxfL$ubP<*wSU)3u*J0ewNh^Y%~o{7RCqM4X(CzQEyr
zc6y6T5w06Znt*p@rvKSgGI;pW!j+Eq?f#Il#8ft)^V`y-?pr+lmW##tZZ@?sM`*mj
zF3T+yo0*lx$<GG$bAInj-g9{T*U`OU+Gbmx=l2(4!CLv2qh<XxJYJ3WS<1pxKi`wv
zc>ksV0q<sKN_N(_W&O+8D&0Y;9GU1$V%)|)d49!8r1h4sfSb|_pi&tqS39KVWnyAt
zWks7S0nQUf^sFocD~%?fiSyrn%G-k2J~sE@;;qzuUwp<Fnoes8f~Q=cLI1gU(4?Ax
ze5hqQoafMXmlCUEB*ErPiZ!`R`y*-cZ?Hi06fiEJ`~hm9VdHzXm97PMO#W6VI5tI3
z{2}gk#iRmN_Md4}Y2l~eZq`vtFQ_~y&v$*lAG|+?t$aQKvDmsfs#{El_JP`5_1`_{
zE#p9cEk3IKCvkm??;SUU5mZH*GzyY%7{da=4nl-FguQO+=*X0zKdzC<Zf_>_LAR>3
z0<)F1xc-@oM9_2p8jIT_;N@|;Q1&JAV&7A}QCH$MAGmH>VLh$?ySvM7Fki&GTEFM|
z9Q*mA%6v+v)1l9!Wjg2G?{#6cjga9?fXIG%>J<vQ3uyFQrj!RsLq0p{d~4ZKC*W4q
z)BU)3bv>HO`S@_Rp=kgLJ#skx?%!CBP{aHn62N&J*?CL@*?3raTDF0OJ4CluU~Y+s
zfI&@w1!?DcLkPO+JDS5rpok@lD2w}4h0agd3W*-ecg-jV51hAqj^m*ndj)eEv)?73
zzU}H<fE>$TTlKXj<I)Gq>yz#%Wv-;kKo$t8ybO7w^c1NF;xwTpg7@}f$7KI8T=TD)
zd93y1d%AeHw{cyQW}8q$<dUdaX1Cj5)?(3NHCLJ_V*h5ink@k<(SiT%gkFiwsFnDp
zHq)rRb4MuCkk;10ofIJo7C5tyWzIUGS*4EaK4&*k44k<Cg148G?bbWflT$zP@uRU+
z`{c{X#pIQSQl|BvM+1Gmfji3SdM5t^8VQfr@4ffNSVq12T%gtCrdW=WfVZ`HUU%+f
z>SU{j&+R<X4}@)QARIX!kE`DPYLk%R%i;MtUnY%yae4XceqJFAF)oX{>EQ8Vk9#AK
zQlXpM@4e#nHTY=yiq8S${dO=K)BSt}G_Y@UPoj8{{$@gtj{YCTl-@xeo+Jh|08Z*D
z7<K^T7|^d`;N#O7i3Wp=9FA4Y^)_+f#K+4U0pX4u2jm$75CqLhg2_W=CMFlGL=QQ`
z4?d6XhYs{(I&gDWI`EkgLMUbZUG4NET)iXF-M3^XBmabH>D!CU1(y$GN6~Artt>C6
zVadOPCiJ-$C-G05o1|Ti!|apKI?$|BU~}nUG^VC>#2|0|F75rYZ9m9WVG?>)8=4}K
zb6w62a?rE&hmh~tj}9t!V22qi=5Xi&%w>j(2KA8X$BzrZ1vSf7ax`sy^uqxYLaC5K
z5`M{vg`X5c&9bmT-LK>1zADrFa--vZnDA#Vf7kVRaBPR0um8tuMmEnfKEKa9@B?N}
z=+9iPv1ea^z6V~%U5MJ})5VJGX`p#;K%nm5iN5~^P;)g~D3QzCS3bGCuz2BrYw}Sf
zGu~`35iLOQ2C%EKT3yHloO!X5PcS%r5xJwM+QtL13(uiB+@8YAovA=z8~3+u1t4bw
zWt3I(<$5_$&4xWxS_Qe{4&Q%BB$M*^E3(qzM~JfC0P;)B3xP0&)aG)hA3Sm7-&NXB
z@KmO~;)i@Ifa{ju?S3YGFJ1&@(9VhT>wete3B_%VNis$^XJ>B_9{0q*sNgdAqK*cs
zTv838;Aw&J08^M2Y_|TG>!QhX``i45!1Kn;@)HLjo|JnPab!jg41m4_xjAeH@EOZE
zX^U-V6O;Y_x)~VeOO*=SHMn`GSst9ND^1B3UJ<mo{U?`j6+WM_>6+!;4?jO#dEC~Q
zPkNzpLByArYxrD__dnh;6|&oJ_u_#{zqMvtTu##gB!wJapgjJ^fWVHzc65zicV}g_
z-OFZ|>u~&*-}`fpy!+3RwNC5h8h=P+g=PO1i8X&XQpf6gmvDn#W|=6C&{2-ZL%C&;
z_0~Y{qD)CRN$61FS(A`~y&DZWj6W>V@)+bXPXx|dS8*I~3Z46GyS2rqAfhUXSqvAD
zvlFdlz(i|sM7*~G+hEgXc&W$5hj01+f5S2r%Fb!G@U(OY1CL(@bSz~@{)O>+)BJXd
zi9AwC=blQym2xYDt0r=`FmU>El50@tQM`qa<%Pc)xDWs(wZk2yfRrSD$f{ueOu0=#
zLsKqAM&7L`D?1^U<l+s~ofU{h7Zeus_Y3Pv(ZScJVhaQ(@&44S&*m@?8;c<4k@rQ$
z;=9}K8{+f6Ja%b$kc6Cm`-nZGZLy)w<UEO<7U*!<0n0F+|2@_7@tAqTA+Xi1#_eQt
z8Ie!|ghS{1QG&cy_<y#^3E+=iXmijO*PxhH6$~$rK-(LUPd2$=M*W@peOy{u87SyK
z9{~IhFiS@rf@CKr?;ji-TwK@&P3a&-PkbMRvPY`=kOPEu`}ddaboU@@KcwtCcoPgt
zOC(N<nw@aA6lkJ)l4m5h)Tc8M^}Xq<4i(XDwzVt3GyFurG$QK-)+hu!vk?an*~HHn
zjw*++b7{vs_BG0A_B&5sST;C$tM2cfz~>CeyP)&_MpPu6bYX#0h=-lR&PqWU=Llwk
z&NenSVq#+8b~tCiAw#8*g@J_yy-J^HvDX1S52fc9#O^4R0mIPx(~}y+Cj#!fa}VJ)
zx@7`>=SP(B)9WvuPcP9jEz8rZb@s1;NtVuOsx_Op7n|TuSDUsonrFZQ+wmR4(^XZ~
z8jIIJehhggMNTKP<8c1;sasB)D}2^-TZj8}u0V(Ha^J54g%C(>QB7LU4}OwRRqYOD
zpUx%+SO=ro+1YC(9#)|v0E!3ku<LjvS%SR@J+=u+;~d9S>53JO)Tf0`yiB^kL;4<F
z<nA2sok9rgsn`%C?rQ-8cSd<I@O=M@Z$3A8fAK9Dv+Z>I(1r2SdNO;v^8bS8@sWwN
z=f~gwmj_WxM#I|)4+s(fLDT0S2}QR@wpfp97FC~WT@f#R*K?EUK;8J5?=!<qlM+?l
zggb-131nKS(WD3xSvbWs1S(q4BeAlUNo{odcC@x~0@%e5y(TSZ_C}h)AO2@I8*M~p
zdq8gY?{vm!6lQ63ePvds^O(uCl5+vJo0Swx0p8!<5>K-=?YEO}W_td-FZ~;l+E4Z?
zO;?xG8ql@C5~s~pH@>w?xBu1J2DEtPrdsn;N!^jl;2H9d);pg*U;K9;n~$A?HzfKO
zCsUyV1gS?jOV}m-I0CLId6XmZL(!0MP-8@MNJDW^(Z9eoD+kh^mk08kY4K->nlwRj
z#%7<2e3mC|oNXq3^3UudtPsmp_=FcoVA9nJi_p9s>>OrA*u{0_Kjb>peY9qcFy2KW
zl(oend<G>)?!A87e=_#Et*k2-GHI0TwCU6RhF|j(lZY5`2+)R=4J5IP?oit+*3ZPf
zjjV*)IFUH&RaBdmIO;X{&SB`kri~Oi{Y#vIKs~JpI0y|`WaOvYTdVJr7O9rkiKG&7
zKpeS$Y3^2GAW$KlnUD}TL<0J`8w;UvcXQX`WcUw$4>DbbkM@y9#vR649aThJRN>8R
zY&A2Q;1pwhaq>;GWjm+ZW$$uznQ;0|3c5DO=kqe;hd0M1pTooOc*p5k^Cge+$HP>;
zFIL)c?tCow<J&L9H_Xiezt-zOyB&u1^^`SB(6;2#vaIBnpcwMxUppIF9t3FWIyMMd
z$v4#bEFpqMeL;}G0A}sO1ql0E?JkyHUhQ?k>`<(y9Y}$Qogzj!Sh24*g`qhQDQkE%
z!K6QQ&IDefu38(S`ZD{3WT2~5(yo+Jm-?c!Ip?JkkLiH?9mpMWt2G#X$UWU(>8;~S
z_-*q6)Dg4xgRisl^74nlhZxo8@2~ZSuN|IdpoK***0pVA(cO^IG(U70MxGbk&o!?S
zDF1p|N2s!=4X_upplFt<lF&<Z|7vOBsD)x0!IJ9L*<Ogl=c!%V%=L%DwdO^~Dt8OZ
z&BNJ)vJj_=iHT68PLVeWgj|ir;a67Es==y9w~irDuAzJ9eY?A|vgdz44z&Bk*IhmQ
zaJJ)3_r`rg*qZn_krMm3Ss~stjv7MReE$d`bU~du8_J%Q<1C@w@N08=ONjTI^?i@F
zd!HpF8LHYX`mtO5qaS06YE3FWXJ#OMdVAcrvk8j9-bY#z<k)d<2zwY8==s@fu_w&1
zJ^%(_AZ9LjJU%*Fq{0Oo1d)*Fci?`IOLSCJ&_}Ppx5C-etgr>Mxe=2Kkm<4D?E`9J
zH3I%%PU<^A8NFpmji~=AY;r&)<)rq<-@tJ8(VaSD|L@SyVKz<6WQkXm?JInr(q<oV
z%OOjO7j4|S<!!+XNs8H&MMFC{F?PO(eWo#&OwxK6d=~mg=&Mw7ome)&h9Ga!O={ya
zeMNG9kaDM(1VD1~0QG=oq(X{u9>fNSL_@7KxTO%d3kwVD9qu)tJxKW6Av#AHY}N*u
zD=Qk{w>X82R8*Wqy#iEJGr@&*MAq5pAAcaNL6uLV_)epI6N6$uQ^v!VbKj4<tF4A3
zJ>5aZ6ov_Vlp7x%)+=q5oEacwxllO|be>}dEOXDR{&`#PLT-rKsEsAD&~MaqeOez<
zX70a?qlvhKUzr560}2BrWXh9+(cNHFgu{8%Hcdt$ivvb=duuC}D@^#4?OL;Lhg+4<
zIgm+=PC#In#eA&*V-};r0Cxt138nwpq?|(Bw<=pcL+S9U08kr_iPiwc06bT`w2z*#
zoFr442UJjdSyp1n&4up4Pwr8(hx01!uI98>$^l_tC73$EV3NI+KxY>mV*5fWp0B9I
zL@S2}BS&W$wrXhSDr$>Ob6<bspn_bBImA8w3JdK67@Lp-9?SXRLVMb6kmbG&n9U*7
z_EH3z)df%tnFt9#V)GN_!hnoTPV#wN8zfEE17{GB;D6fM{#X$=-edzlev}_Apgah0
z2!Eu5Y+vf!Iwg&I6+K5W08Az}>OLow8JeE^SN@gP{p57lRU#4N95&OB)oQ2fdD>M<
z|1_v1Cobd5c^6LizTZ1P>8XhRAK&}-MxuQ@@|w#?%{#o*4h0*#!U>Z6n2zQS_s?k?
z;KJ9Sfo2_ODpjS~HuE9H2FIc2B_%Zmx+_s!fxWKMo59c5vzoi3>F^Ihl*B)nI^T3W
zL`lUYB)q-6!19RUyuKkhmZH^@dAD&M$t>1pcP@m|*lgAYJ1i3Li&dY}!D9+_tQR02
zW{7|D0E$x__TJf-Je$eT0!f9rRsn#;3t~O4V>hmix#*g?sA3cLz;3&KIy=}uD7Kor
z&GS3=Gm`U(R<9>CXrf5*vthSH2^&wupRp2?DL7DEUW}ET2+09@!VJJv{h{!<2u4<U
z;Lu1Q&TpT(a!x1#=*2%pJv<~GP((2H3#RJLAX+V*h(bVYBI|$l@^IKPag~ZP9apRj
z_K^FR$-E1v)yk$d3Lo*5^P9~w!0H=pkyXu!F?emO+#vGMB_(l+WcA2v!qxH+{m_w#
z$1xuU1S6D+g$zkwH4vSI_;}<~kxjhpZ+V6?4l~8!n)rNPjX?K0Iocm(XG@r2Q|5F$
z<GN!}&BjiTiskEpxp%hRtP}vwD+!mcciz4#vYM#{<(I7aR1V`JO8)Cww+l8E#MNiG
zomCWk0PwcRVSW09`|N3R-2rB3Jr=<}dqt0}vV(W#PK2LiRx`4qH|LB_GBEwy2$z48
zKpXrIqgdppu+pShhnVBU0%g_<E1bC6Jy}_Jd+BwV^FWaZD47>^wh(jre-@#_*;1ym
zKStjY3f+>)xS~)?4GV!|@*4A^2Qc0k+Oe;=*KhSfH9-{FlJC2kyG}79n*%L_48AeG
zde!h1ZFvOsLdi{8*}VGv9?=>-6wCs?brsz6iIsJxP>dQnP13O8fdhR)DbhtfJ^p*e
zM?-!BY-|re%1JT93ou^(H>n0bqYSa8g#pn$qF_qDn6k_28Ea0BGj^bNj66RunT#mc
z;P(J_p76ewRm9)Wqn$_DwFL+z)J#FQ6-0@sm&Jn=Gs3rFrPoQ&j3u$Js)np_I4dZ@
zlFtK*2yOj$im$rjKGxYV(Ej+A;z&q=k^lLAjPT$5p1w1H-S~A@HEC1_0Vd>Jj0~&u
zU#;J}GOK|9XD6Vt&7e+*HnG>r(xiah&)(w`0Um;?G|E8gx}#wfA1H5&J91BbfPah3
zF!CKXptYrr;-HOQ7GKZ<<$jCZL`33lJnSqt$*y#$YJP1kFsZ0?$o{!VESO`Z!8cp;
zH{;tem7L*?Dd?x4gD7W|dIh9lg(^)1;&!6GRW4|2F|-hFqr;ByLT2?SM99@n%}~nP
zuWPcn#LVUh;A}A7u>d{+MA{Xxl-Y11+ZyeAV9U#XR5uEzfm;xC6Clf-;rG_IeYAq0
zM6f^AT|Rs}X5Ix}<I*ms(9pt!FnZeBXn{RUXY<y=3Af_x$8yUnGcgC0p_LmF4PSGR
z<`D26BF{&2rqZDgp!#+cOubh<9-pU~Jfz+G$<E2?Rz^Du^zTSlK@)YtHj4u)$mF*4
z1;ou|1`uTFZ#Tw!F@h>+o;zCtFsFB!4zu11)PlA3$F!b1n5T30fTs^&cfqzFvAr=m
z8vsuu@cB0N50s23Wmzyp?2UJ(z+s<nJ_}leDYxltV$C4#MqOWDEAYSkAz${}>4!w!
z#4aL>3e}6LFiJ}I@jEnkm~sdNEO;C>I=Eq1Yv1Q^;jUX6N52=qvs4;d2P3V0m4=A6
zA*4c+q_;4oO<HkQ5ld+oWvW-K)08@I_dIx^BF>|1<U>zpnfF?!S=k1AwcYBXow%(#
z1WgmYY)@9^kvv8xOqB{?F^)i}1egMOe?N#90Fk7>zklZVkm4j}CI-s(<wN%xQN!{X
z<vZT7<WE%aabP9~44XmvObi#KHK&DRiB|#R2Bo3kwFmJ4hJQZEr3tsWBXI49&Uz?S
z0aT|&Q}5q5^BcU*+cnLdRLu*Nya?K!t2GJR4k$?=c?t9ONHYNAI5;>chvaXU2Xx{B
zCmU$F{$08*?mK|sQe5|}{6T{S@%5e3vfj0$;3Y-G@jgIA6V%k>joq@TQ+N~s{-Qbm
zj&|7ti}9nm)v_T}*g9MJY$!F1|NW}#iC~{uW8>$gnw_EHO&>I2Z??^$B73(REp3Gx
zFDPx$dK}>iQH&|&WD!+CEakN*Mm-lRcq`BYnvrGm2Z;X2fLzNIWMQn6VkOlnU$lBg
z!AB=o@#K+C)QoIcjckYnEqZ*X@|qB>Tx;`-lDQ~Fy`~3vRqMZo0Zo9+*XgZ(4*YCl
z$doouvm~(bEH&^A%l?RKMi`#n5)sLn{clZLKeCXCR-RRHSInQ!fQnOXr<V%#s-TUG
z+G&GoM3*7%y*kc<%4zByQevbNkq5h{xpOu`@`3q}&$p{g^EKA26s|NZ+Zpon2kUD^
z`rux>-{y=^lxAT~t5w>Oui}(^Z~o%Kx^jx>*qw#Pmpxwj!UNZoD+qi#lJ-iF(Oe7p
zuI_m}qluA$rmU7!;ghU0tD^pl&-_G024l(8KMYZ^CV)$gbPjmR^MEG3S+WeoP3-2r
z;<eiww*wCzS{8t0c>cfy?);U=2Ecc+9LF2Cx<6Y2I-mH?wpplPT*FjiDu&xfw^kdJ
zjc#d|qj`O5HEq!Fd)s98xs2k(2El@rwL?l!q*Do~<xWIv^?E3e)PfP&ViY;D#fZMB
zmIoYOkNI97M*@T{0m6kscyR<rZZ&A3m4Wk&N8dXpr6bj4K}CwF&=wT=Tb4jJf2v<?
zcI~iE^n7np8{%)2;VA>{h!yBuWKP=A<;&J^>!d=4!*!Jd`pkX!w_H@HYjGzN9+Uyx
zHR<vC&6t~%HCW0XX_9C~5GbQ*JN|8PDH>HsHmjCBDpg$+77rFoH>Ei!l~ILPzu~^l
zwK#(hXaoGl(||T0<<zrVJW(hSTNtJ`00n5fjNDI-l;2@$30H`N`vlL#oA723k9mPk
zI^A(qC8ck8cs3;e0!imASQj`Dsu#lw8co9=Yl$1LVs{P?AH!x0@B#}Z-&|=XKNE;(
z1QD(5Wf0dgi|;b6by(ttXGSUkhOZ^CAxxxK44bBIQXoycXm%B7_R=4zY`Wj{zFJ*f
zO&=`N(71O)|4uDSfLsWkB5ljqq6VHE5PPxJ^98Kf&3<RJU2pwX<u;2o8Q_sFd$`;Q
zGFc0nE(?_?Jke{`Ky6Y_E_7Nj0kObrL;>&|v@*?OP`2A-)96QjMN}_H{NvH=^{D2-
zZR!Y{vhg{+5$U~AnWu;87gX}+2ZEsV1J&F37BC$TH#bb67OWFvjDf$IvUkQSE9nM|
zG>V(-Hn>;vBayCul?Rw{(jv^*q=Kt$W>VQcy%kKVXNrkY@qQ~s!t({;0EMJOYtU_X
zaddL>8Z6Jh+~7G0eSB|5TOOjmEAA+6OqXD^`zsps?)&xeL|nMj1#s#$csah;ebBYE
zE6D#YIy57rYCtyIUD$Jz0&z11iNI=sK=&DxT8nnAx4n1)yL{U$R?F`z;s%<*<Sg6c
z7zOF3bQBF8MZxG|XH-$c8RP%=bb|NLP3aAuYVTd9$lTK&KD<dzAKRM-v(J$ZGaFHM
zltZJuM^#~iHecFcBn3V~R6=tg1w1}Huz^~{iU)((`h1pdU>76<6l+WEGIwv+sOB<4
z%9;;<t|Xc8>i~7S)SJ!QX2DmD=f_#7V*Y)P7k2IDzQX35@~#|a8%kHB{qFA(BUW3w
zc34CxJYPHJ!Yw_FMrr!Jp*Hj@sRpe%pv;G!?EANvOhu~w8xQ5OMybMZvo)w`@7Md2
zRK=5Vh8~C^&r?YL1x5$hfM;da@q=3^3#gpA(uHcisvi+F$hkuD^=R5|Fq{_Z{j$Jg
z=lCcM!t}$5uH#D9?v}T94nSh5Kc?w|t_+j~Efsz$5qfig@SY~bK?FbzzbXT-e0guL
z;6N#@i`ccPpWo+V#ZT=-G$0|wqWzZQ1fdQtD@Bql=<ksOr#uqUg`bRINPxf<e$G_u
z4l^`Hr(cs>rke8(vhLdW+}!s(B8=!zFHq`?EF7F|uq~iM{JAJN^1nb)fZ76o_{Ai;
z(Yez?OiT>o9w-PB^Cn>d71qxIbWM*XtZyXjk;xT@#5yPy2lJ$&SE{5eRfH|w%wt5>
zt5eA|pZ=`L0sqr%_S``=x8C&f7<63yh#-LpmLi=$YXK!_{I_IQ796JyNU(~~v-qdb
zxQ(U*zjB)Dg1pw@&Twk6j5SkBSbIdp3MNJNy>;$CTXs*!LWiPV&q;#xbAK{^Fr-KF
zVia<An_QWRsVQVN;w&i8Xr)!9&QUuJLu(L7DW7GLAqw;Zj{=p8)ojZX<%(3YdDx#;
zF=imt!9wBo0VmQ${ed*l+&tsTI;z+ZJ$(Qzelm+10-gq%2Q0F{vk0glu2$o<>;7s(
z%R48NB6&s+_PsoyxI7>(sfUmb&FV7%wj<c1Ejb~;T%1cn&RMoQq>kW;p|kFQnIf0h
z(}+ZlUau1d3L9jjRT*p-L(*XfUMObal27+PfjeSn(u)4ZUA;y>@?ZbzyF*veK~8Ke
zV(l~t%=5(29j*g578(}TZ(x~IQIXMn3U4okB(@3YBVAef8$frRGl`fVBoYw)o@s41
z10z}=4UPW0NWCH%F1?AiAuYT=;}ZT6+lPt6G>6rmjnpiW$QQO#RCB!OWsA{>!Wl>!
z-78K6lP)t{q>uV!uM~u@r%a-~Q3Vhrd3K+hQ^{PJNpnJQ81f|OKN$wpkCniUfiT1l
z-#0hsA<vOt#J}tSD6uj@Bc7B5QrXqjwF41zf<QUn{Uid6YS%|y1H;$A=N7JGmtsZF
zORi1JGvS~~C?O${pU}$>ULL?Xa-2`G4A2F0b8}HqQ9xNQz-0%_))@nB*Y`l9F*b`C
zh_B*?y>s?huW$i2k6a=4S&vv@A_PCcasnRPv6MY=6A%gSf8V^?;k?`6@IoAYOSPO7
z+k9rAEH{@cTP-1azH9x(3aBSCE+g|t<n!XCcECc(+SWD^dptj&FPg)0p;()WwL2lM
zD|qF+3mRpm17YN}fH)pHf&x*iSS*%E?|Y#nC&%Y>$P>x{(hDiMxVQk9Yp6@`!ua%b
zcr0RtaeyM7iLAN96aHEg@T=1Gg4ZJb8VA&zp^7#wnM3TO){YMFgjmqYy&Irc^cUJn
z^6%d;OUNV4=b<31*K&4tm*r{|im|}ndRib)Y5r=+%!;1z6#`*^wZ#WK+(FGZ1wgBp
zHusb?J|Nd#H0=Q1e9~9s3pqu%mXe3ZUylxlV77hV1}cy@(lD6d^d?%VFaPmAhh=jD
z-flX7Ha;u5yw!n@fMbC_;iMHM@IB#tO&z3bLWvWn+N5Jz#G_w;IHZ8_%4~tQpp5{k
zRXCg1z??;<gPD~B%T1cE{$M(uifz?UQBiSsdrQ<7lT_W=$;-oodmKYg_yEmP%t8hR
zc-(ly{O7yGd0bCJ=7CZtW~A_7f%3z<nWiz-qbhnp!VH+YR&a3WEd-zqxLg<;IS3+%
z7XWw8ma3woqgj+PZP&r!R>F(Fm@BzK6F>}R2L{07MuvK2Zw(@_m5h(;R6ITeiX~!6
zpE1A<0U=e_wrsob$7oL1Mp5+`uv7uM{Toe{eoT}6$8*@4Jt!rRb@fH!wa=-(b%M0p
z?Y0iXF-^x~=K<_D5oWj}V|M=l=ACq*Yy63r+@Z6UU;pVQuL<E~8N$XvkzwJGi9tZ;
z10gZw4B8f`7if0oIrVbuH*7_IyLC><Bmk`M?iv}h;!@&UZ1)D%YVV1gC*Ss-xn;Ft
z*$;VP#<1A6{?i2xbGa(bjYLz^j7{^|*|$g_N7+_np2fiEcqTs}QMoGvf;9k_j%A&U
z!m0t+!#qVAG60IIko-@XDow`U?=dXd(#3OF{UMP;ECB6vgz_%^+fii)vb0Ddwxz(L
zUBmO<<=aZV2%Qv-yIfoYEA|8fmZ7vp6Q?<P%a$<XRDJ&gV1WcqHT#61dkt@pLINVo
zNXDi40Nm+iuM^6UW8z`&=7<n_b-AqBxry9CcV0#=>p{mT*pFaovV@4RKJ}4t5#>Bo
zBqSue9kzm~v%`XnIm~jPdvvXZ0k487d(9Ed&BsMNoPlZ^W03D>#}PAuWTqOg4$xYh
za|C5q2b$eC4VKuumGY?@dy<usV(OI=4yD@<BCgs6;uApJNQTjgx}i%&;$JhCoe-S^
zz;-1g*MB_0=9`Eajyw+kT=l-h*w0bmk<-yN3tOhGaMkmYIA?{hkTHH|GRNbvCuYKJ
zwyFPaOUB_2%4VDgK5!hZivn5q)Dq2vu1YFBLalbLh}|v;{3m?xCHr$Oz^(4M6crIV
z-G3y2RNY2O`|oc&{>9-X=w)^lX2vpFrf{-2U}`WZ>U+z~VO?`b&!f<7a5U<N0*6Ad
znB!^LV_kT@+w8Uq9;#W^Ux!HJ0u;T7MXgoCDcVhdPK5cQr%!|T0ewzoR|e-UDJp`5
zgzReU#*{M-GB6Lg)kiUD2=GUnPVPNv8NmE3quNBczX#+}E*I^Ydu_nW0mGbHi)s#@
zT!sCK)UP2^U&6*RBb5raa&bWqNjjmR-+#(8utAHN#SHjCd%lc>hDv}#4j7Qb*{9X`
zYSW_gkT+`5DrQnqX~wkT<i>Q{MH25)0G?oa<%(3m>r%6O4e*Q@-sMn5u`7DguqREi
zf9KYYphANa6CFqU{yoTwAY;Ftm>jUW<No12MSrA7RkJm;#nrQGu@{<cFrP+30onGN
zKZ$?Bvluj3%u=WUD3Q{l#otqoj*fD3tG{Q_(a~>iZWyX=#ZtN~jOg9KnNGIX$xrUy
zDZh?6ZW4jve4Eo|%HD5}flB5qsCWxv3Roio3Hkx!WPLrJNQP^c2dT8PqrG2FuuXSm
zK8{!-7B|bGfXOwUX!*^UQ_A`IL%>n#PhT{eqSbVCBN5HTixT?;k%SU2`JXZObL^df
zY2NUQIaLA>!SD%BB8E*g^*5@<N5khJb3k^$-x!hQ4&9Ihv5?G4Q{O9`vIn$4sJC?;
z=G5e50GaOX*!+UOyu7?{rQDuOxQISv-!9I~#MLWrv8QhtgM;^oc(_dVFkxP8b;6Z|
zC|-d8IJ!Xo?FhLUxHhvUEkqp=(&*U10+rrUX=im0^-FWg2?NTBTU>X3m&f2*F62I*
zitxqghMz{LF$fzw<zo<g2n-UA`p`=H@C8xZKhG;AogXHgp2iBpV$l$uW0IJ2zwv|u
z8rnA&SV>zUMFV)PGUh&)wHp$=L0I-m18BpnXl9X&9<8i~IZSQ$TW`%iZill?WBHdj
zo-55def8Fy+#71@(nM755OH$nl?f<{4DkW(_;e)6?L<JyV_{~Nr6iITNs%_?Er8o4
z(bisbYlGfy1i*-Giz);9rp9Ww#pQb2-Pu*i2BqW`*6VW5rzyR)QhIW5aBy8PILN`_
zVFshGB)}ms*t$U3!&umzi(s=d>ta{ndy@Ek8efd;Rd&w<{LE-Hd%p}@qrB!VKDiA<
zg*~ExFD!>^J|9CAo(|*?un%p^5YuD`xq3sYs|sAz9!?JAq-G@KCY+R}^bVTIO?cKW
zkJL_yWGemTpj+3cB2NtmA$|7gH}mUkZE*$N{eijB&gy+b?~P);?$nI!`#b>Q*S$A2
z3b#}qW&>CN^^(<POl!I5#0OZG;dMQnCudL^$tWnglc#Eh3=Hz}Pq4wrBr6`<p+BGe
zzcBsx*Xr^-4vJc@PZL_isumBv+dH!7N|qr)9KmibDYpGck(QB`4m&mNe>yli%G6jV
z2DDEd;*8~2d!*#aKT49UwqF6j*6AVtoGL!ZQ+XACIZiWa>EThp=+Adby<M{9cJBi?
zW`1-tdQ-}UfE`m&1s7mHU{txXbd!KTl)fAR)6}0fYG}ax7XZ`vtazzw5t{SG;)=r5
zoRf~)EuFR%fX&wZU*&psBo33;>6y%F1T5cVTN(`zF`$x{PKMRw?s7f248|UMEjiE!
zwOLc3eZ+Iau8C|eujbZMM7eVZEi}esUz;siGY&xzmkCf&=UcV2g{W)|gF`|fjJfv~
z|6Q30&9aa0YG`O|gK9u~bUSeOe4;+NA<7e@z?32;kD7o{gdt#qz2m<M$G#|QYNn^B
zM^famQ=Nu_fq@waY%|631Q=SFF<HpwaN2g4+x#3ke*>iHdLi6k>r0S?B1miSk}U{P
zE1IWqxaI$(*6@pax;5~})NIa7Jk`r`9Km}+xYE#R(H<lB8oX^+!g3C!9=MbSy2Vs{
zFp7-aN2Z!BiR#n96^r9u=4S&bWp58mu$t9K7Jw$ZKfc*4l=q0|MUqIs@Bz=+sCc#L
zO}6GY?oQkH!)JX^-%dWe_PxGeGbrVwePUsal;uvDS7m9`q358BGR#6UG&KALW}&F&
zYyqXFstS^AD4;`+Iq86OVBWFc4M;@tM~^gA|0)r`{pgG+@$6WLZCo0w^@zUHzW>{l
z3K;)KO_;TIu|4utY5H)2+pG~RR1Gr|6Tw)aKfyF-o-G{K1!+TtiIYZW=Dz!LWFbJs
zo(t{Ws2qkZ@!!&|Rh6w%GQ6^T3{4Bo`@>~<flt@~*#!K?J(#97*SYtFK$5P}HHGGm
zm)@z5*eJWQVtS@I29_akb5<~;!oXV$$u=@)Z~p{f-9SXTC<2x=BCZr19#l+ofd|0G
z11`VoOEMYACpiQv^4GK((qX2(60dmfG6HpE(15E}=P}~7#RG35nqP-Cpv!oy4`zT5
z4RE{Ft>pPASX_tzM69SKLW;QgEH&Qame%wRvnMZs4)gyg`=;<nx@hg#wmFkzVkZ;Z
zw%xJKiOq>Iv29K4iS116iS3;F&U3E*yT5O`x~saM?p;-TueDw*J#IaYFK`6$Fuy{R
z;Q>-GyKS$fs7Zesz&K5iNmvu3K;Qhkv9U(?crDr8g%H(p#v0{Q^DuURD)0n0EygyA
z=t3e05&=-&;f=t@qM|Qq{5EBgq0@j#YA+>DG&G*byjZZOH+cCDu~xAB+r1_~XBGh2
zmtkROPDT<lF?}dx_GlzNa<u8&pZYqMle1kyFK-~%JZmd=uKLe`l$V?@XQ?4e^nL;~
z?ib5@!!MJc!;L0@sDdvqbzArBFr&J*WU61oE0roKc--ozDg2MGRZsZ)+qBGp$FqlB
z4lz?Ak)H4s1BCm0|46T*-XbI*d<NSE^21*4;GqYCA8^NK0KF`+M?%-{OuWpI(7=E)
zF5ohC|Hs!U19+x`1RjY6t>2v4a;7X+W{`#TKY<ZJ!_nu)C<+r|S^?`uML}Vtp%EgF
zCC3MF-OkR=nGkAA&8v36v;7=zeNWMTM8VxxxdJ5sGLJ7O@-o?mvkb5v>?Ihu8TqsH
z{Q8QEi%Z$YzU`a@aSPY;QmK$NRK)h39fTz4#mQT)q3R~3>V{?j0ADdH(q|#s{ouKf
z<$VP~+8W8eu(=%SuyrLz(rJqzUhw9P{q_oDULd;=U!}`FuwsmIH9x}w@~95YnrPZW
z)Jfphgh?(1%5jcK!E{#2dv_Y%FyD-5w5&Amg(S|@yceM88iofv(173*NVgmh0A>_0
z|M&uYXZ&0KHaDj!8R<>yPx!PZ$~aot7U2*f_{y+%0G^#!yI9sB#RmC-3kN+Ovz>)x
zxPI~>==Ph<d?&)`O_QT4Tnj#gI)NVr8L^l7e|}scpN`l|ow}d#M|gnS;Wj&s<B|k4
zBckYdtWtK4zptw(-cy-bSw-RmGsBTl;6(53?yAwISTO?Win<LOo*BV>;Gl>411^%*
zA;%kYf+nXWq+=?eB2a>hIegFs(}RJKxm}KcVT9z8hbOKf_)Sb-J-xzqRUjD)ndq%?
zp$SkI($2w4&rZLiX-y0tl8oku#k-;nuQ~Ti1rA6FA51l<(M7@NvE>1Cb3BI10sSTH
zpNA`$=<MB^(Xh#YzcJ~f+nFplik)Y?5A^Al5wN`=HdAUuwls=LMkzyxT!4u84UoZO
zGSF$!{n08Qc9PTeBNBmjpCek?jKtlQ*a1`TyX|I|M~XoKVl{a9z`MZXp<d9FD({lj
zXZ&jX$Ylbw__~Ko-^rUshxuL}%Ji6K+p})HQ$8`s!NCD6+<4B)!^1-~v>KifP#dox
zakjLc&yPpWhwoehT!Hz0wEJ;hl>v#H7r!bZ9oRY;QU)MLoYaLp#BxM(Pf6lS!{?Gu
zf5sO)#{^=~pb5ouXcjz0(^4=Mz@OwCx*ou56`UDrw?yhDfC*g9>HH!EmI#RDpt+DA
zGO(BdeLf&U%|zec(U(G?cUeAkQN_J_@q@YxkNuWU*~>jZz9QX-KJu62dw<|~UCUuZ
zHX3bjnxhO5!ws!gE`U0=!+Sx~QSVpO&Pt)!lOy6ORse>2GS*t>6tEu@BC`B3cNds^
zcXh>FlN~TEMo~|k*3~<H-on(hF3{&Rz4+6ycgOMSI-bFQ<Kz9(KREo9S#y7PH;9OR
z7fvVyxY`z3GJ{z=2h70;_D0!x9xwh5txcw4n~flknEx|@Zl`&<ePv3;^+R-pk(@Yh
zMH9X4o$mVu(a`r>#<@oWrCc@TV*Bwk1dv^m@C$>ba{93dn!<lAniJ31sUVR8d>&v{
z2Y3xHlS@@8B|3Z#Ao2qHjciY5lkWJN8aMbFcaARBAj~(ggkMY)=4`>7NHbA6_h~@_
zWF(02iPq$_38-tm^4YN6TAY^ryiP3^&jQ?Zp5ST5kZ^b_@N7G1j_`4bVz9<?OddP_
zf=@m8xITBsJ3w#eZ(+X;vZQFF6s4s1^5?%`loKy^0>k3b$MK1}`LA}eu$J>!bPfRV
z{%E}IaAf0FuQwP`1$B(CuHwg#Sd)aVhyuj)ObjykUMQhUpv>sv^Uq4Y?nvAOGWDx;
z+%6Ou@ru?qQv$j&hEiGiN-0qTHg_3OgMMObShi9jNlOyGATk`|m_eU>HmVqmiq8`k
z#=9(E({mjctTG?l4;~;a#IrDBsUz%U$itV4Mb5clB!fu~8aE%32Lz(v1OF<<LbAd9
z0eg`smu{$=1`KPpfm3zSR-g`^>N_Hx8JypKS_e?zfgBHrs;RA5(_L-baHto6=u-6d
z+<euR0ce@Varz~UA{8bGDx@hWIphvXqNBaN@4=332D`K|<Ei{%%>LIDF~`fxwEssi
zYO)P0B~a4uiUTSW1oOc11ZLQakU&tq3Prk2EPx3Yk~+-?-<QLnZzHa}e%lY8tuXgR
z*FA!PHyxb~-EAI10uk01WdlkAsMmv09EcMJb_!VlCMJ$E4gGF%{MIMRrh)w$Sk;Xb
zlLI~g0?hvSkg$qazz;U|@ilpdwwFssa!yoN2<W|yG<WHX&0$;g<&h3xT#z;Ywk`eb
zn^xj1$OR%Hg7f9)HOBc~Ml>}kj1|6A-c<5>@4lnc*5q1#2T0ZO_91LyaQ?e3NCW<&
zplE^GwC_*DuQ2?3*Ymsqp7VD5Mj$jmz@&kT(%^H6Wh)TXIc$k~6gV>%Z2`De<*fAt
zro!xXDcibGT2N0s<F5ey1sL_f;(H)olY)+!D;So((UiT>kiA_73=MX~QHUf^=pj`A
zmR2C9<oVQ9Ad#N*74aDmbyOl=_L@9N#2-!<L2m)<X=W2@Q~hxaiy)g+>>BJMCYaiD
zvIuBu1tfYmRaDp1;<i8YMy(GUGmKrtn?L2A0SU`B?Av7iD$$f!x!SJAzsc_$o<VDm
z>`LMF9^YMN8<)E6LWw2#w=S*3lh<4*UzD1SH1*Z-fegV&$L@Rnw(lr;N>jvafVmOZ
zS63)cYsYROEx=KxtQMEwnZC^yShBdLZQq~D!Rzq3L0s~j-`j%F`DfMn(Poe|fGi8c
zmrKjMZl|s|i5>p|X=^|jd>4=$L2a1;AT=*HW>hJ|!1CPB`x+R9&pyINoz^PL5#uv?
z1bRLo8k{~@8Ur;tTRXB|Zd1nOG`l;{oRys}6iYft{wAh|qB$uc*`|pC>K5p4GJ0wm
zTF~6E|1`rUbI$dJ5QrqMzMYER!^FoRmja<2%9?bdrL@Z0HD%vh5G*W@YEUJSIDFi$
zSlw`P0<93It?4*?qcN<hX;ss+LgE+Cz6ZvJ9}k@oB<@&zxF2KD3fYblJ?7frjKdR0
zH=*>yes}*c^CjZ}i#>b5!NXtDg!$LFT@}fLD;=t~2VoF{xc$f#xIu>xs<Se36D~x<
zYrD;DugSX8KwzVg?t4r&Brs*!rh~sMvdj?=T`e=)g!O#YHl2<K$63Fju=7$@xMA0n
zuW964Os!~fG{1MwTV-SC_g>kC-z@s=dJ+U<zYYB)h(UJAa2guAoZ3l%{@*RvYN8<J
z8A9&QyHEe(sO;!Q-2E@j>-7EeV|{fLTvt7#k7AA+Ds>~fvbpq6)i+w3%+^MF{4zkA
zUB4%>;yiE$UC;T8nfFK2CUH17+}<xx`9n~;0G2#y(BRLSn{8`&di=;?b7pNMo@$GD
zdryQ`F_~fv8S)LY=7e!FsruH7i4OPks8*epPL2KQtX{xQsbattQA`^wZJsl_X4iU2
z?D4qIr2a0+>-LEcB_8PQ^xc+<Q!ZV?!qE$YQ?vIlGDD9%Whw+Ww$I#IoIWRg2(v$T
zgAY!PHgwXYzjnLTmjw5L5(|z@_c}AdpEg4IT0opC3ek~<C;P&#2gj`VTF;&}jU2T3
zFx86|hR^)&4r1{h6LJ{;Jwa@MHoXo^$r`K7^lZYV^E<>0C|}dYx1n=q5Z^JYQ=u&n
z-}R4M&w06)bq!0eKVG>@l$`=)&f>VRX~PFt<lF;~SP6!Z^!`aLgG@~!Q5!NOO~NFr
z2IPiTujvUxMTF_t2;gLkRd$H0*H?aR*Io?_S4wK@>h6`V(FT#hYF9r3=O>$X0NKr^
zy4kWeqr*zH_^~5tJnSXG!x+cgGhs(2)}_6Rm4KL_n9qA~bJ)GWL#E67o+4=!SUx^Q
zge<ntWKZM@XM3UZa=}XQk-`cKDGf^ho;tg@1WOX){S6#<6m19pI^3cCc(W5u`k`C<
zz|gxsT-EpfX75bS%8J4IBIu0|;b7Wq!b2G1_-5EO(TK`da|n3HTUl9ISRC!|i&6ZX
zxtXW<DP^uAWYILiZhLldVPA@CQ2Ji72xx;~?EjrJ497;Jv{Cq`LIw*wZ9@*!b+l$L
z9KV5!I&O7i|E*g$*meTT1nvh|n_vR}t=EnjuFu;7(&inPB;5jVrR{d_hqK61yAhgO
zJ3G5`faQJ<F%^`OTBvbU0<`4<D*T>2D=WjpFfHL*A}3Mv?F>K*rqC#-JGE}?DS{om
zlrKtQ|K--tmO59C`H_c8c16!zE_gwVVrHE$O>T?eFA}>ir4V%wC-bmm|F*ZmpdpxX
zx9QN}l(v{n$B`~wd5H6g>`+PN=;s0}CrNbK!wAMiu}pg+Wg1h{rjcovX5PrNc|C@s
z)0XeF|7Z}ZH0w9Cjp|9L^9Y%3H1G5;o<cA|mWkU)lab1Q!4RjIjs947=FB?Z-w%0f
z2sY@w15oZgX^WiBFeGR)3PjWN@A*zPX-?Ohzj$B$ClOg7s#U=x0|x?`U?N-r!3A(q
zrX#p$rHrL4q#ToH!}RPK8<Fk{<S!-%QD+^H8PFXi`*vo)i1>E<Lo1P?%9>kRB=(^t
zDADFE*-7G$PEV~TKu<>YjYNCi(Ap=^&vD+79_MS#qoZv|2ZFXRhh+zqr2t%ce>@AB
zqakz(_@2Po+SQ;VsQ2`8)_t{&?=aG9n~=XUf++kM4+EZ@4<@N6alZ-lutuxjgAYXn
zdO5W@HrMw`Xai&3gBHXJwe#f5+UB?e;BuNBm4L`;JNf$A?0;YX@3H^w_<uG|6GVc^
zX5Jt_fhyP))X5c~4i&3JCZNi;Oq&NFD%f8E#_k^MgBY9FUV2VD*4WzG+JNlt#Q$Ez
z#V=r+&{v|nRV3`^o~KGDFKc+Y2lA~FhX5Ci#Aax5sN;@zVbE2_sFm@I%Oh~b?@EUt
z;?}^h>Wkm<7>fcV>pp@!s}dfM3nbECzkXdoap7cQK;hHS6kHqJ0o@Azo2llAzP@G&
zSSB_0;R%`$p}<ZP0RNc7ak5smj=dZ>m%*x*e8olA^plrQAqhhJFl@Ahh}C=%TfeH~
zq%h}oy4#y0XB%g3uCz+$^QC*QhIE~Fye*(>^4lvo_f-Dn@?58`*bN?g;9xSzQ>=hN
zP;aO-NA;4m?ze~m-aC^1CzR&g$2|sc4qJnv|A;kB8^I3yXC$FT9{qp+eiS?%c?v@i
zWz%%2#O|C*ZTt8}Di5VlO%j>WcFyZ@5koaLd<zGRHdPb~4VQrhc8q4>C@=J}ziYkc
z!2{uLA$)`fFp#l_+(ZzywZox^X;vq$zhN{U-6XC!!p#Q|JSNqbx*{oN`4)i+_3$?%
zyw0%UTKv}8z1HJ@h)OGFLk;AUv%Ts~lUaCuYusSNxE5#30K_PQ*xs$<uh8~=_9o9V
zo7=@xYv#GvN7r4kx}xHAvx}FA7@7KVqrEXSG&GVu+$Gp7+#~sh!SD;j4@60psxgF1
zfVTVJJe>M&H_kCX8&E~nZo1)&5V^IVmzajK7DNishr9|-YLKs*aozP8e~O{}Y0#uV
zN|5(r4ZprJC~VDsF^IxN$TLIORkobA6Bwk6GKGKPJHK8iIbCPOUUVvp8Ylr<nDxB4
zMC8zLUNQ>V+)15tt?jlA`nx@+G)2&GMv4xVgZ{)Y=Gvxcpz|%<{<TGN;9r^3mI(KA
z8V-u%5urpKr-5ddzMiPd5v14AfAp3(#A)=iEwLjlQ`gK*jG=WZy_~)$_Q!RuC@c1#
zQc0_SVX%v09cZ~-U4|Tz{0a!t3XUh_slVyn%T6P=IqYg%#@&=vaXPs^*~!Lz%-Q%h
zLrF$IaC&JtVnv8=ZP=ObvIt8$TwzKX@8u{=x<y!{IX5j$Zg{?G+DlsvHF{$z>X{bc
z?<NfIl@mJHuH>K0LX|i$FQ^U;2G8$Pnrf6;W1D5s;2W$_V%5C3W#St4v3B11ZVqaN
zQ2~Mr(#^CI-=s$*s0W4<^p~)lU8-suLZW3f2oCPwC$EjtpEuTu1^H_mP-ya(_#ZE0
zl@<kA`3ID3gj@TysZ0qiC5?nrr}FGLo&Au|r7v@{cUTC;!n7IPJlwIV9w3Kb{k0kk
zO=fCpveuBBdiiVHGJ!qAgzQf?d%1)`sc8~S$1pYxjmNSM@i2;&C%pLQS9ZTD(dJ5}
zVgDxf5gZ>OlSDm;G?L?P&KM2&69l#Ft7UZt>Uw5WZ+6f0V|0(oEG$JKf1G>^%(X+h
z(}a<63E(7d!rOwZRTW%y?&n<gdn7%OMTD(t*0kw%Qd>&A^KZ^Ut<y%-bs*gORaW*a
zxroScb^+&O-xy_YlxoGp)piAwHq*wjG1IhgQ+mijx|jyTH;Mn*DiV8{G0Zp&#j%!G
zYnyarN)`9vCK5z-DYwTY?|NqS_cRkHb+LAD!C*(gb~Mo7T$7=v>2qtdd4KrlfD+^_
zI29rj1k|SIPjPpH7l>}LtxsC=8zYfI$$FZUq8iRW;xqGm6{$2I{!wCVC2y{sOcQ5}
z%Uoi+`@(@GP4!-^Syl3o=q8~OsorIix_F4rHIXcy{(YGVjFc7@%u8Ihv>z1-+kCnR
zcBSXudBhD9Up}6$3!@X6v}D55%~lI8<{9Q;<}7GfcSc4K7^Y@ur_{Q+3e76#xpWF5
zqvW5jUS3YM{tEGSe3m%&66vcU2ihy|_fAi-ipYUKX@dJeV0HFM0bZhm7VceG!=1!o
z`r?gHvj23w`8|Hoi@o6gcKc9C@@7aG<9Oh<Kl_8EQr*()gkyAcsy87Nm{?4Y<ZoEe
zwMHDY^63llFD*VkK8ZlJaX607t|5IBYv9vE_fODlagg{Rfoj{!J>WeO*N~dFEu=gN
zjls%>6;2LajPRb<=l?@vb$-FsBeH5Z(5n&S8_~cKQG$c8e7w{1VMs=8=V2&fOj+B8
zDc?i^yMd_qEhN$+;^pO~Z^X}{(TtGobN9WGafcPB8l}E4`8Of$wxKSt*-rEfsxl*s
zz6GUXuhH)3uC<1?_))AnkTl67w5Fs+c|YUUIdl7J#WgQhMO9wwWI}m@9>rkH(Z8|E
z<R;bL6S_R(dvciGe5&8L8q{;X(zJ4$^j|WX+%w(vehm1#OltU{R|q7%s7;!WB9y8p
zj4rau?JKkyjj@L1M-%vAkGdYHHKEiAmY(mzOXe3#vA}5JU&l9yly|o)d>=91=w8dl
zpCEE)gi88Jlu(s_N@`58L-CcW#m;N)wDf(9@$WeEfo70Rn@85S3~>|j*5%qQ^K-qL
zjL5<7dB|%Sn50CJ7=HX`{kAYY#}H(3f)|!c5u^n?bV~!F>W5Yhju+>UH^TJ2EdH_R
zE0vPzy=t^C(ydEdgLI)<@DDmEH;>^j1<tXnF2<{WE^e`#T)zCm$S*KP)J~CwI_i1=
z8v$$O&}x?Hd5cPmx{+6g(NlM}O{Iy3^Wr35;@GT#VWY-2E)ZXRFv1hNyH^M{uBq>P
z{fjTPUm%Y;K=hFrXPjF3^S5gp{sVzIy_C8YN(8!>ui$H;SYJNTP^1<Ly6Y&;#gvjc
zioH0L`9z{*TzBYgVpPZw1)$97oSdypha*w?hCkN8$Xt>xeNe((RQ^82D@uoc<9Xc2
zWU6bRiO3~UfVvjgy(Mae4fY5FUSe~^vi|foo2I<Ma_%cRPoS=%FvNO9ALP>+<Zcl1
z$p%HKQ<Y+<G>Fo;qCl=;Wg+)%fyp>Lc~B`uOoNaeHrGH*UEg@=z<kR~xd;?`>AN~Q
zK$WGbNvIe;&`-_(bJ3WqM~Lpi=`n2MyHD|&SG>EH@g9rL_S%L$SI+(wO?M;x=p#hi
zuxHkeajFJIl(XN~^XOPx%GRH?F@`*vwbgFz?9jjB=Cf7vyf^=*&JMzFv$>=QapFPe
z$hD<hmlK!%SQF!1Yt=0@WCI}OzRp`$uC3W`P7leF-j}TIG6<xh&J<bJ8AQMyeY_}-
zouj+&_C2?M+MHDo@i>pN-*Z5i72EEsGT#7N|3{bEbOlB??Fj26k{8hsD!{1!1#;tK
z^ikNjIaLZUbefFRv1qH5X86ga$87D?;BL~UICusI8g=p16=wPqPr@nowC0OsCC8V^
z?e4l^YIQW18HE+~%CSD76itBl%#J!zeDm{O$A+$nZJFviWLU~&Y~^Hxy0pGh_s3Qy
zV?V!*&pw0c*|;I)I7}9sD7OF_q~GyjlSIjBL(Q;lEh9C9nw^LBSxU=B-Jv(u`2?HG
zE)1<{;fuL$cn3ZR)PPRe$$?toA;FSlXtVkALyXS3o6Q7fSt7xxf<zFZuPn)7EY(<f
z>BsqQm{1gDw68^9r(6XGWBI>0W}x<VzK~`bUO4+Mka=rOw`=@8$=1*CLy4Ek8kuc<
zN4LT6H<uIfKu6~oEU8eeb6y__8d1h3LJ{Zz{c!|e6FZ~bOhNPM6%&QJYzy%`0;~+|
zJJBC=lz=Fx5xHBK=~Hakbz3!6nm}72JcUhKikY!OF3flA;dFv}jFi@G=q@cU>T((#
zcGQRzB85*jX=o4X{FLd|CAq?4voh^<UmqWvQHu1970>Smaw~&;4Jy7TQcX__hLe_Q
z+#~TVw5{Wyy6dV!Ssho;;cY8<X#bbay}5lXRg6^Qq*Q6WXZyE*c!woF&lf8Y0$rI+
zM{^AKJ|5c=;88|#ZTQN-<wv(TaJVsEx~ZPhOzsSzjE28-{P_JZ++6sVT393}$uHki
zQt7Th1(>tZD42=FHH`yfGuPp{Nr|b-7^7DxLv;e!ukOobpVXBHb+YTjN6{^D>lrrZ
zApvq73VXDAg(jJ`hh)6PHSN$f`bRgQ-OWBii5!V_18<P6o>ZnoTU_v2uj7$8UCR=-
z2Ey1nGIFqq-aUf<>MOq0p)M=6)n(jpqcbTb^`Gr(ZfQLMxmAaKoyEkdJ<T;}{&(;o
zJ{QL$Nn#Slm5sIxT*sDez2;7N+hNe_{fMUmORu*yWB+&^r*7+t<kDXJA2`;tLOgMX
z8^Rjx_j_RNp$Lh2a7FCy?jD9d!xEgzoySfF^Whs~lyT?YO;~xulF(E$<}kdvhSvU8
z#<Ca8p&&zEgvU9UHzBUk7?}>{31>m4!e8%VoeKM&C(5jIb-vD(fI03f^fu=uubx$l
zF!x2pbWxUF+PjQ<VkGxtOEsoxf!K+$V@6zKT}^0h+%?M0Jei_J^CruqBRx&s&WPvA
z^IxY~qkbqqt$F3?d_(6Fx(C`53X`p@t<}R!4CbjfiauUvDMpdg#;y}AHQB;9%j-z+
z@?pjdRjR_^&8X3IY;xl6nNB?i^k|kLk~!-WqlQN9uQPrSm1gMU^^qMoI`<&)eM=H7
zGehUh&kyitdP`Eq0nBgVn)x&$$v&6R$tKl1B)aqxALmil-CbRylz3o>A1R;7=iK(M
zT?_0`L0~mTNgS|<!Jzmc3omOqW(3Xv9J133&8Xfg+R@92;$jhoo5atL0N>;Dwm=zL
zT*}MH>7^V*5mYh^W~`}74^?1E?++VO>JpKAsyCzQ3!9lPz>7XZ>?wXq^MC)r$fq*b
zNVJ-90Dm5cuO5M{MdlgFXqBx`o&5av7Bk<sscHB${*<;AT1eeQPHHwjFlhqWAky8Q
zX{#vp|Nm%%dxIfo@ApRl*_*kzI)jYu{@Zdeu}0wjPR303-v%Ea8IziqBZ!PiOVPv{
zWa>)Br0Qnk`oHf;+ZkH`J2b7#TrJ5sIhe_qq(D{{mab&1?BB_lM66t0lt9j+4t9<X
z_8@y#GHx;^Q3qQGXH`dIQ{ZD_Aa^TMkfgIQz@(9|vULSH1JAa`t{^dxse>7CfE>u)
z0{CJURxWN9G66ul@ZSUB|9+?E+PS`m5SAw9R-uP*lnRz@L_k!5%GqBs0$%sR`p1>n
zTm}N!v#%xtsb#>cf{3r2g9x<Bbw<MU5qHk7f4jRHeB8GFYP3C5e!p6IKc8^e%Hnd^
zL<+(iqLE;UQjRgfDg-t`dbYfZwV;I{Wp12Tcfy<k%ZK({YdDw&siG$)rfYJFxzAV;
z;SEjdLX~Itdfwbsi%azm?fpD_I?i)G&o`okz3`8Xm)G6Vci!eQ&B>uCmk{%VN|@FQ
zPV0}CY9Ewr7_ey)fjZOq%=SVbS2)zFeM=LwES!@Uq^~kDyrjZBT*tholD4NuzG_KW
z=EZ_MjaE9Jj!X&@cBg;6p9(}<w?Nod5HwS-H!s6GTf^?<dxMz88XewA6?%^rn}f4o
zWe8+12AME)UQy%X<S`4?4(7Apd&FMS8zW|lM`OC+4ZlIsd*gS2^di_EGpj}Q%PmDq
z`dp<{x~xXMExPNJ_SY|`i}<h%STBd~%0a^rA{}-3M_b3>O26ElgtMYHtXtg$Ky5?F
zc*6~KO6Pv03kT51!KH>%68C8v>(Oi9nWJXv=PyT1Akj)liB)yY%nHN`*#ADYARqBn
zvAkP@Ti5Yz%?_lH2jDPp3UCn_Z8k&4(vS0D(iK^E-cE1K&6Ldltfi4pY-!CRe243z
z-ol;-TunZ99mg9c2e+D^w84pPX54oDk~O>tF{Llbw#Z&6tHES3%oo26$l0ow9W0ej
zw(aJ>p5fBG-yAOst)YcojF*xcZMKu@jnQD9V>E60Fmv(I3-Tj_B%D7k%oSn#U;MF|
zL$2;Cwpm99+j-2{Q5km!f^!OxWaikJMA5ygM?q*tACWaUpqR*-ozHn}vi6zU5{7ZH
zDt{D^)@C)crOjX4Cz}%VO(GLb^Ymkq_AeI<DWQh;H-S#VXLcGZ6VG9+OomG*)8e1=
z^k8D7$jx++a_rZnKFq`t4AAi76f3LL$WfxJKc|E5VwzB=jo^3mFX#HibYXp{aV?e+
zDqLg>zjEDOI~!b(7G>h>Y+vl|o^5XnX?FK*E7IuB$@Vlape%t2Fli#B9~0px{mQ&K
zm0w%Gb9t9uN93T|q?yy@y7X`*`yP9r_VKtgnbowAjH^k#oNqiZ=8Zj;lYO80b~@Y`
z?pVHUVe!hCWr*?lu~@<UP4?Fd(FCg!e?0E^OdFDbB9b_cUXH=KP_`Q1P)^3I`U~0F
zQx1Q?bDxMs!JhgeQc>GZVr27L0TvItnqz(%u?Yr2_3s}O#MTKi)?ezR1}Xe7e!(0y
zL|hrF=Ce5!zVr@NOd1~RF26fVpJV0Xn*~TfBx<v>5!6Z!+WWF(!BNBUN3jXTI8(RY
zPxl~r4%5y_U%^o)*TFN7OK!{EeMUHcBfMLbly{FzKSzvpq^}eci$~yVu=5X9r6~C7
zcc&VDB<k*z6CknD8lfMe3}Ip7pKR(?v^>kRjz~X11Q%T=j?)2V$cy+Zvb4>-pe<Bj
zZKa<TfPLL-ZW2Kuvk=n(Wp|-AR0VYRO2afQ!uH|F?T!5AxUE>+m?O>#V(idA;wy}C
z)*(IUq(Wp@mUbZ#1KF-gFEVKCQ7aKT)@@CFAu^7aS`zF+T+oc!Sjf_Rs0w(E)vxLP
zp&bo?yULgl8W}JIvN`dfMnseV4X+3qK3=dEf<>PQ|M7GhQ8-c!Uj;^nI)<lCFgC<7
z<4jHUolUcgq644!PMZD%ZJ$-a^a?j349@<X<rYBN6*4ub?Q68FTV|Tr60ZgOG>SQ<
zg8sFNaw8TlJ96*2cMz*g^%vzTtS8=-&$iL$olhT|6Ei8m8*kh?x=Qy8(9c7{9{piI
z&&J!wR*y@X?ny{B-~#sgWFSn8%*H<=(7zb5oq^W}OJxw0@lw+m?|__~EfS>-tF<8Y
zB!`518xT1{)-<y+qekZ>?ewr*YuJO0t?GA|Xs+bGW_gvP>J$jIy}V&EhM4OqyTdli
zl?y8uGRnA8bt%tE`q^cEvu$5Zxb#qay0@n~B*@%mC3g9ftx!?$&x}FcA<9Br%{Z@L
zVZ{m8BNN_wt?B#=_z#Q#fkntucI!pZ_LHHzq0Do|dv!(#TyErt@5=c)cztY%X3gTd
z>b#DpO_iX9cD?;kD0QwCk*$SjMtQCPSTyN5p4OJU>XY97_Ppm+i46fA>G4m6DPF47
zo_<J*g7$^^mW}~Kej7tYTI}!Vr?G-p8h^L^gBx#+JwrYjkc-jy(bV7UN7b8~2jVG%
zED9D`l5uaUA=SUy1|9BLM+hL<>$j9Ib@QUCll3CM_`2+(NsejaLdV5VD!2E8^|8&m
z5{_Yx!sVAh`mEe8ACalzqAg#*rc|u-st*JbH{Jel3K2WzIxut|TP3srP8R1Fp>)26
zE)A4DseGqYZWPGClnEVz1>cc@T0<m^cKrl=hTRMlm4hhogZ27V<bxUgeIygXtPRQK
z+KgYF0YU#!L6A4=Z;(iOg+q#MselE@mB#9;E7g*B{Eak?9VSuG?|lr&CVi)>ra5V-
zg<SPeYpC0=3t%CAgjYj%rn-S_v)4FdO!x(ajrW-8hz2b5n>LGkRd_EeLOx}ZB&K+j
zwC|`kXRM^?oS5RgUJ_z=Ht7m8o6M6MHv`BTURg=I*pB`z*oTpy9Bmghw{2!k-Mtyy
zBnR1iv-l4-LklDY!;j4zW;0KUcwDJ*3WEo{S@2nr_YF@YVt=VC^Kj_5D9@tdC-Mi5
z+#{Gn6{vJRqi>X5&zpO(669-p17nGdcC)k4C^3I}b$+49)X?TVdq-R583>J~9l(0O
zurmWdzW7`@F;r_oQCL+fBWPSsgn&QVbSNyWv*h3hm6rVURxql3egwFj)(GIt9Th(#
zdjKO<O9^a-wsRwEkGTDQ34SuLAMQ-Ia#gtlhExUqmeE?hU-D6Q`y~d*!MdHk{gY$o
z9WbRR2(@FU^a#~PjRE!u_Qgb!fU+a~lnlm^C#4HXn43vTU5pFAp@F$nq&QLi7TDv$
zY`$~83WOi2Jphgu1*3;&8jxeJU9vHyn2wA%{FQ@WBk(v+7#5i(Mzmr{H4%(LTPTP1
zEK%L{h>RT)F|QzRU*f&-_nd8zgO$$3s8E_&vAxumwSMZ^R!!^1I@Q?<4!3NIMbOcx
z5S|%qt36qq>S0tUlo~dL!d}5jNtV>*!R`)n;Wn1eN3Q8bJ_bQKF*AW=0H%UFsSWj&
z^Azgdb2@vR0JAbj5}I~v7$bK8iZS@O822BiHPoXg;W!?qZjV?f9m`;nMnbcWqIQE_
zVWwmG*WEr)ROI4TzDO!4L~<4Sxiw2(yG41(^-qMl8NNxufnX37g^nCeVLRAEe4(~M
zvzA~eRQhWB8S^!++IC`z)HFU^&zrfIGT2`na*vR*80PB^k$g6EemW2Ae|@;4orGf*
zKQ#sfH!tHrL6J4n$~)1We0h~5j1Eh3ggCM6O7SBd>lO6OvrM&E%uZQ+F)pIHf<3dT
zL(Fjk#j%KmbZI}RUd+=VsO(e!8KoHb8+P(WFi`f{L5?1rCp$|}Aw)%H-*t+(HwnsW
ztkqKYj}#Ke%3+~lcHLb>=QKxW7e{}*H+*KDTA39XrAOc?LA74llH!!~>cC(zDjE_}
z(00j9e1n$TC{XHW4rWa5WRI{?jb_A$sJq&|LMTp-JW!EU#ex;)Lz=K3k0$QFBH|wI
z4fn&GUn^l}MPFs`rNDv4Y%=n!?}{PLPN$FExA&wv4*o7&oDj($GCjyUIJtL4K~VE#
zrjF-g#WTweHqpa9RocX&{JBWw5Rx7noIaCeqwVW~qRMDW?9GWOkV~|!UXf12!{AC0
zHEMCaK9lmpz|jhcO_z}O*!u=`ozhkY6%!fG@mfsfj{%^RePPV5rw$5qJ^^RfkoIeu
z=~-ZAQ92&w9!qg1nEjoS!_5`oE?ZrIX*9Lk%?Jql{!kcWvP>#!qZ1x?319rodAQT5
z=(q7k@7CKCQX@#W{Gs$mY(ibWsQfAFB8xA_FjzwsxrET@qrE$>--2S-S+aK=+wE)5
z%dCHH)%!X9F<Z*&w-<DQEJN8VIkC4*xYS6xMXUaOCMtV3moI)FTF6ug#AfIAcECbO
zVdnxZGiZ7WIdd^#R?wp*qdR^fJaN1?CIoBDhK~PLj$+eeF)zC1d18avMG{YbX{|p&
zCb85LAv5xi4mk!(SBnHl4s@CmI>Iy4px|aywKut5ycjwmCTuUbwRnb;VGFuw!R?rF
zlKJgTYYHYP4hxqSi5c=|FinR7cZnLm^SVdh$}$Z1BjbC3wk|~I*kGuLnMny}+Mj3I
zq}I4%<G468R@v`1)$z<3ZQ>Df^i|ay=vO>RkBr~dMvS8G&Df1U-sND)S+66=mZ#*K
zX^VD25)L1B%3h-KOWCV&^{5>P36U1iqk|7?94a4BjjjF*LrPiAvKBZyIkZ8hyV%04
z%9_#Um6qn0h7CiLLWR)$COF1xs3N8&L1Y;#hNUVchiuW?PK1!vhu2&KJTHv}dFhrn
zLJ4w3SN$GlZASJVM39g?8a*k@(ahzc3F(#@U!<r|3iDEFOYvZw0JDid{PXG1>HG>p
za!t7A*S{mx?o0%NSFvfmr)NAB?)Q-XS!JjEjGP#Qvd*+Z8=;SB4NL7}BObjXc+v98
zXP%`{PLLy#Q07R*++>TQtW4{JkM_w}Tu^5UWLD0j!#U;0sT^kZcBHSZ?K}MEnx(A%
z_W=tAQ`Qbkl3f1t@H=+(2=gh$kPm!tWHDQVeCj15Gu&6Men-UL3i8HiJJ{61%-1o%
zA4O3Rr8-x{VhKAiWk%+7&6zObK949yOZU`Dkh6uTE~-nH8*b6%JFY8u@2!k+yVYZa
zqO!-JKu5fnhOKT1_~voO&5v9=u|{M>1Ew=)2qvdTCPwvR)&a|z2xEK#W*zCboVE=M
z8_M2@&YQ(9H#sYd`quB2f)YYO<^XEy5KhxnQrG$b@DN-Euj#=H;2H=5$A8Ny{9q^{
zo<Syv{u4=1<t5xR_G8!2%qI&i<7qT+7`tT0M&(3;Jwqd!2bo*=5wOUz=CjyE5@yVv
z{Y0W~J|S|8*o1UHab5nX;^4^H%=f|V$ym`Z<G?aSW^6iPyLr<j0Do8L&}I-x6D(q>
zQC&%&2G*xQ1of5~rG)&KL4V6i#Jku?jM~&m4Zae2Q)>CAak2Q3GDi>r!*9f2-dQEm
z%$fLmYt8W`<+Nr_n47G#FI54N5jgIJl0Jvzm5C?4Nhbrze)`_Tg&==RLBkBo1`_Nh
z-=b>SI8js{POd{{%U$bvODW><icwXnB|2dFxAGgYnPTS#J;4{oexuiM__U<yhvozA
zDnd|Gyd*B66mf2ma#eW*wowkaKjgS?F^7I#yGo3t!VQ@cyY^tGYuF?RZh>fJy9i&s
z<2IcUTgpt6yPLw`!JOr|L*;EmTJiIml#sBV?}CZhpz|6X><*%)@Xigi^Gh3jsV(Xe
zT}kj>zu$!SwH=V8B0&?tlod%=vUs+7A&D@YP{1Km@o&w+KZbVm{3BMH224IY+4!N}
zjMGbqQw(gWWf4tJl7gYJhvb-kE;8Vlt*o4or*0(7jjXzIViP|W=v>agNHfy)5z^Z-
zSh6!?=1gS`HVBLrJ3nSbjZqo!hOpEu%_!b6JdsJJ<wE42(YrN$aT=ioL}g;+%!J0?
z`&vg8wFv}+CGh&_$xgLx#qVR~iY}t7!-g;&#u%~vNc$;%jnU*REd=h5PkSy-&;e`6
za8<rIA^CNuNymVl+e&cbKfKWSNS*{fFP^aZLY3Y@Wh9{CAUjcxo?vQ5M2OH!)Cj&f
zTMSPE5$Urb$22w{I%*c1-SJq0b_K(VIRz^Q%C?=vJ8Z<jt#$~8xoW7wpasbUp*I;(
zQC}6?EEeP6=W4bjBTiM=o1B^0rL(fSNUfPvXwj^`MU|{R9?YK-O0f4X_{e)gST;Y*
z6=PycfQi&ai^Wy5Y|l-0#exEYBM%FfEn*OgqW%qKR31(mgBe{@r_RK}r%Z%nmWa8Y
zFnLXfC8CH=>{N>AS@AC_LP|J3<O=H@51y94bFWxhD>Gb=1)8*9glXbP^2mHHB9!!?
zu5cAT4+6}xGD>;`QLykIv-jWRXxwU;8@vcCytahV==<KxAl518zQG~p3c@}zv~>P&
z93*Zj?<f5Q#>TreoabjSE=7!~h*WWbvzt(dj_H3xdh=V{wTDn{O#~5wL!n34!<I~2
zFBAXz`&yWI<S^Ic$jwhl$!lKQHqzu=vcLVdu(Z=%I=6*XZ8%`Xl{-T!92}k430V0j
z8r`&5>c0J|a2Kzaj$iMG0Dg^u3)j(0<3wS~EH@?EE493K(;A(=EY@k(2ah@4bMSih
zMk#1+A((zkBdrq;it_x6PRSI5NkJj7MJAN92SKjJp>e3;Gj443Yv~8WryL0zOC})&
z1<*!0YCw-{1FUfPM<y)AfKr?2{i$C5p0&%9>>#^hTsj<Pjkt$%Alm6*`7j3^M*DJ}
zNs(;+#H#Kx&$VGUhVn)@B?X19kx|V7SRSt&ltNYguhL_v7f}62t&e`$E~hxMn?21{
zv(}VCO@w7G0;CHGEL=V-*$qdWk6XRpmL}W%gkjTfDA1IvX13<9enkp_TB+G`LTQhK
zG<PL%=)W8|!?OmaKm;F<%TBAJvwZ&y)z#KUG?jzv%3FMR^dvR7D;0&X8b*bbmli#e
zAJDSw){bC^;K5adoyRn>M&g(E4NF7WpziWnjmr`(i46ILE~N~O^zpr{CEamg$pOC5
zvWjH6J0YiCkt(>};oL@=5_3lZlv}`x->nC=Dq+cEt3E=~RExd(6RKk2G}QmI$FO4F
z%)FIaNysgil^?uY(>XsLQXcNLw^>1}c87HdRm$*{Xg=LrX$1W@>4m0)46eaA7lpY}
zO*55Z(Pu*|KXFpixiohXmuoX$IeD!|0$+G44C7AjHkv}w@3U{+%w>Gyiag+J1)62E
z8&s(l3MPsb=GPSNLLHwyV4{<HDDLyp7Lu$cM=NK0ka|esD{uT?`hj+ZF^j}!3fkBO
z9ctDCUEX(%+7j2-6`kAInEv^;X|wx+6YmyH!qltQqRW;9HG5I!GS_RG>pXaB%|B>t
zds<zQqkkIRUpQWOg-$kmNigy&9R4$&O*B^{t;5bmIS?hP?J2t%F`PoWU39@6H{Ye!
zDvPQR&1|c-Ze5W7`4SiN!zVw*OFfviiAe27vxAN<DhHi?2M?ut*V1&+blK5^{5C2&
zW_dW)+hVuSYQ121?`sU_`(>}GU%5?RgVMXiA*pCZZx5q-+9~?MjMB|t(UD?-EnaMX
zQ>A||Zg190Rrp%9L%sOAQ=g@u@MZT4xhhSWOk%$JSz*7D6G!~tv<OWIu%6#9^976^
zkK1a7_~rdGyzIn_P{o$A{QhNoi?6d|(WOf}Rn{?UQiO(4p{BBjdhSd-Tm?tulI!&8
zSo1&XU+vxn8IWK6uGVMtb^euu{SI@YV`z!o0TOuGE4_pua*K<jxV&6GT_xS=dXeYU
zb)W9qY7E1FJ^wg_L}%j_7DAcrilBRTy}bIJA=-_2<=&$5P2KF`c;5BTZXwg%kAudF
zQz0qil2O^{SJTmuZh31@9xa;i5V<PaC%4SDor;fkpVy`^<JClIM|&w`0kx2LV;Y9?
zD_xzr?8?tKKg-LO$@IZ9e(#I5cU%}8F8+Kvh+i`AI?%Z)B%4vZKF(V?3}8D`@1=~J
zGBR9OuHMNydL>8N9M)mD`~5Mfk~hYzj<$*I5|x!oT9P&3-Rgqh!?Eh4O!)Wbs#r)-
zntK<rZ3o<Jme>KOUVrE6DNA`PO4~8673<QUzE_yWD0ah+$QqidRdmbPC<p}QL+U4N
zl{78AVU-0>y<!mvrwX;)!<o2ckjDNs&0YrY4ZfU}WZUc6w-jHge|oovZ|a%;t)zQ%
z4l)fexhG~mTfMJM6A8Z>xebgge@MRm7@97RucN+6F%3yubjW%z#TJXAW5=|(4vcj8
z<}j+U`x0<Q>g(SuyKt10HB=uZ5?|m&_tZNmH`dNB+Ag6Z3oA6w9fwY$p3A%1mM)Jb
zDF>9724JahU{u6xoPU}Y-MW!i4LJEe*pQ)$$U6AlNqv8bM3wUuzF!eSj`QjHyX=K$
z#5q#&tC}1)MZUJAC{1A0fus96ATq10AbXTy>J5iyo{hvY`s#OBRj}a-RJ}|OpBzb%
zP_hy~q=k2|ZXTi=1rpsZmQhI<k--Rng>ySplVXNDO%Iy!R#<U|{@Ud$<vkm{X5o+@
zA%LlmAeY`sH6hs`^(`TSeos+N|CZuLz3P;elYi3L_C4AyVRLEmwQUN0irS(gl-HKj
z1)eQQgwugq0o!XyqIN-F=j%WKEj@p``L2G&$sI%c1mRer%5=bd`bsv!hJ&0Ra}ihS
zJ$_tpWOg}4>><Y2l!s#b%kN@ON_!7Nc1tFPwf)FEOZs19RjE`ihL4{zc@v;TpL24q
zrCdO%G%e|#(=!xs`t5)xRDmyh6QKv03yg}0=P%xn>ot^R_O@PxutIEfQBcMjqF`><
zK7$=iZLDCy!=T)6CQ$M#H`}eGHHug<(=T$q+H9!9!gPmJ&9;EB5_HZ|wIzsEYVTB0
zw0P;q6R_udM_snajs~Us<h>D^F)}*;-LZ*y(K7O1gZSvJr*f6IT0@TbJEiDw#3q}*
z(B+0Rg<cJ+a7)B3gvVjg4suB;CyoEy1Jb)54_2<m7FfcZu%GxOnrZJ%;=m47tRdBW
zGzDDHqPR50jxqk>$nIj2QuX*=pWYi1OLu3z<~6>Miu-jJMB-nYbkAS*cwJL)TpRC3
z_9eKbRMhSj@OW<8jzRm#Q8GFS4~kX%ObYW!rwvJ-0nb;TsG_$cCL&r;^SjP<R)|M0
z|H4v|bUH+zO+<87U43_gieDG%#yHFsJZIEuV+%>Y3ip#l-=7k5q+*V$am{?TQgMO}
z&vT1$3Z=1xIP}w6<LJ1yXJz2!`&KT9<KqVx28v;9`?RG*LiZm;W8JeX1^z^lq17jh
zPai|WFd?<<r2f>Dq42vkQM+ec8#j(Qi~ePh>zPmLu=LA);usv)E*iJoGMImrABc^L
zoEx=ccuc_h*XERBuZ%`i(GUGv3+VYgLrkk1RB8<&?&Q(6*<v2HxZbK&nBm7zMpw3d
zH(J3bIe`JN9$RK3$$^%?RKb)n_&yX>XbM9zK5@!ebY!%5-~01Mp3a4If9Av4Ot(#L
zXQZkHzcMDrue@pbr#-ySlJcQVef^?{I0#A6V&wU-H9hes{mzC#2P|_r%Ow^)jpcHq
z_r?{h2|m~9ESAJ4*JJ6fD&k5c)e66%vJrl;c>ENW9hI-PRcvnMeXM9{1@h8zQ@w+u
z10|N$6@?T(=_<~e)mOp#Eb)S460JKT>2SAbZ*|xNpWWosC38jV&efC}cMbh_bGAoD
zO@+K6(Au;l7q-N{xzPNaFDc<ceco07=HQ?@DT`WsfhPP)mi>@wMB5R3r2C;Roe&>J
zA|jVK8)HH%q(!81RWb<)DaW^?vAvZmsXN~5`(<53cVX#ggYDnp$*k6kmET@2cWYaR
ziiG%!`9Dj8zxB&zRBV2>Yb3FWQDLvktvKGt3c%a%qAqjAAk>=@<hxbpxI7BX(HWMK
z-S798ozg2tQHY^o{7{`#coX4`Q;yUN;)&hz{`x&w2d;SkU*!{}%WH^dU8$?f<yE&!
z++fd}Of`?x$H;yq%N`{JF<PM|M#s?Dl7i(FrX>Zh$0tg#xx<Pd>xKQz_}zqFs30Tb
z#K#*a$I0Qd=z41W3C`~M;N&mUQ?+*EWo21Ylwa?Uv$sw0+R#z?oiu`Ud-sj6nhNKC
z*&Q6k0y?t){Nd1_OI&a-auokgV#40o;C6dmIZvZUi+uW|;oQ=~zJn};iTKZ%WDz2`
zzX)s~2%@r`S1W_?1R9$#+ePt#u}5%Wtqir$ks1dXcPK6I1opLq(9IYtZm3gQO}r52
z|M1fPpwnaR_Kv64+_j8HW_h}W3%*A5A*M7NUblfMb<_Sa`~kz$p_Y?Dw~nQcg^dhF
zgtZqcQt`A~eK&WDE!;es=3Q8-!p|g$IHq+x=~9K4KRnWc?tADJba(|rzRTH^fAs#z
zTHS4sp*rTO-cF;a59lu5nlSb(SE#&e$lIvzn{^B>l`8)m=FoRTc6Gs)%xi<w2`A>u
zqlfk2rjV-Il(Y4da*p*|Tru879Frp|T(N!@KVa>OFWQfoEtnXruEyVyd2f!dlivu6
zc>)>RQcrPgeXw^xiQjJNXBBc-)OAp-Thl!>{@GkklK*uoBYE(fuzzr8n;_P6Sd6PP
zLeqUNNlvZZ%)N(tu#tSHsjx{$LJF{o7(WfzyMf&?)6Bu~7(ZnENTH6*W%ejNT5Du#
zIBJMg(~q5)#zr>YjThQN_8cB0*&cOYZfdC6$N>6Sc4fTXtrl9sSanCczE4*DX4;QJ
zW#jN~#DD6~_`Re-lC6in&-s8^DW|r4F3^a@itd`PsO%!~*SB@-Ha;JBAI~{e6OnVW
zBh&Sgo^*1HDAsSs^y_N8rnZGi+H@}SvtEovW+15wZ88s4)I&g$n)L@L&Q;jYV);;X
zPBiwJ;bibQfN9mCHiN_F`RC&?Bqz0nRqXGT1Hn&+#jT8Ec!ucIgB4t+=r+>=`LA-F
zzQVIX!8zO1imKl*aaVzsDjG^MimcTeG*vBZZd6fxgdvNC&QGg@vy~;e;e0<ATM8VM
z9P?mD9b0nTL<UYZ8)hcO{t@IGH^a4O<(YS8O8mRO8obc8t9#{+AF7-h8Xw@(m{R`-
z^Qx(&JuH>;ErRH@j^+{*Pp`JZ`nM%_+rELX_JzQUWC1OmG^1#-SxkoM*PhIa5Esa)
z<^d5%IEy3tim;I9%B9D?S8k?Z1}zXly;GrhiiB}sNVjf}=@wsLLOQpkJpqFTB<ikO
zS0EHx1~vsOs-d7_Kl?rj5l)fI=hn?P_oM76LoljtxhWjnB12L|C5Pu+dzsC~{-#3o
zh4f?Q-BU>xonJrlbBoWmm<)eowG!mA?g5H9q$#e*io1yJ=NnZ(k$&s{W$@;>i}9fc
zb#-VrC?+Iylhyg%_oF91H1YX-{x5V&Guy0(*DhD&V|D|j+(XvD3hvgRLE^?MyRkIQ
z8ltV6wE02_H@0w=?0Q_ZB6ZiZ&X+00p6_X%L^I}p4&oCfef_&zlk?3dKRxz^b<6|X
zo@`D7-L3f-;wqw?b3FIhqHwx+|LU6`Ut=ulePC}#^{GLUOf*U4R<|1Zk4-)&_|_mQ
zY}b@-O5G*oy4y2<<akeYy{g~bM*kfXzZYJocH&?3(aiBsyupR8MQQr^ZqA>c&18&m
zL<cgVomo~uG($6Q>;aFidKezchH_90RmB^={OZsMMXHfX`#E`HFei!b<!0b_KQ!`d
zBmSFHTchdsZIiGNI8qZ>OF;=U#7x&S6_Wxv`_c0m5#@g7;T#Ce0nGia_>@#$DD1bI
zfcT@j_&nheDUs<Gakv}7-M?Eht=>d}G(P}q3{I-JBwWgGS|E9&0(Qfeu-z<mVak?x
zgq&Ci!|(Cmn&8m~OUrMZ@>_P*KArgHCz7<Bl!@$$sd!jq#TC!7!ci4HF^7v)dlRK^
zpohPR)V5izFj>=#Z3)6?oL|d)SjvvN2pTZdw~lqajw+%>x291uE7SI=Q0?AKU|7Km
zfRz?A;+%m;=1lf6iK3Y??<hX*jrYZcBtrTJ)e8%kHKyEKP!KQ?W3OYX|M(+9jk%6C
zec0KntHszWU<gI3XiU0nl#7GWy_}6_1|7#qg276@<o2Yb%7!UD>Gh!djqRGy*mt1A
zu$nB!gRI~dneK2dxifC{3o;Z;7scBRPx@BppDxtV`a}&+zjSmN<=czFTZ70a6%*u#
zXBwkj{_ocs7~SDr{%)$+g0&;v%n_bwp2VZrMBqOjOS0z9c;Y;jZVQqSF#jc`r<+A4
zLr(K)cQ<GU_`Tr2-ihu03BiyBvx5YKp(8%Mr_}iT{jWq2b2hh^$*-t3@ooy!m1Q<d
z?~`oPfD-cv6T-fEe;n?)1Wo~hIcbcwj?3E2#Q%nrNzL2@-dGMQR!=s9T&x$SDVrAa
ze;9kG7}3IRO}A{@wr$(CZQHhOt+H*cvTfV0Rd&}oolgJB-r4D!dHrSP%gh|_7|%22
zD_$9BuK!Z-+BPFGB>G^~5@eu^r%+wrh!|Ni`8&Z@_*b(qB`OFJNHc+SAwZ+MCVhGb
z+axJU&}x1bhl?HhRg1jZV*kT@#y}-wBVc4yJy5?+ieKSwkoPyvmov&=F>3CUgGQA`
zRl<SVfg4yR*=Xq!9B9`BnH&nK;QhD-aDPSj5Yg^~sqd8R!4J^RyeJQj-1|72zTY-K
zxEvkd$b5DC9>XsizG_Qjx?gV3>p=@b9su>o&F^I_+5}mJ%mtIe<ZoN;6EUv`qFd*!
zpY<{bzXISyM)r?S8?zdwrUc5)!W?=++2J*4kBJO@@N&nEPamEIDS5b0Ywpqlcl}EV
zAG0U3#Zo>=!Q&%tm&K?nS4_P9>vlLJc%43G<UQFZd=dOThSSX5A7(c@_S9O(`IwgA
z&NvNkO($pOy#te3YZZH%f=|{3yf_}UtcI#s8-wP~;B@JahE*npH1xI<=KEnV8NvC~
z5}9t$M^<&FL-Fw-gB61riR74~*1l3CcEpC=-g0{n1%)F#hNW-$nW3qYqHviwKDx|r
z@*wW-WwtZ#g1c@34KkzPdW&eCeGm6d%szEMn~1rYp^LqbQ^tu`lF2Y9TA3K)aJBDY
zK36dM0e}TuE`wu~s7nij1BzjQ73d{h4KnisuT%pf3N-)xGqaFLQ+c?Ij9AXo(yBhz
zQ$u%|RQRzO0~Z`hbD~zAdQg~7$t>6!i0ky5$I<X$0tOn~{u~W|fdXwbtxOLoymUb6
zN|VoJmQ)=Y6lDcQ4Wm%Q$uOdY0G+(Ko$n-<<t0hRj5d%bd-usqkC<*ZHIBAa#u>CF
z(S)!*JR?ZYOQZM%v1(spE&x2KnW3`AKeKubcy8^)_(=vC5C@U_1`@+q1F)dLgm$gF
z^|MVa%!oex@b-mV0(?%^W(TND$o2rte9E{c4(`1rU}Sk<VZ*#HMQf+xD@<f|Dv+2w
zY07%NqA;aDs!n7;FB@nyOLkIW#S-<9q40&of%sJBL>ggoz(t%L)E&$AA=|xQeQwHu
zw-H^C$N^cO%}n<(S6Y2T!?`Hv`aoN%TPNZ5vq#?o$L>@)?MbWzeht7Gi29lV%?U5H
zOk{Z?EW^lx*y;0)FF3+reAmR=)vR+#5@5`lXl_<PJ=>hBsL<`kUW^S&r8cj?@eb^j
zbs}EDwFcoW%#Oqv`}gYuhfPzSp~gf`Lf<P=X~?$bHfv4x4?=tVZR)ck*(gFJFzqp(
zy;zwR_W=Ni=?O^Z@3~_lK8!yzvK!)`qCb2YaDBbguE<EpfXe)os|&Wxstsv8r2dS>
zjwU}o>%s~KL<Mq%MhNJD!nt5#kwD7dby6>OJ@O5&BS4iNz*ZGh0@;F&#@jMD<g9~&
zmU94tt;Rp;WJju)%_L?&`Bm&j>e1|qhC~|2#XkTkZ0H{|!qUDOXJA^lE@?cE{1cl-
z=^`7ytxanDYSNo2;q?@Nagn*ZI|7Awc=$+m`?9=913bYGL<>^8!o*0TdK`cGYLAdv
z-J;ZtHPow~SE)>a6!0_>vQ_}rD3wE$9b&q%-&4X{ImH8v?w3rR&rgdWT6FqFFjVp@
zffZx*zGS-ycjh3t3nVxRm@k%~ypuUUyD%VnXhoVGljvPDN1CVI7zNhk@CChx++;`C
zS}ZX*=1Mg%lLu#56e?{UHoB4Nk&8?^X4?mvAE>>tZ3F$AW@VV?kFO7cnMzj_;YCK=
zTSq}`BlAxNFN~A{`IC_Za4sQ`B0?=ClVVY@qt>mn2FpCeqlgEc$WbM4D(wR$(_Lyp
zuAOQ(t-X9b&yqR<N8*3}F8Ew&=m)x|K))@6VrbnU^b?k(?%gqTiy97Upf_dd0PAE<
z#ww$-zTtmrvzFnZ`4;qJk((=e3c<|5M^>ALNQIUcN$VI+em#@BDyo%nCX0iNZD<_W
zUSUQimq;%v@!)DFOG@SjaB0D>$q|5o!(22bVxtQOM2C43aqc^C78)PPT)}wHb2o!R
z9MB@gf1Y}A%L_*#%S4;zEs25b+n2E>%8q=0q84l}@x6#6eV)OF@ZkyZOz%;t_`H}c
zna|3gSdwTTl`AMS<QsxbyE#$(i>bWyl?<e5dO+;M&C3rBBoSEoYJ*IR^aZl#;!IEF
z;3Z|Qg+G#1gtVyezZnStBU=X>Qpq|}_#<d=>9eaF9U#=B9#X>g^IMaz3U-A-xRMcl
z08W1uOIL6p8lsF208<|=hKlju!`I^}S5M0Zth%>{4p3xGqy)`B7Ev2{Mc<({1Jt=I
z!^-5yzdmGXNL=lSj(eO00LF1sMVEU7KfUPtt(2<evM?ZWtln}_rJ$J^VIZ3aKwnT1
zL#E`MlSbd29=KcsZ=qE4EqI7Rz=uV?KEeg57J@J#izko?G9c-*E7OqWZ&;ZXV&YNp
z&^#1eb|-d4Dv_xvP0>fDEwIk9mv~^9k_EL<wxy}0W-g;>8iNKlrgIF7VI%h!a);L^
z!;EB$qagVNtbUKHO8tiJwd-1U((bBcK>jZ9jM~7_;K#?tyNt<ae+v}S7yxn7<57>X
zlm~|(_Dq$f^Gb(Ac$VD6)I>M)M6R&1@3>n)55Ih;npyk9R(3U5+JGFYl^aIwgC_<Y
zJS+mJetTB#MyN8)-el0l6i4Ku+iPPyZzHW>WJI`kv31OX29?+W_E5a^^sClz1ghK!
z#JhnVuLgga1=WJ*XN1_1KkOh9k%oU-T-?&aj-QP_Kvnx~-Vhd*^zS-@oE{;uGS*P`
zVpHM3iO#j4wkbDF90<B@lG8w?YL<AExwwh9O}joIkf6*ei-=Uog5IuDo3dCY4Hw1|
zZJg}jGtNq5BlG&OWm9DTw5EkI%>%ik=+|DLV4=M|^n2$WO+SA^|E`nyiwuvs*TLp}
zw-lcDN7$hZ4Fp3+P=UwU?y&y$W^ig|6d+xjLqm&e!rVM3Pn3HD;WI93?>%(BChHR_
zMIWOwHj?NZiG-ZZ>n%TCq6RWRZ}?}%YSWIiw{A;p-`Cvc0ugP)Z{mhL)|E-o1&p?{
z%lW$;ztm<_nWI+X)!YlqAFwktR#wfX?0|ax@!Zfbv5tjd9%$D`K5Lt)cs&BwnsgDU
zw<x>0+*iUQ-IKg7;h$3VkFax{eIBpP-Q(ZJqe)C|ggDgO8w56F`|fpAMN^`gE@@q1
zZP`tWIt|{#UO9?hlplbctWJZV-2-D=%VkAi<a-aicTzGcn&#_b-FR)R36Wyb4mB6(
z0ni+Y8s>jk<(oRbao{2D2|enSGDGe$QOfx|lC&v|ifAg7WUXHq(Us#Xt3x6~=TTBK
ztt^B>p1W^680%Jv`4#bUWGgCk7%)o4LSvxzx4{BUXB2PQRv8h^#A?OE3w9q#vjm>W
zjr1uS$f!Zh4L3*nc?`doZv#phU6s7pBj{!9xq@gMtYLut;4cZkXI5=wr0TA+TvXYJ
zhzZa6co;GrL!JwDX4Pi6#J2o`SoredlTqQO@-hV+Qe66lkGKpw>!w&`<#lmxq{J|@
zl#HfXX<DRxiOsE~(F#tkOiqQ%o9^qeID+b1X)nv{X_XvVQf>N86vc^@j)ZKL8|AjP
zRl4ZVF7Knz5<?;eNrE0y8PD1ksB~2lT!<}xjUo=nxrEiPg%=xHIh9AH-t8ThM8q$J
z_K}gL4ia6bgpE|;Ap4tY*72rN;*V5HnPV$SZ2sl3rXl;0I`B|;5r&|kNB0gfD-}K@
z5V5-iUDYv@DCPQJ9f&Sf?Pb2MCz^w*#3kjMq(r+H4wdJR=b>6Kf-xOR@oy`DAKq{d
zhA){5zPvWAl}4w6gIf2C$==T|?^Iv~>eNHiL5N+zf)Qlsae2ZM<vQ}%Zv9fpT#5N+
znpB(qSg8<qF4CVUmK#W=!J+<twml@8T%$A#U2k(DX{c#Z5={1jFkE>n10t8gZT8zy
z<kcJOBhr3gMdgc@<NVuctaeWszQ`}=JG%Pn`OQX}4$z;$ufsPZLo6S|@D1lA^nP6Z
zA5U8E00Hz~-lu&u2?E?YPF<VYNBS61U+uU)IkABKFS)toqrHH5Qfe2ey2{0vto_<}
zrhhh5@yf>a>cPHUN~?GbVs*t9=h<#E)B7&f{{0(*kNR}!8`{E!M6y2F%9S3gaK^$9
zFAFpsvIq5z-lj960-|m;xLO2Nit|~PBrl<5XuP$zD1Y@$O_i@N8z(#-QrFFe&?_2f
z-ux&H%-y)h_}4<x9(@b+gvzrWY>uz4?FC;wEww=RBUWpU&+zd{y;U*UE&EwkfPvsT
z*3BW%dXX2l9mdhB5&Jh0*2PIAZIze!TkSL7AfbWV@k6cKNXn-94e%z(-^0bq);#7W
zF#`}D@neW9CTduJTpn_B)0+2JSEoGCIXR^A&HOA+7Uj=Wu~TSX2iI>$^+8>_oaG*W
zu_nN&I#p>mx`S&)nakdGXN@&a*d148bBB^?#IES^7KINfKysi&UM*?yRax9uF1jqX
zLp6K2(XKsvq9-+5bj%TEepsuxyG(8fF7tU}c7ly|QiVf$Z?{29RhU=?Dd^oEB|=sV
zo=DpOtmSsQ-J%IwGP$;LiLyMSe!i*pJx#n33d)s@;PNMCYpc$6R?imLM&j?;dTDKE
z4fJTIt(TZ9+>2soh_OVvUX2ez(nPAAu~Ltw$fXxZZbnuH)vFNKrt!^Ufn;)6;LXX!
zHRUw1)k2NCNd&sQw(6ZmM5664TQMe5gpp}sQm7as!xE)v0wkx1=828^|0RmXK_Vj<
z-T|1u_=V6ne3-|z$baQi7#W%W15o*&@DUbPhW|^C{0BbLqBCh{(i6M$qPzs%Or+>u
zV9>X03O-IJc^l-PYGB43E7&-GDUKu>l{C{(g#!{BcI})<kwnC==kh3;?*H?-`z);g
z{rO?<)n&9ne`1uhK&NB01hx0CO@H_MUiLUA*ZO6+?mF%M&)0raR3P2Y4gBA!<L{r{
zZ<X&{`Ny8wM_(D&kGE_0+tw|YiB5p@{2%5qD8Z6Fx_K+pK3%`>&d;YxR~9YqIjo(Q
z)?@QI`d!<IV>t?UpErBsH{QP*%K2cQmVSG)1(({7VYKptu)9#E)fji?V;rqrJ)h4H
z@u5P<behKip*cc-)HorDZ0V*JLp8G1i8cjpU%R@}Ep2<_s$vn!CQ<D_j=tnbEqLTf
zI<dduU%b-IJOWkBVdyEB%r~{q9nd<1WD?wS41_QnW-qrV-7WsA%yjZOT5Z`Rv+cZy
zU#gNUgMM!BbcO`vRVLk=`L6b=flU=ZMoNykE<EEq96yWE!z;ae4J90#_q|*c|EB9u
zOBAp|0ksrPV{#=Xk^wty;4M`0l>U_TEU{?*YCUL!D6wt{tA<<DjfGBrV?_(|>j&E{
z%#@4zPj6z*7Yi`5j?mghnEk8fxJYgG&4LPw%Z;_M1o##Tv};Q$^gSf)Qf~z;*W8HD
zH1aeY$J--XA6(Kk6Yb+EY!B%kM%Sl2j-?s6*0Y{Q7M0$l(UOLAzNyG`9;OS3#P99n
zdL=q!hP!<F8LOI@z7#h3dcEGtp0M`KxQ<<oz8Bs`j9r9*6A2bH6+MAA;i2KJe%d+L
z+MTwime{L;XHYK2ks=f7Hy#bsCW_cJ@p={3V#Kc2s`l~Y?p?rDfg(isEX{GSw#yms
zAXrzCBz-T`-XVmzUh!^fba`8H$9ld&G~zw{JV8B$Ul`w^Ba~fG%Z=?egwp`uNKw8M
zj%y_Yg377YPD;e$n>GO`0P&hMWJ8zr1*x@E$FF2HR2P}fw$2uk=TjzU2bs;1O&L4K
z&%|i#{eAVAq5#+AmW3aLHdbjlO!0QA<BVmwm41Hv7@m(Op+O1jB-jaQRf@fubVpLV
zdrxKI$*_EisXam6ipPTqJq+3$;dklFeeUeu#e~y<XaZ%jUK4<FhbR-((U%;p($W|3
zLREs!%L4yWEJVXXF5Mowx3|<iz)eF<c{ibeV_!Q!t58Bd)?!0M!$?KwMQ1yE^9zf-
zn+Gj?71vrSFWdb$mfH-;FK&$GEH>zqu=O`FpQRp=hY4)DF4-cwo#rcQSuI_W=%{eR
zxE4ULYK<qXzN%Z*g-$=!*?Ewg?q-4~#E$}K`z2(sIavaekpu2Rhm?mRq9-lIu(+CH
z^;GCjoR^6^t6`yJw$7lO6m}pD9b?`tcMITsgLkg#Z}2jG6><MqmNk!W=){LCvK%}V
zR7i~5WDKC;6nlabb>pxh`Eof5I2GN-qphk)8yafFck0D8yhFrxS=h(b7<5Q05cWzr
zF6Nfv)-FPD`%VwEVdf<%>@G)#Y^dtl;yl)ew!4lK+B#0{*)@-b=gSSqA4U}dkl-#Q
z$6Qen*H3s5U=mX8wS1oyoRI;6$FBV{5hz#>-2i_=UJP^FMD1h1EnJ>!65@TRrE~(!
zU)<pGz(G6SLDwiCSd3Z-4U;^Z)<W3VWeK8hKpfAcP)_jbN{M|W^q8c4I4s3-Q%N{9
zq4q3t@9PvM1UJp2lG%qX4?TSv8!_9Aj9;AZ87<QpCyvI>euuzbfKe5=0y0;=H%M--
zu;_aD5U@YW7o?=%(-?M<_mfiO#=$cagA!7iH?s@4*;(x(+FUYNlR%Ej0Spes+BidK
zG(2*8EI(j=7KpMg0p*v(*YhZi22r46cU2QcE7;h=;x>p3@p=wi<rX5QqEo=j9FjoL
zw4zj#?72fComT>3mI?)Xuhbm6r6o%kZH%!QL~*sqGB@;NcK;E?J+Qo(%D(p2U${;h
zEiI2El}p1f4H2F>yCo)*Yrf&E90N_o*u%-h4>!v$1&Y9AuQ!Khk6nlxwh+t-_OoPj
zgw5vwBW(*&q%eYJk$4Yi!Vm~#`GCwA54(mpD3d{tZn&iESZgIu#7Qe!O{XA8D3X|D
zEp4QlBl8v*PcqOhDmn-g1aS&zcgmXG$V5D1c)>|v$xkN3F+L|-3R*`&HkpwoAW*_d
zZADJy5>z8|#-|A*c}!Y*=dwF(`d)8?=j2N8G^?8>3~Ajb`noYzq_E%c+voct7DYN@
zOB@7<RBqJ{guqaaJXfKK)k(GcS#>e=@=i6s$sejj5zhO;LlB_a%jsReVeU>+B!Pjj
z<d06s92aE*AH>CKf?liN0Az-J<k%9%IOHI5$fi$V<~mP&bkA@w0l|~pQh4A6{>?_R
z#?jb391?4bB@b+YS6jy*g(|~HRm|s%A>CNcnu3=;ahd3|aYYoGbI6qjG7JpUL1J)Z
z{bzeZDdga63-z@4Vv`lK-hN|Zava5sLdJ_=zJ7i_o@Rf`?rW!xX^VyV)Ii-C9|KWD
zoop_MtN<w#)+@rYgMHlR0EQ?=MuTgjw3sj$PN3p$z!3l4o@mU~k9LW6z=b5LOAL@<
z1A?$hRUhqjCg5w;2!<2WoFeajLS&XDoFVc7q3?_o;b=qZj$SxPT;5^YR5qr4pfN!V
zljCOfMOJ-jKS)=zXS_Y$>am5;9!JDtl_EfcjYuUt=bY1acv{LuQ9Z1Y=}fjuz|3o$
z)?{!eA?|J90#e;aVSlaI$u4^*xS~ellAt{u)dxZt>PNym=|Lz#GIhulux=3h7mi0P
zvebb;d%xZx@Q#dr9Lndrx_tMBD{;e;>`o49>d_5ZZN4a)=OQ^_NxM#nt*YSjV~tpj
z*=W7FO)Gm0lbf6s86?=20ASOvO0TaU0aNsM&#mw-E;_)6qEsqnuoj@CX9<k2`x+4Z
zxWvY$E9iYHg$G6*M&h#TMm0*v3%Z1nUu%)(yL}KSF-dedp{55*9U-N20*Z+L3^x0C
zRVXRDNC>Sw)_k|nXp%%l$2@ne_S7<bRF)u3g)qER5;ch%XC=V%A_~Nau=N_)=|L*x
z;i6j(TV+#G({La_x&`IG3#N>$c*~L%_y*k^32pYMmPfs|bpSY(cvP5li?eWNNZ?yD
z)Y$C|+>j9^Q_Bh{8!-5RlnI_CJ)C_+6}n#)WUF3f=mcJNZSV5tKNG4}+WI_n-I2@+
zzxrZGnwHovbrECyTAc>M`Pz&|TTC?*=*k}lFT~W#!hM39^`wW2Kc>QkEkQZz6LSe`
zhfWWK4)5`vM?#9aZHh|5$aG?wm|4<tOTPY7fzF{OrZhK7_znU=gt_Y0XoXQLv$%XY
z6d|}@)l{A6u@o~);?s6Hmc|5d!1GhKOkPYso$6E+D#VvzUZkfySEGj}wW0xRn=o*m
z*Lt*N!-3V>4p`-)&~2FMor6}R!1dQKiyXJ94x7%JG9L6tg0IzAri1DNAngT0&Crb6
z%ad{|Zcf#w8vD%`tbZ_)QzR`(31ot?0QGP5^{SraF!R$Z49<a%y7#Oa8SCGmes4h`
zvW1hJ=no^Yn~bED0wYSdUnZ-bcdDz7>jDIPeM%BMrxS_?m$@p?-;xBl9-+gupCPiK
zFs1{pI~9(P_S7ymgVK3$#mzd?*3e3F392^NAB<do-nKhUvtT9$Hc?YCbv8~|#ABqf
z#iBaH+Pl%wC{<i^#d5DX@qa8Hhb*#ER2{aAJ2RpCfOF4U4{&n_%&7Q<#bZSve4Zso
z^ICTp<QqV{t&VtLF+6guri*K_%z^5!ijymXR@8Z@@AWYLT4%}kkWVqKBr^x&oT$4W
zlEh@<@=kqhE47AvuL@r#d=wv!E0;YR^2A);?{j6DB8S>-cTBUvnIh|c$5#5RHXv=U
zlTiYPO-@dEljHb^_R+LZls~6Kt3uXOX1Ebfvs3f8YWPRAf1ak42^fa}l$Dtx4iL`|
zJnT@fitZpKlP#@;Qh?v)i3Ey`xOv$VrMiE$xxi)hR3`OM1d}FnY2LW8t2dbb7{$?$
zez_A{N0_PNgL!3E!ZLy|*GmIJ{o;RS&3=FTLS=T84sxoY$CIO>x43)LG*3esvgf){
zf$Eob>d3=MHCD9tA0ghxxp0SxFj{;iQ-%doZMztecF=0hMe9+iHSAGD$dQF-L)!Df
zJPfEPVlR2*3o@e$Y8r?>=D21!Pa{)<{a|F5CU)bSkOBFqrIj~;8`VWek50<72S^b>
zX-B>NfLDsKaIVhh4bRp{#xttH?n2_JYVw!MmDKobL-_O~>S*LvYJta*QZH9rHI1cb
zwojD=wi{kwwLINzkT}m~igE~&p{)(#?44xab)6dC_iqj+QJeZ;eZnr<fCh41XN4L=
zmkGMe@j0uF>YS-k1Y)jSw&OMxqW2eg>YLRaf&CLuxuAwcH%keUCqt3F3f*~8!&Zw>
zz2y;c@O}nKaZ|!Dvq{%sIhTXs5dP@U+M;hpV;C*k6-Je^woI&9J7G2kTnMnd%9V6<
z+_mMab1C3q1v1M&d#ZX4y#~rofY2+Ad-Oi@<sTZw50kr?rdiJh@H*|Zin%R&ahr>o
zginIvle#_J&+Ea7+3`&Qbm}K|j~Z6tV;|>wu|A<L>}Y1U-SaMVZne}Rcf{&D%UMek
zc^>g4c`l_dVp|v5v0XN=tlV0Q$(E!+Sw{h@bCjHx&YSW%a^0XmYDOsU=Sd5afoxO*
zn&XCZTe|`M#JS*)0E(Z=rbBc~x{$!#@)Uvi7$v3}Tu&WPmtN+H1#L=}I+mV1mCGjX
zY{QdsLB^l$xyKWU0P3GCSW68~ax=I+Z|OOZE$hW?m3R>`B}4BtbwBVSc|m3#5G`{(
z1X#T8mVR?fxAbPJbycumZq>fN@Nzb>Qd@FqEHm|_?QDw_3?j-?c1B63e`aN@IH^0)
zI9yUK-Ec9(dDe#q{lOD7fh+w^w|rt!wij)Cq^itFiPs~t=i30K9%)N@-rO=W615+$
zRYmoOvuz}2D`c;$zDTOAx~!zJD!j1pa{-_F+1^yP*ygAUa%}twH0Zy#%r-~p0Mr;%
zjXUTfw=tAzQod0Ip~FiP_K8h+LHvn*i3%2W$U9m#Gub)a+2f1fKH>Cb8@XO@Q`B@S
zBnQuDZ!0=IZm5PoM;N``l6x7!r#|nE1)(8z8kdh_49c9TkVqsgN|HlY<_BXq=&f|E
z>Cu@zo!Ll8<c5l@&N$Cy#s8$+b=>5x5hp7ibSs=6z6?Ikww9ZVX8wBp6YxDB)-ubC
z`{?Q9zYqUT6kyp$qc1;u?<&P!u+@j-3lYM61j6}v_Ovg@@_hs)F2%kc#n>xU4^6d7
z;q5|D(sF6*>hTXFKL43EpTn8z#MR()Wx%4?)u7_lpwsj-kf?4u{Hyj{v|!WeGl~mp
zZfSWmmjf@F{-foj1rLivI+}h_rzye<VFN4$-vdM3<J|XL`rgyLS1CS^S2kS{Y9i{P
z22Xvp&%u=|Y#Y@2zQNQK5>;FG`Gn^x=d5B7<;Valv@Pzp)As6paC&Y2<%O6Im$LlS
z1OWw=_O@AP@BdY!^W**YGb-HxO{f38ocHN^KT;V)KY9D({x;dhzmFc>-6N7}96LWx
zXu%GD)x)&oqf?cLHUs1REp|PggI|=a>D2%UQh)&_{-&7j=!E{#ZOz?t?a5Yos>|@0
zauvG&xN;V{-!Ju9Ru>Li@`?$@eA8wFsO6QQJRYwUL(Ai=<uVh3fySl`6d16mBK^1s
z=tA$lH+Hpqj^-<EB^$s0qMt@DJ6{%BE<PBCZ0C!5`hHMwRS<-8k{a{5vHx##of<bi
z-K*teaGi$Ya0NYEjuYfS=S8gH<eUGg>%2i}H-o|Jc)OElChsT@)ZVT_93n7o)vanV
zo;M--55*tOVHVAz-WRGl>~OdGPR<?6c`*cCmBMuD6W8&4lEhfHiDFl&e7<J8mRZ`7
zWw9Kx>gervc4fh3rT-@8VPpJ%j(L88`u`&d`Hz@qS!X<sxC620Mg0ssOPDDLTY*A=
zDivnd_No{5#AL9y-Ea;4r?&$XGe(9|m3NyVMy#k9C3DN@caW6nN%Hj94dwk+B=7$|
zNZ!+7bOG+nq-gnAhS1#j{Fq*s|NSIcl7et~du9AJ&Y}Oey*Q|BlK<<zic|OJ?L7VS
z_FEwTLn8lFA)P+ouM^W>ipuly{_)s8zga5L?4-vjp$tq7^iWQ}H_y*^enY2u<Z~CZ
zre}FWPoF=y{}VdzVR@{MV@%x{yZfYyy@t^LIjikX@wvZkr8Y)!v+VOlj8@0akp0l$
z%vSpE+DM40-Lvj9<8gA_o3cY)!&Y1E*@hW3x)ro5ZryzP--buX<K#(O=j?=y-G94p
zD?o_v8<Sd-y0+goI~)2Y&X2yn$4xzRK3Ar#6kQ5lE4uY8UYkGvu1)P|Mpk!wKc_$@
z%gYZbHEM#SU3}p5Z81Yi!`9j^n97vu^w;l&G!t&s&jj{`{`PgqH?|e19b?6PUyjw~
z)7GTQ&1k>N=x)GHAdSk(bj;^aUBizsPA`QO^d8rGp0VlRxljFJaj~8C>jWKb33x{6
zuYyp!A<Qz>{T1D{R%Etug5hp&E%WVEx*B&MrG_?hHu*J5VGpJ-pUenqquLgvt;h)$
zEtYHNn1b*sk1%;>ziWjVEm6C~S3=34XyUi6f=kq=>qm90S$bQ)*nhCTSm5Ba?wZs8
z<(~D5Tcs4UiL2lIo_Hp1WrpmOw&~Tpt82)yyhQD~SXGYuyl7vIJ;$GJvpi<uIKGO~
zZ4qLAC{%?p%h6peY*<Ao39MF^>SA)IGdI0y8g4wTnf*|KxccCqnRL&@tU=EVXmsDx
z-?U6pD%m+*m`+V)))}QoE4=Wz0Hf9*jH%~rW<$+5iMSf-3A_;5JvEFO*2^IEtZ!IK
z-z3tBb9fiy8bDB;=#&XF%!GGQr8G{Su9T`MBZP8;+T9YN2KC#DmU-fFSgQ_gBmL0e
z1Mg{G>-k~1nWT5p9g*?&Ha?+sQmm}~^D$&c6}n2}-?i#v5!O^_uyN(4>56Q!!MrJ-
zAmf<%GKAlCGjA@O^{FsGWH0Iq6oZ=4dKls!l<(JdeJ7l3H&1P!i&IvoQMD!2q0DxT
z+T2hF)2Rtgl{KwXg($n7c)Yw}B;or3uS*HZt7<H|v8=9aQq)We%;t98xZXlrjUv$S
zN)#_?iQPeu1vD>;#*F|ap&)r15!!?!lPUgEDW2+`dbrT4RsOF6ii|KpT3-!it)vxy
z2ey6HZl%^|29L0f1;!k)vqF1oS_zcGRRuz}L)#w<bgWD($gwpT6T-N6)zYmJ==_ms
zU)q(*Maq_g8c|6*VJK`le$-V}mSyAR;DJXNQ%R7Zyn7VrgGWg{xj+o7kUS?-qBmIr
zj+qHzDZt9Psw#x}EUoNQ;nW9{V;qvjaiB~2tw1e^5o+Epjk3AXtJ;29W)0Chhbly~
z4<{ZFVv=WZ{e!zQch$1QD!UlQY5HVTkRmE=7#B-i<xR^S<49&BwQ|@I%6XkANvqS{
z+BW3y!ab%gu{q`_B8Q*zi*sS4t&@#e)=N_zA<k1XSQ-~~1KpF}iGK`m$n^#h-gNye
zv930xM1yspSH8Uu7>nx?PWsg{apIyt1_#kQvb+jj%-zH$`8=6O!LQ3D*s<h0yk8F+
zbdxR4fSw2HIbpz_ZOXK^vs_UX)f)npKor#T+zI-3$d!cj_=Anlc)z#_gb^|ztR|*~
zRq?VR?0s6ttXR`^pKIO;-*on%^XRuOvA;kuN`&6{RW%-6O@~|)Z)v@Tdtty|Tfw{)
zX5b6|F$FXIUBy0&N!;sdH{@Dup01#_fq@>>XbEX531F20s*JnBV2jOphS%r-JiN*^
zXM25Vm)(0it(!!+_oqpJV9M4bLt{y4Q09Oj1_m+v->yDjgBA<27AGCGzA?e{W4cH2
zpBk+Z9W5r~njh?<kXoc=YpBzeJ++B?cc?~Zl2vGCziNWck=*Nzn9b=hdWL2F=HQg)
zIy{;O#|MK)vTyngjH%yMHXj7gh>A^pgiJB)<|aot)R`8Q?ICVgTj|qoadBG`NDUO=
zea`y|aKSza+Ee@Wv;=DwcD;1mb5;sT4Owxz+9|^D5yF%&W2{2m!O`tko6C}yuv^d~
zNUH^v-r`obO30!If3&z~$iPZ>w)FRal}BLULu^K4lC7)D7EI%gf^)X&Z(%vh$iRCi
zZ5aJ#;D=+JKDw~r@I@{}PX1h(2p&&gB`nS6)+lA<g;v;<zzKK?bS&5cbR+_f<uV@X
zU^qR~a-hAtbCr`hT(?0=-VV<DNDQcJC}bo(qJDR7SRxIV2Doa&u5Wn^9k8cG^t1hI
zo>7Rt(-BZ10?ikgdG`}A$%l;?5EK|{r7I#ry)r{`Cj3iJVMJ^yUEB?see5^dN3q7O
zf*aFrW<36nkOPM?rp+7tBj75*0n|C=GS<M}JVZA>b}Kt1FQ&g48|-1qVTaDSW1(xv
zYenGPr*5_nD;Q4iRzL%|x;VD230vf5;0&pgz{~epa}`6gGvJP5Uwxtv(ZcM3`*z!S
z?k87G=x^5b2O-kQ-<@mZwm7tluEyCa=t_rX$KdwuSlCTf=0c!V$VF23z6Ub96WL#i
zNe~LJ?2i*|qScU_8AL!s<RpgX>r~ldQddvO=y5Y)j9xqPs=82TI32i#c}}`K)mdoi
z@j&B{Es!w}_Gbi3p{B1>_hmAEm`uW<!jj<3E*@u$5ohj|s@!(h=%J>HTG5r}Q8?&3
zK0yaUDxhZ<v9cZHi0gEncu5z)D;p}1Jcy5U%Gs=MA})G@C;c@Pq{<gdYiljEwQ}TX
z1&xasD8J@KyaStJ8fP<gYLlgxS9Sx0o?24KDFow;5ZDDY2z6E~Wb7%6Y_%x@>oemt
z&@%UV?bDN;=;dZ^1cF_pw4K8v3t%4O<QG_spoRLCEtSGo3vo(cQo!P7EC+rum9Wx0
z%y$3;o@RysBFz>=jq99nBKxRl@bKS9jD93{IY5@}tgXes?S^q`3a@54H#FCXchEDi
z+|$|5F=<HS6ptM%26BPFcQD)Gl2u~3e9~I5%8}Bw_BXuR^L_ylY~^QRT#hqFAR((g
zg^5IarlG94zBh3y&W8MZuZ;M=Z>?#{*vgFv+%z>3s1X-^M##20c*a<Bv}Ul7J1s$W
zZLdhc28Qqt*>X%qIc?t<AI~VSm9HJpaOkh~W~NXGIzR5BU@fNNtLSyqlQ~MzOJZU^
z)x_7=BU1-=*27p{TE+9>K66M%JGzA1J-v(EHz7z$nuofgcg8lsXiTL{uNov$E`Msw
z^2}4fLzD0e5kL2s62Bwk44#@`B0JUii6jgg&p)|wdLprmojoDl6Cn%?Z7IUSV~C0E
zhB+hN>aF$a*uZ!kGZ|J3=G7MS!KaQULaEk<Ejy{8l5+)Lm8$hg+z2l2@@K&?PozD`
zg`ImLP);4$$YLb`%$U~$_nKy5YH`kWUVWr|Uy3XlX<>c}A<m2a7wPz33ZiPK802X1
zwKUWa<9UqnxE#$juoh?4H-U@dd)D{UhY1V~?<06QQ-lfRVrq-!@?{9i;LSUHx@lk!
z?sE=}dT@dSHyo-@nxSTn27&bb7)_-n+eElsEM<o#Ep3}t*Xli$Ir}7jqJ@|^C6J7H
z{2Rq|^6lfD0*H_*IR;x*&3}BX$#culVtKKdbXR%f$mQgmu;eOm5)GqcxJjMJoDt=O
z|3#9SNjz|5*fij6QX0P-KUJ(NmW)!6?rc#NyBj!(1DuNGVa*=>Wv6n&SJ_YllhORS
zA_I|_54<s_*t#{d=&Ds<9B0eJ-(l9aac!kB8VDfAt$8&2P9g=jnQBnly;3n;62jCL
zRYZ{RyGUlOHA6jK!NIBQC{)|{fbp740_`azziHV%uAUO7pOx79LOaAd*E91Qi%z?s
ztre%tBR39Oq{c7%P@on5px$kkHPNljrai3GPRVepzsS$dtcl{qaox|S*2X_!3cVk4
z@&081mWTxG{nr!QU6@`tGY*5mfj=*oy&mgDrpJf7iegWHR>W7U?y^%leVcccIPbh+
z#slkg%P$3fMcEZwP5^6*lQAe&8Bklz5rjw=TlCm?3-FKq9rSuMGIS*wdX+d(EILNx
zta2>jG&m|_YvX3>zWk~rwmm@&O(XAAdqNN68%SZ?20pqZDl%+?vdD!l*QOpb1S_&1
zip-FrQu?D>g+?Q^8k_M~rjV&L(=8rr!j*quhhO5|qRdy2b6q5L82)Ui56@M75xKX)
zSj&Fks4)F<?m9U(ZuhCS_D<sjvz#;3k-_%;kTAogA~(8hTkCl4Ar(j@zC{-MI7y?u
zKlES_nNwbW?hc7d2<Yzw<Yl@&HyE8E(4*<CXTG$ad+buZ4feOa)gyh8u}FzNW4zf0
zDmUk(qgqj*T|bV(%D{`$NVu{`tXSR435ORc&i*CI7tMA>&_g9d&pnCwF`$Scno~Y1
zLHIQ8*}o`n6i$s_or<&6D1YDOY0{_HjA6o9XxY3+k^36~UPCSO&&zhdVEHu*#bM2s
zqmbd#IDXHG&ndFMU!z}F98zPy5subr8ikKOrh|#xMLUNilcPJn{MN!6Xs-=EBqvP?
zjbaLPwkYl~XzmPQM{-XTu><(w+t#ApNVZ7(^THOU_NdWGt@`h7*i|1k<K*n#u`>2{
zuvHjTtO@COr-9oNasde~Bpr|pG|RgxGsT}SQcGh=d60>9@Z~i}b7Fsw<yj6{^Tr&*
zlw^(B6L)?nMimun)ZUK=$sL)I4z1okr`QJXoj8wfeUYa)XY@k5LRCVUok(8cvuiz3
zSerGwJ=D=FLCckI*;e+4Hiia~wU*0G?%0%(j~?=?N~XMgDoR>UfPibc&vNiC(Q;{N
zz=0PYeP-nucv_t_j8M&+VWgW=KACZXnIHy}dt!P@vXz1!SOeukn=6lW04B^mJ{L&!
zA>H!bBA|Rb$Gq|CR!)$#MlAL7X|OcGp~P|W%d+js>W^Bou)uq(&i-+7dh2|q7oZ4_
z#3!dmt@bfSqTVf8p{qXtht*MzXs9<(3-0NghhBk5C!{Hh-f*vd2LXFOcv}Wg6NSmy
zj$;i1t%3yJEfy~VPlD^sn74juy8xBVYCrcw+z~nUHEWQ0!*mgOB|;a#^mdG``NBf2
zr(-dksbDbWwxeGW9~HY|Zj(HdMm&Zm$&EAPPY>cb;JF@#9RjBsh8F}}Ua)(}F20_2
zoHngr0(cAs{}9-cX*a>bg~iGkJl8uG{27wfuT>yho|6)#8s}QPYHzUb6B-g4*5E0#
zu{HB<jv!THh3ICM7$pl&Q3?jp-@9)5AF3nf<<2$bMpYxLl=oo7VYYrkLs80iY2MJD
zv2V3G+U^FfJO$Jy)V~B0mWDixg?fPPZX)fKQ|E3KFl~=0K2YQF$Nn1O+gC45u9htE
ztu(g5K7Bv>wnM&YAaAyq?!cmFG?n4>L7Y3(AWCqu<(Jqz?tsO)mlv~JeUnAWE+Z=J
z38+drXdlvkS0;Mq|Fc-a18T(|+A`LIXGH$j_u%V$C<ZNL2=eEda>^hdUTD8R8+1tc
zN}JaEM``Dg70*Hy(=gZPx!hnvy36m2R{zKP`^UMJ2W|13zvX!K0(3sr?}y{-ZIS-^
zS`SXH1gX56qg*_!eP@l+bR}eDn+b!xz)m{wX@i5InyvQ6cTOCYNw8^gpFXsld~qSD
zDCG0bM2jw=ADh9uY{BkCf0OxlqNyPC+w5EKJg{>ma7GI#n%9b*_oj0(SYcL<=66m3
zesXnNsduN>@*XLI?RSTbTwB&G@u3{3lvQSt+}Ct!98}83zaX{`mh2&^Qutk6zlxbD
z;wU-1cyZt=n29*mQU^2MQVp~=)t^OU4|%fp(|~@;a3{~*S!4QsnN?nwi6(REOmTQ$
zN|iz;27=TIl8z(dO5yBxo=>hLgdOg7EMd5klGwY`54X#|oFf6A|1|_+`;Q3df2(MJ
zO|;4`wgmKY1dI&7)>$!2C+A<i>~8?{KZ<0T8JIc#yNKvoThn=y1Icf-uFu%43h02Z
zRY^H@Jf>-VgBW#o2@jKB0f}}oQ36$R^Zxt&3#-_|8);oi#{R?*Uj_+f?d#9{8FamA
zkZ37sa?o2}@rba=0s|>(avo(cd*C_2eHdo9N>pY9K1~zVREn{jRw<G52qP?!i}IBB
zY_pkQaJymuA!9wpT%(K>d8V5|Sbnql0u}$F!A+>atALU_aR!osQG_GRR6vtk5MX@c
zo)y46bT|0J-dub;GTUA1Z-lO44ZsLs5A+_~0?ZGM#z$dx?+xM{O-Hl`mj*>p4H#XX
z2`ULq21ldQ%uIxsW3Cqj^6r^{(9eJ`7&Q``?T*1f0Td>}9%?^7Xce*eC%Y8`o5aG_
zg;^5l5h2%gN(`YskCVs(*clQ+M9OG1BR3R<fccaN3o4_$Mvz5o3^zk2M-WU8y4xiK
z7$JD7yQl*M{`Q_tEI^f|KL`|XOGyF^1%xcLFcZn6k4H1hDNxdRe)iY9#0deLJR%hH
zh$<l&p$f$?AEW}VY;9A=HNWo;nHtFli++JBnVLd4f{~x{SX`Y8h=Boxgl0IGK_vO;
zN&@w0V$q6>1ylx6Fq}hqVHcT3bQ8CfZbFBnlEH_j7ua2l&r2r7_0Htw6MzXqHf4_m
z*ejm>Dg~RHV3~uKl=?{CBZN-@6(+3DVMIY-PC>)C<oA34IvJ<=-1*&MwjO__sZG<_
zSo_^~?0)reYd-6@yt6<1LlweNIti-Fc1T(#Eu4ONYxYr5!MoI%DCR{a)_Gom<^QUR
zA^4I-SJt-sX?vARH~FD(Ca?GzOlSJ;R6M^`>hF_oE9C6zel&dj%|CC-=(j{RA0yu;
z=83QHKshJ!a;S?nO;HSD=8HW_Pdxj$N#*N;FaJR~n|`Rl7n5_L2ylu0Cmx57-Tm5E
zXS+-%UW~O$?BIA!$CrsS{A><g$b;R_&r5xcodn(Zf%(u|o=>>morar=+fC@PVsv(O
zGsx8b0kyX7qLR9@)>QA2cNf*xcb+CHO~%&iVbw_UgntQNFz$WLqJs$9mLwdXi%<$y
zqL1_GOJ7YDHt0@y31s%=sQL`EmumYxuJg9Qo2~Q|J~+p&^6&ZgO8mg1ij4g5qsqXY
zp#0S1kd6G*^WXKnjO)n%|IzW?HMhvqL7LA3llG;+K3V7~4>BzITy`b<3<bN^h@7-s
z7_3sgm%+k;7R&wSH%<$-m$Hv#%$sSbM0bK)Zst!}1;38SygrFekCcC9n)Q9f@o~J3
z{qF6y;=1`9ui08Ht;b#^166BaW$UKbMjKAq_RJ0ObYtQNs4<M%odxSMN+U&AfjJ*i
z^DR|f)yUnI@gUkWeqpPaD;!+uvrRTOn|;toQ@c@Zvl(4DU5XVD-JU#nlt&aW#`?K*
zc<*lhL}uS+3W~FS+2%jPF9*27Rr|*uforSM)uIFP5h$=|>RVamsquskWJvlz(Eu1;
zqeV?Due+&1$17@(s7&C<`aaLjLz5pY{T~9Jn&uH=nt-+)?r<((*2l6^b+%}Sj8n)t
z%Itkvo_NiG^JR6G!3ZoRNm_cS7STUg#AMb`uiDm2AZSad5W3<Ld}1_&AlF-hRgQ5e
zstYpv*@<Ig1uhSCkkm2O_JTV_4ZH%-4QsqFb?U=H%73-pGy%@a>b6P&t{oJWXla&y
zxtMf~J>}qjG>j&Zho@jf%7<d!hUFCHZ5n!1MA}e!+997SgY|>yxhC|OAGRuUC}p#R
zgVUm$XsJj8HBLNfo}z*SA$tjhjz^U%eMBX!U^Zj2ZW=E}n@1VPrL`t5UYjY)g&}&a
z$id1JPVq~NRvb#j`FC|_Z%c4jehyCugQC=b$llzd*QX^hOVM)Xw+x=aY*wtQiJll*
zN!9EXSzX;rLsB-KddkVg$Z*x52C#}}w+J}X055Zrj_}UeMa&QoK*dWym{`YyyiE?t
z!0e8BnYd^9ft1c<r<O8;P7*K{fgC|laS^68L7cU?&USbYvg#g&e^o8x6;FIb0U~&{
z2q~azeAlQr1=x%?!zr^5rjUx9A&<7(IklHnk$)^R424;%2|V<5g+!3WieD(y8@35b
zgPK7?*joBRLJVMp1)^W2fZ>QT3=F-1<Nsy%Gf6ptezXLeCU9yN%fJtKy?cgipM+H=
zW8Ew#Ra)XD@K6r<0L`3hOTv)x$rcx=LMq1-tQJp@3MpdZ!(RsU4SEdHzF7}fc$tpR
zDZeR@Qonylow)32iBMp<CmN9$6erPS$?-8RZ7zrUe_!qfZ!*$Y1Tt(lTp0EAQgZri
zY?kcg6?pp_*n=0v{1x_AW&mFF)3667`Dd^Zwldw{Y&c8Gl+K-7@z_|KeQ4WDp`o8S
zwsWGZ&VqV*iMlS2-?cibyMtO?XKaAEteXD#12M7!ua8@@9kh=bb$NZmt6CdwF0KvP
zWiG9a^2MxL6h1MVs4k)4KC9A)2gQLgwR?J6>^Ss34Hn%W&);KvG3+Qfk<s&Y4?(5l
zb5*X~OQpxF;FP5F(|RJ(ax#jNYY`czO1=rd^Q>#bZ?dT#p~G*n+T{FIuNy!5G~gl}
zT3&|qF>8rV!0OG^3_iDe?4UlV3-8Xnb$!=?z0uW$e*^5vIG4)p2JCD4j9T8yCMBX)
zXn85@v_Y<*(F>M!Xz0m`GvJ%ZFZ5G|b~9erDvxMH5%T5+Vn&iv9t97qeLUSrm4!lS
zSI15AYOd@}Zf>A&Zz8r6N%qgsV$-^hNqG=ZdjrdP=E2W=rlffKEhFbdk3;?VT{yC@
zuyD;aPaoY@RmU~COrqFm11Mf}t!RHJP+7sduWL6m-wr}kU2V1Mdg%|54w<bsSNjvn
zsWm)%nRtS?;PM0C8oGQ^m_jAB<%7D|tDTfM$N|GtP1;t4w}I`flT1-ev{zHjrcW<j
z9z`yR{RLwr+P`Q&kWTb9OrcfTfztePw?C?Tbe>@H*;5vf*KQmyr?#%j`f0i!f4a#?
zgE@mob9wOlrHXkSNzx^DQGzs3fA4R`$`hA;SC-6W=U8@SwVLwel(5+x_2s@eY;l;!
zE|7A(3w<LR*<T>Kl9OpsNn~zDIx4w#3uRdW41Z<$`DVB!K~W~kJc;>{D5z4^zCJ1l
z`Y`PJU6h;^hpmCQy_l$ol%v5tx`zvz+kdXPX=kn1baL|K^hy0167aJiqmeC^^DBz6
zzh4OM=}9xW5{iGPQr9AFqScxLdDlm(u+xZhnF4RfA`}Qt0+{agw}GB|`V@)sc;+F+
zpWHlX6_6s&40+k68>kO&qN}pDs|K_<Ii-JSL&`J5kP{?UDH)W=e1z#sXi;I7B7QN+
ze^Bs*+DFr9o+44$zMHbWJT(kFL>23Ro0zTOfw-ralik$F3@;*J(-{4%jUrTSAfC*1
zyC5l0d4uIv9z@`PmNY<()>wMUJS%?NQrf@Za{k1CSL*YFxJmQz)gd~8vh6C0=nIJI
z?A{rP!BJk^ljrVXh_}-{-ekW0MVi8V?W!5yS1+tBMzjO4P1yrHkoL;tu`<x+c@J9z
zV({%`lv=z1Il%J|&|wOw7}xO|DC)?ri+}mh>-x=D4rgu|Z*y1JIL!OpO%_oh{aLN?
z1|V-ex9#s5xVhji;`SBr;KZF0-BjT&>kTN^aQATfM#BZ7^Xq}L9icK)rSn-8`5$uB
zxj)|?T|id&?pJbgV`jID`&%k}I5WfJ`vy5URJl$Zgyi8#WjJvYl3gFZ9{fu}CEMHK
z|Be;3Q-vK>J4bVhl+)Z0QJvw?<`)y)isxN?Hi?H7^<7zxZ^l!$>E?@Qj5d;-rYxsz
zD?2^c=$a{$*_3ha*RN^#_0iI=*Cy~>kHFO?6a@!F@a&VUxr~jk505)5FPA&VX7JtB
z<QDRna4%Q!p0`T_=Ac*uzH<YLAG4sG)cW0|T&olaJze~Q&i@v|xbx1wo7f=*CQXzw
z;0J}&6ao8zQGdLIs1idZ-HC$T2uQxaB>lMtE-dB{tR@6j6L5=DI8^GZPZ5Ii_<>%T
zr$_s*)dV94!++Eh{}VaD#KFPz-_;AtTGH_(ZLqzwb!Ytbjd)8%0V4pq%!`6P&W-GY
z<ly7Oz{DL#x8lVjkxHuWuJkJmL88|#s^@}=pJVYV=gBia-w*WfzdhxT$><#(#>c@N
z86~U%O0k-1-mmw!f4Ar@G|&J&=)Z8}eW2ZEq0!zb|L4<U&W^wT8|6<I?|eGfVmjaQ
z`_plF6=azxZ9f=??^hYZWO?x-C0AA9eOX<8S^lh6_=}Hb4Pf2pW=)-bj`J&7583DC
z1-8xcbd~I@-0A5Xzr>g3(98K#dL6$RX5nsfKUhstLlexNLkd+iT~j7IPOfQx8{I!#
zoeA%*lQMO}E?Zio#LaG3&xT>Q)Nd`VgNqY9)6dRA-<qKDk2HN*>c(65CxDgH9tUUo
zc&7G`hmHcFwdco<)1(&yeZINRZoVqm3#!*vZY|o#SR#<re=zd_QRA(!9doUI4U-)f
zfI@93IGc2dFSO(2gs8YpV*L|*#jBTFqkDBBe;aYraix6AEK(Gny>qs!n+rG4Sp(jK
z<G0`x;zg0yfO6^@JFFX$7Qjaz73lJi)uo+jkMU^XB?btt>L1$%^o%@6ZO2oSVcUA^
zSK6c#i>P0tblWqdk<30*rWaLRA%sV0S*12HPd6$}j0fKq)Ng^LvHo-u@@x_j-!NB?
zYJ}OmIdaQCVQo{V`~|MP2+}PE)nAK0VwWSx-#M`@`P%;y7{9ZZ>H27n&aK5kmkpa0
zI|FX-%t(32POS}ZB2_?RYig)C2rY%`{ML(pV*?mH*2!brJ8?|qKc=K0h|q?C_B9G9
z$YFjMYX528lwPLX;J#o;nbfPov+bvNGii=#=@us(A1waHrq>aH7)Btco>HzrkzK^R
zab>frQ_NY)Rr4M@Oce65T+%C~8xX(Nbamo}Q{P1^RE^fodUj=G4loixTvOeMhk^nV
zRt*FfAvD%rm5A^IEwCep5=*Mv2FKqAX9c+dyuTR+HgAU7V=pR@SzV1?S}1CH75LDc
zvLXHo9A^zMnV%AiBK}|OeF;2O+xK{bQic$j<04ar>ze1e%tK}}ba8PrT<*Q*DO0A*
zA(0^(ND&!Irphd1B`F~a86!hV|8uV)yk4)@_xJ7fzW?6m^K#EQd#|<jT6?X%*4pcw
z?TI?jG?v^pV-U%18;>5hhx4h9CyqAC&eO)VxTwmNDi!aW)TO(`-g4kr*jsMlSwX$G
zeN{vDSM878ls|H#<U@7xSlxD_{kWPF+eJ=2m{Gi9M7$Si8rBk87yY_!d-Tqi=8ErB
z^kIdtdoBl4(TU1-r}hUD1CBF=tG`S8yR)<x%?XcP()c(G31S@KIb3&KtL4338(yCd
z-FeplrBo9!#_p8v#kdgQdz2|~Z{1u-dNhx}5l)X>unFQlOP&6ki&ce@WV}%>uUXx!
z-};mgk496@PVY0`an3Hh#lEOb&N}yP$kn-;Jsk29ii<RUM){X+yn66ZcvtL0q@C+7
z+RU($BWB)YQ+Hc?D50iR+7($z{!QS7DTN2z1fums?$E*8Nfx6H)mLft9t1RYe@Ho4
zeDI7mOPpIj%B^Iq=(rohw92I$V{!`(w(s6RV6rr<O)5i&ueT$sT0gZbCrtsTZy%5D
zqWqBl5}AI);{eXZl!G3=WJEd>uZfi!Eszr?%zm%0vr_Q|-FPL<(7j88Q3?u`ZIUpe
zV8X3pW&|}5ax)4mXJxv_+&;*9>!Cr>F~QFg(sqQmNzFU=nX5E?a?0t?dvP^@qPmQl
zgL;80>{)$)e*Oc)4{u~Eiw<+n5c{$irD--x8ok2NPfICim)hNx?`P|UcLTb!%WY3r
zsh{nM(u7{uernv((+5jqkJ1>2h9pA}mPtJE{h$|c{#Mx&#X)?Jk|={M33N%avS2LN
zZeDJad)AyKtw$itOimo+_D*C5b-k*3FlRWXzu0-l_`?iO<QtOaLsoZ2Quj%O^QFfo
zKiI7;NG(K$uRnL)9nX^Ga;Vs2;!$lrkF!qG{wYq|+sFiNX(h)a#-0&|euyV`4cB*v
zuy^`99n&T0&Z87!Y84yO(n!fG-}fLMFI$hLw`Fcyt!v93uczcKDK~}(o)J;#`;Wh`
z>8<mwhggqtA+My+j4F)Pa??CZ8fWRnw>=Ua>M!<8*I<(DXf}O-HnLg`dh5`hiV%&7
zNP7pVhWTiQCqidaDoEUSy9;&EJ^HviQrj|){`H^>of+A?legzTmz)|)$U+(k<xO8b
zl`8`y7nwhtHJg5duu-9skKSUbgYeFQs2aY+-Symjme-q7%-!h<oKD$2l}_M$^Nxc$
zrj>bnPElg$8;Aw|buZTK&xh|hjnoE>RbEW)`rsDrMmrWDhB)Cmw`Z&-B&ynnDZ{pB
z)Lw`DA?lf4P7VsP<Gz;S(v51$q3uReBn(R)$G1;(2tk&@{V26>nK2{a(xz!yxq`0+
z?B$nI&&KZq9X5Giq`2$i(<86mIyJ#l!$GG0cLc@5mfnrIWT>?AlanWU9|hM?&&34g
zG?;%-guh2W(oW^s@q$cP1;5jVrz8<qYDkHw>c-=W9KQ<J7n*$`Z=5blDg_7;@+(+w
zAEu3g-5yLD&^)-*FrZ0TnoynBdJ*)f<<zt6nlrTZz5zsaZ*GaRm~Ss4;U-FXVTpu~
ziwC=wn>Vv;Q@>c4NFt`@su$ag$U42xzrcs7y}kNO$zj$x8k2)r<&un_o~zeP=$&;=
z1*%yGXCci@zKGV2NQLx@V0e=>ZPL^D&WKElxSN8JDLDSx?ROS!S|1N^v*yF>4L$ef
zn1M`Kz~0+lxcBcYh<?8(9vL$~QlIFbOsfqXGb8bEpL*7niDwaQ`hwfv=a<}iB5=*0
zsp3Gr<<X$F24?FRdM}@t8sjl{w&^p>^W?kET&+8u`RrtZNL>HJ%d@aCm6?zI86_%K
zAx%~{%k=I#nQ*I&MucTN_OBSag==Y_rFC^__T;-Up;(95p*CSUf}Gx=7mnOlMn2h|
z=SXwW2YZkE#Rps$;`NE}d#n>JZF2mTv%3U^1g>FO5B<C;Z-onz&STf<opzZo#bi<y
zqTY+8;TP6L5QsPst15_mNZiKM@HEXc3%9s^r*=^z!KsIiip^uv9C}tbE>R@LyTd9`
zyLSnGR@?u|((o~|{lk<cfeN_SX8M3)u9;sJjhSC*pXFYThs>N*Pj`+T$Q~YnCdgc`
zux7{8u07W|bqwN)ny|5KQF8_Ub}&k6CWjaeJ{M^m*J8b(4;5pp<l-w4ef_bjC3Boo
zeQuY-UXggAlZGA<$1Y!m5+GF34JQNQ8CsU8LjBwyAHlB)M4FttjJwavLQ0k-g6a%g
zq4LLFtjAt`s`iU0nX4al&NHYZAIov~D1FbRv;U>62>slCv(e{LJ=DA-db&G$;$F#W
zB4i{;8RJ>uWR4$F-?GGdi{cla8K8t)-Ypv1UGLMciepiYTY@T*X53yy7nd}bIrQp8
zzLz-H)a%<iQ$(qU5)-0QREkt+D~T;k5M5VsdJO$}8Cvk&&Ntbl?|8@Gj5_1hRaY8)
zcEU_`nFQB>$VO9Y%H&<0{>SFhc$)amg^b98q->1;HUnN<lzR8xLN6m|Xx8lHqWinr
z?B>CDlP}c)2QJ&oXzT@h4)aM)rZyiqCGMOlmD7~(<kjX9&?F(fO#j}~*$$@!x6jO7
zjcUodNFH-uDo3Ws{Os-z&)RdvhoIZuyIdvBSA45ea5x~SOV&)DA-%M`OZK&@FOgr@
zv#=ehO5}H54j;-@P%BKhIYDzj>vdViQ=*|s<P{Y?t&?Cia#nvFqQ?iRLQ00;6$qv8
zz`b$H$D~8=S?s5tIa$kl+?~oE6_sUP`7y!^PuZ)c?OG1+L@5+x8|B#2n<UtqjK0^Q
zl2)hmW+9q+S7F9{VoK?NXP>U(z|5lw8FLf7OL}d5?V!6&rGuFWj-q2M548nfy<pX$
zyBOv)M<OJZ(%|#fH`|5?^fjMcXoF9_lH*r{p`n&gG<3sxa|UXvp!H+n_=E+7`PSX@
zwPWGLrQCqi<!oVSkU7j6?jXT9Syssif?G;3>I<mxs5#2PY~YF>NSL;Vx{ig1t%Zms
zqm(3`xSOb(y`w#V5#(lX=YSG*lVF4b=itFaf!~<hU`9zi5J((pX(g&9e{{7Ga3#TL
zgGM`wg2Aq?uH3GC+z6yK7$PDf0_Ncb^YU^5Ex1tb4rr(wmjj9k3u6_EJPc)lggc_)
z2nP@bD%2d|jFw<z1h9PNzZIe!H{f$XabpSOwm{f}fd^m+Hy`+$y@3I0VGid*t<$^-
zh8xrofWZSsV%vNhl%>Tw3`b|A9d^`~7GRhi%pN%B4+X%1d>eqH4FZio*&rO(;cTL1
z{V|rJzZ%uoVXTekuX?MgeY?NC{kMmLLLYGf1h+a3HVtEq1QcdMgSB8Ngfr3tcEknX
z5z{xx0^|wBOxX={3krwZWBH7E26jM!;g;W~3ux#3Eg;U{3k6VxuMG}>@AlV|4%2GO
zNE{qtsI`%}eb-2kaFqQz5ssKE6dLMa2}N3B;W$9;VJOUzk|?k_6a}{c^8wTNoBTZq
zby<gpQCtT&3V{WMvao^KLs4L$!FNqP9Zl<XKwlL-8tiEOZCX(1qYmFSe~u#lYT76{
zqAmCVHJ3vmVe4c8HpXIQY-mgH#unHCVhzi;3B*jeBW`Gz0}8NJ>tHc#`@@`stu^~=
zh}u|Ae;eG2H2zvuG?55L7!vLNZD@|Gt^OJwMqR%R3k|gRYdA;YFuT9dVk=COwS@s#
z8a6RobagsnmY$XWunUkZ)cVVv)#bzvj$K6@p-2>Lh5R*hv%2W4L%?8IVg4$3Ng2#S
z2hs$VwUztgnDt~!n{^5>aIjRa!f-^wtl(~vK#xt`R&W1lw{@M>)l!oD6Z3+#E}Nik
z>h-6v*95(hNm#Nm>#R1+Y3u#J%fb~_Vwtcu^F%R}N$Mb>j*c+P6_PhU+C=Qv0JuQy
zoUxLV)B&_;3y3cd{}PS`aDJ{e0_iRZcK{C5b%5G!0k!`8FCkfe@lYg{94z5LPmHqy
zw*sMUU?59`g)?Tc2iZVTfU5;_0NEof;lOmZ1fdZixPyhAGjN128p9E|gEer&3W2o8
zLfAskMl^q*NgRxknH_wq$X4$EL6$3^*JtL6RoK|~Yw)l-^`GdEftn^r4Q2_4g4Cc6
zz=6nEw*0s%0R~)6ZOcfoI<|>B8+CA#T6|5J)*KPQm)Qh(Q^!9A|BvaD{OTN2lHW3C
zu-dkX-fycEU}UyH_{TL$S64|M1h^}JzQ8R3J;JCC5(Fp-2x?~sa)BW+t1SuyxH%vs
za3nVzqkAY2!U_akSG_Gm`XNR7KeK&G4j7-z0`~uxtWb_H3qW!f7=s4|tS#0;{<=ca
zTbWk>v5bDjC2GxBtm)1MwfdzLMQcr)<?Ns+)E3hIu3`Houz%df=>UF$J=DPgFl=xK
zkgE;c!e)c{0>M#O(*{HS!Dad(Q}#>w1b{aUmH{cPTJ}}f>BoHnj6)7zbu8E2o4@ZA
zd^KnT5(z7_a24<`?i2ta&Q|yM?`n+_;2K-QkXv-`A6FMmU}TsTer4{gr8u@w`9rGl
zf40h7AXfZ%V4>ea*WZ=FuegNBeHk6d4(0;0`|n8O*D_REO4@25Sxnvx2x3+;?wia3
z3Y)0e7({(N0f0%%ZR+?VDS+=x?rp;Ji`y$8KksM<$GFBoM!*Vg_rIt(n%GIS<_xcR
z!C%G=0s=_Pz%hw|6(;~~1FZe)83d@M4d4T8b%p*}L+9vhZnv88PzO>wzlO2Hq*wpX
zQaYMnMhL>F(3T<ov+RdjTO(oC*olP6W^SSFA6jevYwX8lleY}<pJl%}0%3b(CFih(
zuKxwruZ|3}xMMPzn8ALDPd4Q?U4h&t))Js>{%>NG|1(F-1z`uIU;hV%v(YTA3=-tv
zY;XPtZ~dQDkgt;Tzn~i%*$+pdoMEfn2f4z5McV-g-@iKZ{#j`_0!eo*sDt(Yq&@&+
z10gYk!zPC}o2M0-VD15qGq4x4nl{E*HaL3KbgkqbwpfaPR$jowa@a!9|AOu)X=|&m
z3kfR_Ai0lSKGwF*){<bGrTEJ(GBBn84+I4iEm@$E2nQhk{Wr@Bjxzt>)G65wI<+BW
zSx+ecOLWN+0skK~dGc^;I2vjP0=~xpXwiDUdvzysqiD7a{-0fdfSScE^!(lRM^{UA
z3xpp@AOD|5lmJN}D0)4^vY|x70_p&iXaGx-v%}Wnlz&zZ=I&^i96;6oXhL7c5y*VG
z!GW?DfccmN7z*wI!-Sd$3p5P<izU1OKl6WB6zbR*65|+Sf&~;t2w!!Ze_4}-ayCc9
z(Z5Re`2W;!t?k^iqCkH#${(^)|99IgE08vnU9On^BWu;JTkAA2wOOJ-w%^j(0)~`S
zb%rX+>c9**L968#AfP;+kDC|72HXUx!d+k>ULIZuI~M4dZmZ=@SXe4BcUJ_`5+%9P
zX8m$w8(`OP6}}{dTVNH~mwRhX*WUsVRf8gJVU|GAqB9UT|Kau@+v~&3wb6hn!<4tJ
zm26rdVL+1B1qLMh?Sc9vY@IQXHwS8kF~!*KV1A$ydn52oFxF65T8OR`2TB6Pf@@iK
zObx0y_{*aWjj?+0WkU*Ezb9jd{&MF_mJKLx7PSL{X9-4_1DCEgqojhSGSp2JBm+`_
zI-^j)whRc66Hqyh0v$yHNpOJ6*j%8pjKPY~H<li*+}VH$OW!7H!~uf8I$8e~k>88@
zE0{0@th~YTg&ltjw;aOJ9SZ=&X2Fh`n7o`I9UBBt6c5sbJ6OVyoFI9q15mXLl1Eqq
z-o)1W%&#+PGf5aCv9T7ARz+^1bAy(x*X#m9&~OAQOqGH6CxD7pVB!2a)(w5JqWJ?3
ztayMayanf$I?~mGP)W?-xp;)Qcm;KMc|`g6M0t3P{tWd75Lnc}i%aWRFjz4u{w-KH
zP`h!YD;wkAH^7ZhF<7wbh}g=~4YX}Sv{9;m9M`%MNq&dCH=w`{e+wZP+E!kE0m_mc
zozdKAm>ag;>dg(xjnP(TV2S=dW?{AZkF>Ac*}z~dvaO&oRDG$v1SqpYhyr067{bHN
z3xV(mfKMV&C~h8pegVLoZf?6qFkm)C<uEo7CXW&L*JsR@MjQL8VQQw`|Dx}PMw|Ms
z?$-ZR?~Tni_Q%qVK<Xe6c9N=4%vyxW@q%oCnTiC8`mttvWpJyF);g~gBZ*=PaU}Ux
zmO{AY%H@_O8_Go_W!GD7yt~!`;E|e|=$a>>Broag48Y_UwBWaZ@k6+P4{-=`@e5fA
zaY0uug?Y^dg{&+sL;%tJq3>p(Un)~qfI=)SU{;n8OD+q3L0&F?b8|s1s5yj>3kHQj
zg@pl-=7JV$KsWT=40OGcQWCP(_RFo!O*hx*Nm|0Jpw4zc#Ir*7mZxi7v4s$#U-r;}
zRSFY9NTO|!Fie^oQ*H5;$D6xv$Y-eoiyP*x#m%eXszJrL#aIKYjeS84q>ZpbyF!uc
zc(%Y<Q%&&|Dz@y^N}gP?(rZL+eY}}OOe$ECmtRPLON2*IaIFva?&gMoVoH8hF;|Gd
zJl@=4J+=h2QW(O=vj+Ri{mtz$aRD{}lH^}&`sLPIQ$S|GD_^3RSFuoQ_*M<&25Ymy
z%6{)#IoP+r05`}1eH3_+O;S}&3s`a;AUs@JFl%Q!C{U}mrNt^m7*qfO8z0)k?c4zX
zAS?vn`<R=%VxRt5XUr`5Hg0T-uXT4>DJ|AU0~-<6lDynJJp5vs+Mtymz}udH(z^hT
zFAq1rh?qUp7KT7uaofSof#=*1eqjN?j;xGfGn}ts!ZHca1uQ9c_Q+L60rdhHMsWjx
z*I;dG`DdUo^#zi^n<5S?yYOE)iG8r4*B_Ue^(FF$rK}BgK>bHT#sK&s8UH!vUr)rS
z6)ydoB`k}u1DxeAlMAcm8z$GE=<t6k-mPeVS=pKc{P{Pjt|IV_0niWHW=!ABbK$Rz
zGN#Yh)`IU8DDXx!51-g^J_v;03c@QOXu^dA_Pv0&kHkP|G~kb*5y0j?8pP%T;TB>C
zaoK?cL0m|X01t>u0f+&CKW>1L0dXmUAi@w3mpsT0igo~TS*`xH2QhA>k4r~}8|{W>
z1aY~76hK`1ATBhB%K*g4xHjm&Y2!D-<_5@J1M~gn9S!6-e!h9f(EWRk{GKB#KHAE^
zn>;nZ(fd6|e%le))>rhmxCz)ff?3HyO2A$#90*RgH24cY0@HtUpYL36FgX8+@@FqO
zyn+IMzT|u%@xQX_d;$NjTX-;}Z<ab>ZvLAyWh1hmKSu!X<j>p^3@83;v*TYkFaD^|
z-xG`cunqYmx}QBaApGC3A1mbjS7yfw=zrb(_#M=&1UJ8f8m!0qRfqNW+Ojq~ey=U7
z8EdSQ`bT5`du{o>wqWPL?;z&e4(fl=5y9A~f1fj=s)pU75c*oy^LwlcumiS|Jr~*-
zm>h-vb?o~s%-`v}{>1$+?B<9Nkk9?MZh!Iq#}elKK?#2q^Dio4;9J7~S_$)T{fNW3
zuE0OAgN-#Ee?<F>%9e-A95B)UrtPn_`h>4)(05wr)%t`VP~Mf=h_B-Q^(4GfpYShL
zUo5AE{{sWPB4pveZ=nAi^Dio5;eV%r#`0P4KbA01y0vjr;(MH|KgaxwN?7n;CSfca
zE$r>#kWDsrD|i1->zdYaZc_Q5gL2ax{23@w?whOYeh$u`&RLZEPv<Pm?dNkA=JwM$
z3v=5Poc{Ock39-%|Fb#!PvQLOtTq1!So_c5{K>4fut)y<B@Fq~tpW1qnOSUzwwmYq
zW0?G%0o+P@?RRv(p}uRwZrMLxSN@-bZCKH6Qh@JIoPHJW)}?K|-g$E^=ugg@{+nYq
zf&VyW174l_p_uKfn14M1BQY<A2>e@PwoSs>%H01($cA$N=_vmZ_xC4{(jH}Q|C3Py
z?$Xc0_}i1%bQstAy0+=%1Un!G!oKMB@5}RJJOPpaV5z}+5+Z+prTJ65TW#Sw$Ny1J
zLgZhz&aA2}HV6Bkh!+b0^aJw!bIiY<fU%y$zfHneHe#~Bn}oBKx&Iwc;-{m8x&366
z|4dTr=VyBWU*hL+cB3!x)!p5{m6+JH>GA#HE99$~x9ZKWvFZS{{;mqadRpIK*7e79
z*DosDe_vGyw#M`K`vbq<ANYDkbc1;p=F$MpS8>98d2VA0>aX8WTPq6t;oRg`@cx?m
zFU*BlVbQ;<E({a5{g&>k0&Lh0IQr#_H7G2L{<YEtzE|@VCHh;g*^loTeFgKcslEW0
zeoNQ^z4}S3@2LmFSdOpHt=|&f_~Q2J=P7^qn^V}#%YX7(#Y)a+Qw_lPg;#%$`PY*$
zHs|wid#wV?Mr_63CJVcjyZ?=(k3D?Tg!maK|C#c;pFYfgrWWoe5A&bNr$Oz1`Y?Zf
z3J<db`ICo<DcJhy76|z_2f1PeH+;4e`$^V6eBM)h^_!pSnwWzFFdqlqd;|MQl@(a>
zFq8!n?udpX954&^%1tf#qmp<)CsClAJ@75wjZ684IQjd3Zu*AH_jg-w`Mdu*D}Lz5
zZ~pxd+txe%=HEIie(1+<{{0Z!);s;?-#ROP=*MsV{Se#MJK>2-xj6!#<b$C>Fb7Ks
zMps5DNjyB@CkhRmBw&yEg)l#f2LbVbFn`6xB_-GH@qXnV-}*fe2<&KW<%qO42OS4%
zLeZGR!a-npm<w<qi<W|{3Fr$Th_LW#x8opQA<%J<pr8<lANaPb35ZVsxD<d00YCY9
zF;DnJK>YmtKr>;?EkWRx5HIgq3lU6X5uh*7LIC(stO-Z})1Qw=5PONag#o?tNDu>0
zh({RPnU9YT3zHWEUx*jP3p6tU2|+Lb0mPWwz$HI$DJ&!a5)$EG@3{g@m<I!lZv!5n
z2ZnAwVG->ACLmytnBfV7gs`-WU?>v=M!*Na7Z3pc3iAOtAOMX3PT=1azz`k`QbB+j
ze1eem`+@+l6|O)am|^e(&w##=m9~5<3|zg3LA&~#7XzCg;OQ!CUMzP3s!c#9Yf^=X
z@O+UWSQ|L#3HT&A2rQ>0e-!v0Jf@DEM-BuA63Bo+fxp1^5=<UAamNy9qY87d2F`5Y
z7ZDZ$q`NxB@Kz_1+cRf_y~d<nPtu63LiSvJ(AKBR%6<K~;iG4FwjWOB+;MTw0gB>$
zx#)w3{Qdm{FCDIDtiM?5x96g4ydF+<z4;NcaK@9DiuJBnEE~0EQ7m~2wN-S@drt;g
z-fO<#*(!lre&f5$Dz2;@H!$sHo=8aN68V-~_(o6PLAs*0NA=yT{5A?_bR?`=ANE-t
zMqfs0vpY~nCEuvN>7{-#DL2zks-s_M*e24gKUsSu@AAosj4Q{;ChKvKkIwP#oprW~
z{Y>zBxLMDvt!puQzm2B2vZ`99=`kj8wNydGj>&+}=I${!vqxAH3PJ+w(@#~|G%#v+
z*3e&zYC2?lhIy{Hf}x=FYBVU+wKVNPuR%eJ!N8$#CjEdzC|Lh~dU(3))k^u-;BL0}
z(GTR6m@*G-k1048a@D;yHo{NlSgf$2Wk{fxg0kGD-eX!)N~)!4{95}Dvc*5WWB-)x
zW=xQR3T=SbrT$}Dp)oP?`<dgFcVuQVB_yAdufw5iV%(|vu|eV(ONNEek)*3{sJey4
z?D>(m3U9eQJ7G_Ike+Kg@rd%rtoB}+jAON*yXq=iE?CAudo!+mvXs;klY+aUHPJhc
zn1-BJet7Y<HvZKpc8kV9BSo%@<l1S+RI*`@kSBM+)D1PSU48tBx-qukqFr{;S&1QO
z5v{jutc^iN8n?A*u11`tDGGg;E_tO`+w`o^klf-it;5lH7yYw|51+-$WW1_)I4tb4
zdA8eOzt|b2z<OKeBCUfQY6Y512$sB%`OI3fh!YV@UJs%ME!3<ZdOS3PQ(4qx@wz6N
z6|uk#ZO7Y3J9kuD%EX><dmj91M{Rk#ONV5Rx9`W4>D#ZC(3gifpRhk(G;vLJDC(t^
z$nH(vTT5EN@(}{@GVgrwl9fuMF&V95rt%r}NxYw|fBV&T?WD7H{`xf5q;+Bq<*%;I
z67pOxxI)3lbaZlj#FJq76a`^~D&gG4@^sC>!gC5dP3|lrkI?*5>^Dd)AxZDX>cz$U
z=qTbY26l6u=%{{mhTyDY%(6nw2UAkZ$=iJ9q{$~0x?YqTJw3y%zjULwVPg8;%>1kD
z-sy0{^0`6q=Y0pt>m5FvRwGQ03NScJR|(dl1}pAKu(}M=c-a^p{B*$7z+*~F5TY&L
z{D@<ogF!(3bGtM`s+o>hxzvJ~**}!EcsndGP|jyxnb7f@CQ}0CVb!Qmj-9H@iRk8b
zlwV0(vMb_5qtcV+Z8Y+fo+1pD`5oZ6r%m|*<SpWQy#_sDq7Lr;4ZM09%&H>;sV>u{
z6lxdcN&;T_zuM_Sa~Ws1)oupQsAqZipQcBa&qvPEkK|1vCW*bgi>?&YT#$^r`3A=m
z**cvS7RVR%fMi$yVg@b_?mf8zxwe5L6?3)*C|jaV;a9@1WILM#cPw2lt7z4<x7S-1
znKL%EAL-T_$WT$HxH~#PNO-DeTiWq%kSIEQ&e<j2GyPJ9cV5#|zu326UpakSt)1M3
zx-Vu8=cC100}b~+sd*N|nOgdara3@q*vRwP3pG8W7PixU&C$!(7?v`wGL$i792mUs
zd8~oB{4B{0HDhJ^s>vhwz{(1(O{57O9*2F8(9EYtTrp33rwf^q#YyJ&!3m!}92j0l
z$4U#WqSd=smtWy$Vt(r3&MC{>Am%+Z5S|!M>79!S1kr?sO5N?z7j*8C5B0XElT<#(
zotVRa<0QD7C$Y-@z^f2&=e_0H@#K9=nYU3^(3mqd^!3jqKkj|ZHgn^Jx%Tss&Z;op
z!KM9K{r<w;k&h1ATzCaLXHet6FsMNx8r^nn#0{+ofy~La<L^>d9C__zMnm>8<ECU>
zN=nk1{Il?&9n`1K$udU-pUXQTMA(_+@87%wXP@dZA&G&;K9Na#xHHb#6Ho9yDAT<m
z<ZZ7mtocMDgd@acybl-DTsKR=8Fu2PfeV6VIK3sgPF?V3@E!QPuFFWfP0eeuJQYLX
zj>GrxUi&zA>36k1ehrG_FDN=gDS&3UePP~_$CJ1jchUH8?typ-S6xxP3UTL4Zydw>
zjwFqPE)zdVDSUJ&RjpgcYM$-6ct<m2nRNT>BlA=N<o?s?)9*^~1ouvabH=3<8!@6j
zzwW6S8y|?|m~RUxoR~vC**72ly3Wu6B`uh$Z+iA-rJiJN<>_aLg_cN$=wY^~A%bU0
z@@|MFqPecQncFt;30D`$FT~OM)YDUWn+vceL>yPlvx@B^r8W>Wm!9W~65dJ8f6m{9
zx>NPQnJbeAUP>ASHRjB*URdnB9jt8dq$XfgQZjKz&k>%Hz)!7(F?kvtk<!H^UXM!I
z$1YCu)!GGK_B?kw=km<>Ffxwl2}7v~-KqCvo^3|+GhO?i+SAx{X$yJne%vy;Oz?sA
z5&0fNL$=%TPU3E!iLFEaK_P)i?iq=aSV7oSQ%rnWiVYiVNB*u)r$lhcZxw0<zBnhQ
zQGLLL!jyuzf+>3_D=<w_>$4X4WPmyK65ptyISVm3bEu~{%YJFH!Mm76Vv)>~WysCT
zT~$D7A~yeCw0Lz!UwVr4Y{R|J(a-0-D&<_4RC1OcmDNdh^=DSlbTLfb@qtbT_1czS
zFTYWKPfDxQ!6Aq3N?friMJ2~ME+qo4xg#p%hUaSwvzKvqgt;hsYklIKAAb=;YD)Qb
zKXsw=%YD^37c?%&5LM%}jV;XW!5@R)mF*8E=PN5_H`-U&9ZYnvp^5ZRd`dzWL%-+f
z(v|tsQ3Z?ogU;DCx043Kv~SmG!sSl;*G!<~twJYw!Gn{AB07!x;!X2%<r2B;oA-C$
zBT>HPU2<fK_tx|N<{>XZ)@%8m+q_%%WS_-LKOl&_7fpBQkesqj_^Wct3Ci6<$Uaog
z&0D4qp1u;Kt~_xkBO>L@{yLscd^!HRCvvBW8wwfy?~i0=@d&3=Dvl$D&4fV4P2JS!
z{i74?9|lC8$h{Rj{b3>ho@+_^TQfVK19NfHDsuDfDz|kNsX3mP?-T0qm?C@r?0udr
zyX3Y@_mIQ8gkq@kE}h8_0wp;@-E$j#><Yac7i`lq!$)NfcSoquHq$<#ACr1k>GagY
zV1er#_1!+>5NIX|g%;?9SQopfsOsxI)^%V;y1-gC-NHSfN>7gu>}kjGX5>MA+#M>E
zoRlIOtX2=WtDe>O6yEZ|r6~?mzhQ&#dB(@RW8W+!_B8BiS-2<vNs{7uu^htQ-S`yk
zV(J+#@j|g0dGUEK6Hi$}7KcLZqDi^Ro|cy2Iro+;YZ(dC5a`aPl_iPYNuCMgju-5H
zD7ilf9dMD(xHzr1xr!`)uF^f*2Pfc@>(b(bGASxIJI(h}76FLer32e5a@_35_y~F<
zeI5Bs0$6JuAkCJ%O`UJ&OmFp*hsITC$5p4vY`+<HF%-0iz54F`VhN_x17!pY_j-JY
z=|r4wf-5_CI7q`^YCljPx9M{a*x8ruqhXr!Oq6=&c#X(Q3zNaSrmu^W$}4;3r1um&
z?;|CyryI%SO4`MgynN=9P)$U@d|Q;~*rn`T(RZI0N7*DFFW;r@Ln^TcL42Dj?G{os
zta-K0owjxi<xgD@qF8tk<RMOI6?tfxG-EWvfd&5~4G)Fr)617{X5IVfZQ+YMG)#Wf
zy<Oh0ec$_W@pJD&S`P+WGhV$7$(+%)EN5<h87@%x0$x(5a*9>K_`!K_`@>%CE+fOc
zmdP=T%4B37!)VI4<)M+{x(7W;t8w$U8SMESTT}ryO0cQ)V1BmW#IPhhK26FX(UrXt
zGAZkw*{AckJ5Im-t);m044t6;=ZR17&i6X)kQdJpCgQr(cgg*HNnhXJf0q`DvI~;B
z@6%&ITS|B<f3Mf+6T6<?lbu3XN^y}xT@m8D-a>an?fK5((17#%G>NoN;hx`vBX_R0
zc$7L#=G=S2?8G2q2F4GK9}m(+P2uDQ2q-e^cQsbu=GjLglf6Vvpk6?}$jGRdP?mdC
zu*E)IWwg}cap!A?N7vB$;gkib7cZ#TPBNTO_mGy4_7ja@jgaAtyookCf79Lforja!
zllZYG$Yy$WlDK2eVb=l!W4VjvENC(oWGWhCRY=OO@ht7tP!L`M@7KDfZ6K3{H!Oww
z&|(Z35Fh&5#Q?vbs#(n@3%YwJ^>y_$ma8@RrU&u&DZ91Tl<{nn!`YQ-q<y84h@S#v
zU1uOABv@&9L3X%YQl`2q`e-nHUg8Vsyi*>8+8QL;B4RF~?+*`Bn7GKcTt2{2fKmZh
z>p6`l)i$aO`(y?w(cZKOAvO2aU|bHvUC@j`$#-y|q|r1J6t^w}I!UCz*y-ORM12xR
z{S3cdP#kk=AKTuvo2UrF3Zfl$Za!sw2-`<70?Vgp-=jIV$yFZZz1BNJNFGxbCsaQr
zES75@wd+YsMfDD5ZQl<W1w<_93b9Yer%kG_n$yOI2g-C&pV@mq;TY5nZNfRrFRhUx
z`LS2?39+oCtgkKvqLV<A7tLh+@cAj~J8Z1;Hnd?XQHj?1Ev2Vx1<r_@k|eg&UZU3}
zW4M<iaOT4uaVCihg7_giu14tNne_AQqB(3$%t0``Q1iU~cs*|5G}&g5DvQxb`F8T^
zoo&LOhSeL=`&)<bEAJN)_ll7}y3))wd<jaj^G<1^PpvmYfikZv!(g~IpS`ByeY@Q_
zT-<s}fv&!cO5;>@p7Q7@5>}Qw*3iqJtkSNZhPA@qwi{1ap9hr=JaqNWD<phhnsBa1
z^Tp0!Pq*hr{xzt4Q=LZ#J(;Hss`NA5({vm-l^Y{+pLv4jbWAhTGy9%I(6_z9ug<W3
zIvR|QxZ>GBeeoeg0O7zn;!!ar`O2hlTo#$^#UlvL<b9`T?!=?yEp|KN0{c6|6yLhj
zgof&fyPvP$CH12`SCW2ADW5@gB+yKBw*Q*gglnKiDbb07tLXw`c2??&<B>a4-6w)e
z-n<dQcOai*$2;b%;-Oi12P|A%GZ~E=_JEIUl;C;!ZN@C|io=)1%b70;UKbF7#Z|Gu
zFEQ@rn$P8U+AxuS$u;$40wH7ngATnT&V0vR$K2$4nf<fRd^DJ-6B9r8`CO~SU}`8{
zsHSQ2A;b2*Mlv!hlIh&06S(&n%<sbZdYRpt(X#lt!F~FzI2L2`!TZTcCOF*q6`z9f
z8ISNR29}7r%QkE28|LQ^zH17WO8?Lae|NGtKY+VuAoFwwft|7>!*N5e7ODkxjkXAm
zc(sRiJzcDoobDh4t7!6Dp0@c>(&VIMgJD)@t5VJz3tkBs)br~zoqiZBH>?cH^4)%X
zxDxC*FUa_*S&_ZZ^{w2D<Ewmgg9Sa^iv!`0az5)HRTn+aZqeeEcZ2YJAfuxF<E$Xt
z;3WV1Su$+_^MNl@ivyyi(<V^kDx)-Gq>{mJSmj3!qU{V$Aosk}5L3(2#~-{pBB!R$
zca9X4Wv`fcz%S@l+YqUMjD%HS(v8wHF6u3}Dk403A}LZomb9EomipLwqwliZVjTX%
zmt@X(sR_?!n~%J1^1g1;ZQ+1Myn2CBM!z>dcc0rOF)legzzgx(CupeGM14SvK(o6x
z?Ml9t*oA}5F%i1sLQuQ2eD;steVW*f<5A`L-nNFV(naLCpOfphJF_AqOi=MRg3q|R
zKX-FYQ)=yg5PwOCTQ7my)OOT?-J5aS^;a#3d`f6-hg~Q|a^kr{Au%S~&lx7TL;i?X
z0YY&F@JW4=p>tE&5!WsihQ=G5^MCUy6p1T)P!D3Cb33yGCzTjJy}kZOo(MuK{7vLS
zTb6ip;SvE?;W!RD_EXXf0}|F7A`~Ho=Pu)67^qmDw3~XWKdIfJAU;`5ui)b~mTSAt
zQ@siVw_kG+#SsNvYtB(|w*!l_=wF}Wd&ZmHhT2o~Tnw3g;l;R;a>4x7wpZJ045x34
zwj7bK_uMgu-#dI;T;@8Wp!L%8UIma5NYG+;$b@I@a5m@i`ONbT`a{towjt8@{q}CB
zL&e>#Pf6kP44yZM3IN>{;G-`)gma_LNQus}FPy{klse+Y-fK1F&CCVQcD@#x5}AX|
zv@SZIHdH;$*~8Lxjleqev7Tz}ta6NUT3#&W355F2J13$|tNI9x1Fy)DNQC=2yAfAX
zwWpi$I+BZ?V3=4))WB;AZ*`~Y+ZDZSnphTWkxykRNRvRaKiE(h=gKRb4<v>=jiq;%
zU1c?&d>i(*H4jDqU<Ci|vIlM2E3t;-^GES{ljKUgs^Pw=1T3NYyojU5g4|hMDx^xM
zCkPt2LQZIpk9ZHU@*a+EyHuRfM}Ogvba7a+V`YGAZ0BnfJr!AR)^rL))4Q>XXlk_0
zo%(ajp|ZBB*oox9=i35&ALPw1oC&>T#Y}Pak?(u$*;7+o`g@7wnTaAlQi`w*N1=sZ
z`x$dGl8Y3k+Fnd#x$*Qiue(}1b+vy%Z^+wWhnGP;iQTmm77_<0ajxOXBgJl8T+OUP
zC%<hWK4kos&iG|w3m)rbv(`#IUJ!8$*BIwrryN~J{?7xMGW^kqnCSa-&p~Eowg(xm
zpT9Y0pwCYKPIL6yamzM_gMRr!L7vS{7u&-WM>Fd7uzt`-<)`scjtiiv`df7l40o6u
zc@rL$`L5vBS@Yrr8c|OK+Em!6{F%`UqvYk!PmDA^@lDC^8}Z_gxOi3LvDD~|wy5kr
zW96#y0mWT>FMKSnhPDU?wYXiHn_76n`L3~rVe%vG!3(E5yG2C04F|!&9FN;SnRyX}
z-26&97syJkC3HbxOi~X><pJrvwLN15Cbb7#V+v<LU`=fUB~>+!^)Ga+mOWy#f7U1v
z|4Lg~S%e#qUgY8z1Wv!;;}HQt_yvJ~AOieAdJ!yx0?y$^uUvxVppJ?#xU~%$BnV_u
zfu!)tLu@bb%Ey6a9IWkNATA)g3DyQq`PKu8@PWVvtGzJ!T`YZ=I#xc=O3G4eV@6rz
zOTrpx`Hf_yAdoX%&sBoHBrCno9!CuD5FdRx!*Q2Lah8Euzf5OOC^9DcnLqWdQQHtg
zn!PD@+(aLjiax0s!Heve5AWlfqm2S*(l0fe)sJ5NxBxjppL6}gWzFSxTzTc~Ho6Bs
z>5fd8uxAev+oyF`WQVCb+uePBqxjAko$5P1ibT^KhPN-3z@x^alVrz(BZx*jUs|X2
zbP0Wa&(-8ivwL~Mjoo*?k!b9yT(^np+-K6H&;qezuP-lVpD)gx<-?~B0P#rW78&d5
zm)(1k9L=0e(0fIz>9kIFdzdHmfu$i@e!|yD;Cewo;qjjAw|Z%Z5rP(RM5K0QuV3fV
z-)*=Nlf(W#DPmIid`fQK+wcNO&Z*ZZmd6dM_8zmTW(Sg=4k0duCcGForLU-?btp2A
zc}0Bh2(#|-H$f+n-g>>w{0!#`yK6r_l(;lxHoYw~k~Lk4vf^?lgU-iB8nI&w<}7&r
zhf>6hL<iE@b{|RK#cM%a>%@2_oFZL-m$YqnQo67_=dL3mBpIe3arTEewUL^i71`CE
zV#y+Hc~qX8pY`^s*s5c<cEog($y>?usN_o)@=!4OTNmI|$C1WR>4NweXE_Z6^g>9g
zMUTEGOKsSVqoIz|N-36buam$LM?>bUd69u<qm0h>CnPynhHrG6D-0WkX;<|#hV;b_
zkU5s_VbqDbNkseDVvI$hoNTG?m<Dl%HRPdxYK;z&go+cMum_8@f3v1cG=2{LDE<@f
zt6=%zKxtN-W}G>_HW4}8UE?Hmw;ksjUod&@8n=Dt{3^Tx-t6qE5n0wbCzxpDU9Rk`
z+jDH<Rcc{Byy#Yso9{NNhu&IW(M#S98z>v}5>SG*L_9VF4+0-?0CMiY;ie$4k|m&6
zaI`yc9v)a0I7APlx8Km&6=k!80$<7p8lnJ=g@|Bwa<nj7s)NvVfMb*S5Ud7m)~PQW
zJ{xVu2HG}m5_}giTWk&lYnHS?V13|dKg^yG@DqTk?T!LWfD&*bHb`6?1P0#efdiWg
z?jSZ9U~dY>4gzZ+furCt%wSt%25?i`+0oGsW{<HtJQx+<u-^pOk`?O*Y{+1?M+AW5
z1cBWoKFsb2u=xXQ8(}v}R(@i4kbs>M6A-W`gxQ+GGyom}eFZQ(HbU6D{Fn`*m0ciV
zV1I=VvmYaj*^UvwJ`us}RPhP&0ozj;RI9r@*jB*V-k4`VW6XW*78bV4%5IJj@0Q&e
zfmPBmoLHX}8yWu>yE70W2;Ucr_r>7Z>w0VIbRF=Gda-11zV2{tc!!Etp88mNU=X98
z8smWjyKn}ZqRt&%e2SY}HBnGD=D#N_Y`?J%nf%#v2V^Y#iww}0Z0utF@pf*nRQ7K;
znro5tlqxO~MOSY5tnAaml}<5KtLOalnU<vsFDIYh=FHUL<!ov(Eob-9)v`0tLLMz{
zzr*k4tsaw+rSr~#xlt<a&B4JpjSOc#9_DC$AG0fn^RVl*<bx=E<AL~2v3kcnb-`@&
z>H1Ws?Nj@UeD`cK^P%&8l{%B>ez0p{@l0bdG^ty4&&<IY*Yn+pL*0g-Ki3_*Q(r%<
z%YwRH6K!^aG?S5uqI{yL=|Q|G1R_?_HkeON{lG5MRP|A4gD9kCFx@nO@+5DVh?b1Y
zu~eHzQoc(M&9ik23~Ic>vme=+q?7XD6X>-=3&_G|;W1%ZcGt2fR3bbY7@y%CavDE<
zM%5{z!G<)B;C5{L;n4p6i#G9xZdO_GRg#cEIUaT;aW0a$^Vd=zo1FyN2F{mF3B1+j
zy&tz<ujZ2FgAqHJAf(lj<b5nxkEZ40Pa2`04KJVLtm!^-(?dJ*Y<l7u!GP;28BBqD
zEiRs8Za&PCs$`{=#Lp7fU|v8aVR0giD%^j+lEOC9zDqg{WV{k%s)A|u7Kjh|l7^Bd
zVJD9G4@7ERdY*%3*eCk<UHd3Q)h8a`==fIw4st>tmQ_0x+0>sI>!utUGYoA(d%N^=
zj8)wAJ`$8Kl*sv9!xW4h_J7&?Y55)N!f0k~Q%O+yeo@?+Te!3W%pPSq;$~(gG}f<+
zTn~~O?Pj=~amDmfLU@is72iD{zKm!g;|{@Ig`km<p$1{u&wV3sXl~JG_WXr#A`bi)
z_ZB^e?;Y(49XiK(NO?d*hurU~$^<F>p6bki8*^{{g?flNZ|{0e%%Oiuspn}Yzuj`?
zD5*;A2-mr*p=pZzR({RAQK3Az^Mtp29<lQG9=ZK?e@=np?CY{tqg@LN2?;5EwZ1}L
z)4N?>7Q7wB5%3x*XxhFr--&JN(v>Hf?9sMY492sUu2yS#Jw0O<eul6JAJ)!x`kJz=
zMat`Xn$GR@_u+VS4cqq!PdwPh=wmFla4{w!<FfkL=+*0Y72BV52Vaq;%}rU-4@%jM
ze=9vUFMSv7^o0r8xlcpkq11UTm4naN?aWRq-yqU?UJXw3#>;UDyIKuPQ%2|0wWcd<
zN9>Xr)?qZd7t=;h8e=q(;rNE&n7WJxJl=nIpXi5oIuC`D{5rMmY4~vPKB;lrd61s(
zbGWyi|IxtmNrKOsRMyW~v!Xc^B_2a7IX=SzFWez$EsB22;*)>&(1}>-Cvle!xh6}q
zF2u1Jgfq}-ogf)34GOw)>ySinAaUw~#Xu$HM~3UiPnEvhv76kD;A7fem4SCQ{co~|
zNp+Zn9`AXbVz!%L@4Z7IHt;ID!_l>4Nkm2jGU<zS{BHuy=(_f|+~JqlT^ZktJDe;1
zrbLLYFStaABq)MqjNzt>BDH^>g26nFYDbO~?ltm-NKVbbf$I?6<6U@a<)){nqWo@+
zYNYQjN4+nUpqN(CyqiuOIGnEO`RLh@+m%bbFbYn4XI7%LiXF*sQa-o6=nb#)r7ydB
zrem3?af}>y>AC_#UbOW|4M_v75*F*u+CHMxW7Fk<9;LP?Y<9j5u3&9BQ6vuaWs|C?
zyZ59HZxO%#QL$v>+iXYs$zB()p(Qw1O<m(Nmozqn*po{i?z(49iY#mA30yl~I!oq!
z86kWiT#52nGZAW!CvsmoYVpO(!P09>u90FbV^Z2)1-T8kTGN$pgdHy-z5+gZP@N3V
z$M4KU@a^;mplzRboV6~Or!Q%Wsxt^VuTu~5&9tXF(#dXoguvh47*$wEB?nS?60^&a
zE~Ys8LXOhJKI4G*(YTUvvyLq}bRxn#2zRDC;mAHKy-aYq6;4FJc<pRuS74!SndkXH
zntXhj;Yttjvtqc^%7IBXnpyLl;{AfryR*6vni*DppsR4bcKuQ(hn{mKr?5fD62e!#
zF`MnY3WI)ovY0s+T)22HAoVDetm1O-<H5aQhu?D2)x6M%dLN1Rc(P;exjIbIIbXYL
zP-h$&DIcf75g~>*-|hdl^uvTn740YZ-m{rzPL-yrR>aBnmvSP-Jh}*&1<H%8FT9^D
z%1Lk%x}u$J;2RKeunXbUHL&l7>yu{}F1=6aW8x51B=o4wE>G-MrJ9kt(nV5uj@pc+
zZ6a1%<6h4VJqjTqNFkg5yjXoD?`#}J-mKW$3oReyTI3YYUl4zZs&;=gg6<gVm%IB}
zYCG%>&HPt<Cm@vg{+LEyUr|m~W;Lc+e?{=mLK<P<1Y$lQQUHP(;lCNu0KNV+q!Io`
zNP}7DH@jbg8;19-kcOWh>+k%b{=z#}29e1_95lInpfFf*#LrxD$BV<V%1W6>PA2Eb
zmzD4Lj5WeLE#O#RY1Tt(1dlg4L`**!;ZD~&*kenTYgt%#!lkL7u*`~QN5IG1C%Y{R
z93E(9PM8rhHy#_G=zq+A@d4ds_nX|eMiIHU+Xna*X-&G+h}?8OGz~28>^xY3`_@eA
z-4)%bS!2CBW%$KQaUUd3RGGA^zh|{w&Lhg*({*Gj{G)tl_R*Je=q8a8db*<K<Y)GI
z7X8eOffmC79UaEVd%c&ejuB{uF3T>npX_1PH%=|1dn-#%U2H77U{!NN$U1HDbkK38
zb~f+8k@q1_wD#G1%2MD8H}OWh<#Y5b-_;S29iB5)yq2^4F^a`2eyoK{c1+}A>BX3*
zMe)+wBTNOC?6<d`xP4Q<R<o%Rt**PYGeh7^-Z5U~flOY947L({r&Rf*%mqK2*Kl%D
zt`Mz@4?4YG!z&o?FmbUOkxF<NIU*fG7(2N6=g%TF_KI|#3rWFmjE(L*JvDnF1;zNm
zOj$%*`NCzY@#m9|1R5^Py@?5_ru3xqIpo!4QW;;{nhj5WqPy4=)gwF~SxhsBo9?qK
z?T-^3@GrkM&o{V(-x(wPnGSp{L=eJG$1NcOW=3wFqnKtZbz*-rMEEmWwnhk8C0~gU
zuwMLsD?<2Ek;aFuJLBQy#<=u=`UBNis~12e*p~|kWaA^=O}*BOt1zdRe__zKlvo%7
zgo5jdi7#&Ykl0>hUEL2aySbA=pg#W&@5!5c#HEiBC+>ISJbZPJ&E0~*n|ty6PaVI;
zm2eC$U|4+IP+Lu#q9@m`pgieh4m(Fvqg`?W{H69s-#5>f=!oP~&RZ?dEPZ}C)4Oy5
zRWUREM(pKnBEM}_#0K~T$TxbHmY02m=M=`=$@4~^k!$o6C31BWNvDL&+Ko4=e@eOq
zvEP;<b>e08>Ear$?s=07ggj?0UN+V|w56WdcT$-8#4P)^vF=(A4U^&SqhhBdcaFe>
z%oT_;IBu4c_e@7{UAR`0jb<w@VJuHfK5i?K$Ztly_k<Z&4_E4Adpirt$eKjPii+k=
zt<Ua`Q;c>(JD)x&F6t^VLA`L#+fL>a_R90r3`N5c(6$FS+ybTdygIBi{1|nqFYbbb
z(fg3vqy^ufOJml(n||^6u4w!=lxFy7D5L)FThXr0psw;yhqvXCmyyxk7pqZ`%$!nx
z6i3t@S)g8hben2*s%HJE21ja<3&yfHSdWLwNFKt4b9#qWH0-h93=w%Mk{fPP#SQB!
zJf2idXDRkHt=p_yXiQ8pOl*N{Vj$d&YqC#&I(fn5qs+3f?|G>vUosN&GX|jN<}t$h
z=pZ$s&+J~l&+n854^%`h9V2=YH#_~|kaw7*Nm_Q|aQLjCcZARR3YkiiN`=ao<Q9cJ
z`|W4!!tJ`AI^28#?Qtl3Id^Vctw*64Ja&Jq-Q~qZ{*~^i;?{g{flzmAej1|kk<V@K
zYoF+rT$bYID8}i#rCc2qLYSUIQO{$fNoDt6W(IcV)P*JbI|ASC-^aqOWE6diaik-Q
zOR0<XMH1S#BQ2Ir&4U-N(b1cw5nG{Qns!%YF>WwT8^p{6gP*h+y*f)WqHu{d>&j#~
zbkalumx%7HsdUc>&%W!WyX;GL*}JkXo|<hVnN|8gy|c;1y70YC8KE6%S<H*tJMAt+
zH!InCwnq(3dmRS%3EaJ4_5jgWGL5^SzD(wOS_*uxh7Cez@0v0dGSlV)4*X18BIuy-
zT;EA*I6e3+mB(dC!?ep32E}h5X>^p$a($qGzr@)S-+kDQqWebCi<4!f_VcI4oq9}r
zSc}2!F8AAAl;%R`#Ftb(i^^+!E`O3;Vp+0U61?>ybMDPp9!Jlvi{t1z-M!IO`%fRD
zy`Y)U7Mm{!)xJj#D?57Yh1IPWgi0@zh|<SXHH()(BRzwtIIu#`r!2dYL%LbDY(3rW
zF30Dn8(jwbN+<((&-)FYc)%SjN$(+QqDC=*+VSv#Ck|nq=b#?~j#K4#aM1YFr!>&n
zJ;9&Gq;|3#WP4dTD*aA~!aJ$Gr*hKN8PQ#$DWNct{CWAqe3zjG`?ksVcT4&%Ouj5L
zESi}{jwhGqNf%w-IaX6NA8BPi`DuS%)U*dtp0}s#xcA<)sBv${$?{3diE<Y(cV789
z*16ig19($&6x$HqA+1Zj=t`2CZ)zou&xx16Ay?gA>6_X2Ug71_`P!a5sk_#T4j-Ng
zmAQ`loSe5P6B?H^MY_AsHPyJK(N;<=qvlGDs-<~+W03DexyPCYYXl29x*w12E{=Zz
zZLD>eDMJ)~D89C|@2;d=cvK(CjSBoG!i&nRsddog<Bkfy$ybx_i<n;Ce>vd$h91?d
z|4|#FJO?!%?RA1uu<!6bWVPUdhx1GiZFK$dj!VQueL;qXG#4Y<+8U`(rXT1ps(R3I
z_>=od*c08zM87WFb2+DH<2g>?61>~Xv%@9l%`2Y^XA(M;MH3}!fWTp91n*Al9Rtt>
z8Sq8etbvg#MH~Hy^(=>|#;x#hr>MO_iXncPjuEWV`)|xl^K3`*=&J|sKSzj4r;QAH
zm9d>%j_;PVnEj<)x71SPugO#72J-c(3RAQMH}xS)sc!@w?-wjRcn3;j7tGoZU*HqR
z1y^6oC+ZEVyst1tWlv&{ruuZMSME()-G~_^e2lcI=I~9^+*5fqGI!EUIrYNuA5fFa
z&)8H5t1w;E?tCD^Pf^i+N<AZPSjTAldGbB9{!V6l{fYMkxAV{*y~pNw&p`1s8RhP}
zQ~0h?;^+1*&VtD9GaBq+tqRR2Q0_-K^5*&rIEtp=T<MoA+sAzx|HMwaKpH5%F_BP}
zM=6m@)u*X#;`@j?qWX`y?KBN|F(v9npizCLG%rP#KIk#=o-4AHDVKJ^tBJS2snQgt
znNOPmQQtX1V7#L;K!t&v+h6!84pW~jmpsi5!KVb%jt7MMl(}^91LRA0&tEwXA}qeA
z=E#?If0E(Ajn1R+o6j`4XnFgI&hI1bkBTOAl_h_k5%Hw9>JSeP$ID%veRSvWTy0&1
zsV9OM-6&n|F=_7c^*5ZtX%4Ea63?Y=?4!DRs`4a>(V^yGiuC=dT?4l=s|d@|@k~#9
zxZ$4c<9e7;BX@9Y7t!6lMN@+O)k|%7Y0@<4syF#ND*V+Kc2orJaNCpP<XUq70MAak
zfL&9Jw77YFbfxMkTrH$D4}#1yh0SsgT!E{;8rVLa|CGI#OepO701NVpo4okp-UE}N
z^EEW)c~6;o$tOcf9!LsrlQ^{9=a7>nL3?P#QzcF+HClP%qo-pj7sx9DCi`(0sOIj=
z><A=Q7T(KBVo4c#=-8q2XAWi7??IA<vffvsN(wg2^5+EA#lEVawqy|T<78A(;xjyI
zq{snhx?s6$1jGu3wW;TXB<C{6TM`<osk2;()^1QP2unGd(-3O8*HC?b1J$9>l)X|h
zi<X=s;fLZ1mGBG|v>HSnD77%dEhQFr-oB6(H>0?Fgz;7VqNVulQ<ky#4LAjUs)q~*
z8DrJBS>hbQ4;d<%Xc{b$>`&N}>**T~S{}ZAMwKZsF0;W3NpRa=l~q617<r(9Nk7gD
zN$3<~tF--gB-wzZaiFs*<hEZkNGLX`l%thVC}uyBkDa-awXlK9a^LOvX6BN1sWvhL
zJbGECySOTSJMFUscO=}Gy47z)P#(A(>5xS;95k!tLF?SfG4gnp+T*|r)x7k?PTQJ>
z0Wu@`s4ILUoJO(~9SR29!ZQj-_Rh-RQN&Mw{79||PIw)5zk82h?2U`py4lE^j&@$U
zl~f^}mJaT)9xj|k%rbwV^-h|&nAdh?B&`UF273p;Zfh98pT4x9?8Q|<^0EUyYvVz1
zLA|hh#=u?TeyRnypz<a~exw`u+x*0C?l~6EaPL}4YqNZCch49hU9;Q*F<pzCJ0*SN
zPIuxb)R~Kvk|_#PQ1SCo4;t-pq6UwNpRSwEEN!uW9$928u5ltQU&EbOV^GP9v!`|6
zl#`dZXQ*|7{?lt?0>#PYEpt<Qy%<}~H424mM<~YFJ`hE8fJd$vydSosEQE;ZNmg^F
z9%-jbkbY?$dMi5U<{`m^pbi9O=440O$o3CAJ;QvOv<wc1Td611b{OuasjDqiC2wHK
z59JJp6)Md#CERD}oTto4D|F9IS(dM~I*=Rt&>G)h56wL`TPrz3VOwaE0f%j-ueI%P
zumQ@XUa(E$>Cq1opCTv8m=>5mN<Pn{WlVhu-Y<VqaiFKhAcj+GVJD-m@Db|Iaf%dF
z=Cn^tlFa?j)zgz3G9FjmFJfMBj`8xlgXZUjjcA!Jha1PRzFU;L7qn!O!M4A~{Peld
z_(>Im@^Zhj*E1jY>h7v+ah<6&mzbwmw$2Snd*%Bo>RsJJ+(+GIYTua2fxIaS$6)Wc
z$}<Q0_E(%0njwE4bKITAtVLtMcHv}2Sl%Fq8>>`OT2t4|On#$J^5oEDefjyyR;fPv
zf#r`R6K!czWs4F9-Y=WV76q4?q(D1rh?hwxNn{vx95bfV7G+2{?=ARjXOt&6Ad53~
zeBt`XeYrh^K9ZHOX4QGM%6+oAID+O9AH;mvE6<tT%Ny#N=KAPSap2lg^5W!8^>Px9
zB}Is8;Ee6Um5=lv`IbdK%M#1%x-UAQvV7q4dEfg&(@hIQUWJt07v%{)C-~M?zL5&C
zF|B#=QP^jPS$))-$8JdB7>a4}T>lw&Na`WHMFL4uj((}Bo<;h_vReTz9YePc`Yc+D
z`IJ<V`&<+!f*-p}-Kv7W!^opH1iY^zd_0b*qK8ke7@{C`+H>~t{ycEPy<=R@CnxG8
z9GRkOjb1m=m5jylq*gxnwh{N4l52R|7fc>s7n4XhVqouU8qkP7?1Q)`QBRnN`@H@d
z%&f}db?3G#4pkp>WJ_)&+6DKyGoRLTGLDBpo>0EM;m}t0TAzx)8s!Ues4gmsJa3Tv
za9hcCZiQ;WTn3ZJi}2k})E>@q=VdN<TyPlU<Z~Nv>Zy$-)`!|>WE{}zCP?Vxma1rc
zonPakJ95;J%4O(|g)@_atcXKNuf4hF=$V^OVMS<*6k2*#h--rG2%B1>f{=exPR>n*
zM2qVEjR~4A&UQsT<B><;5hwJoZy(a#eOA%(0kS56(rJGf$Fv<$uu<qu6X}Q^lQ&K@
zhaQi+P-^PwPP$y7YR+b|PVJL13A~ouNF3=UhZ@X$9b$CB%8`|Br@e<s$<>Cgp^!(A
zCua?N39sMDe{zCdkQUu9l7ml$jw0P_li;rV7<Mxi=^B6E9`*L9!;u4#HOdwSA3F*9
z`F9v+_e|ZoZg&mZ88azOFY-=e_jNdrB9}5W<Sai^Rkq5p;B;=<lW3*t`s*$#XGeE1
zpG-&5B$VCwf3&@2bR5~1t!s<f7PH07%#y{-V6iM_X0{kDW@ct)W@ct)W?9lp+jUOW
z=~LaeyWhP(LU!hkjNFkKvB$_Y*ZgMJ52WOAHQuqdHE8C-2nD20KbJ4Zej0G$H#t?b
zJz_LyE|gG*u?R1R7f1<|d~gh_f&D5Umy46$a`2>bC+Z@;`ftJ?S2W6-84DFuCr|NZ
zkI6?E__D#v&VQGgr)0x^FWj~a5nXCy<_8cB(1P{4Vnyw0K2u#W`8O5amYJ-g1q+EY
zbfpr-Dz3Jabzoci&T%N^qr;CyIV%z&8yJ?rJ_i)OiWsu^0={$mfPiM0tw9$5ydlzl
zqoH%S*m~L)bV4vLz_(J=lzS<&{Db<2rV3WPlFcd!HH@i$7fGG7s{eZuQudGDh2?q0
z0PC+T9kk%8!}=X`%N4J3Z5Vq!Md>I?O6Z$mZmkn&-kqph9p8rMiz8pSA|pY>w}1R7
zJ314gnq{RyCJL*~&dri()h;v^s;N#MFoa1`<qqIzPIL>ia<F~GIy^zSVjEg49W*Sd
z=J?Dl%+U<IYa3e4A;KM<3Riv@$$_z$nUA>x(kL_)iqzhTw!UFwOFL^xJIS(O0oTXY
z2_gAnPyWe|yf4!T&^DLOtRTyMT9D<&E+hCDDvZ3#@Ln-yTmbN7G~fxW5b~k?=|XAw
z&ZIqovB=pbXfE|rIk;ZXgBbYaGsY>o?~Ljp=nXKzx7`m@n}gN^66~ZxC5#=c<vz^;
zzDPq(4D@nMph46nU6iD?Sm%bB-J5#sx0FPWAUOtGNS}rf<4?X-4R{1Ls+qfj_OseF
zF^Jj)!!G8aBTf<)ba#+%m|X7UpsS4CY)(&wo)Bc}=*{Cl5#Av55FQ-5RE|IxttM({
z^oHq3;3+mNe(mdJjFc3yL&9JKm;Vw9?`4ElZS2i=4Cn7gFc7P{9(y)%Ypfw+Yoo=;
zwVGpSCESLp#!{p&H<<dx9(6YHlPjvEUiQi({Un&qm_q@27{2;e+3RtwDlvyGNmkhr
zdqAgp8aq}VwKQedEdAtd`Zn(*?F&cn#3<a3454G;=lCt}8?diiwDRQrl-{o&(QLOq
z7KzB3#Af(*xF#lwxjJi8N?b1S^=n2NEq2ktjxx@&=z$bn;tN;mk+-sr8uAr}8XE(X
z7f;<9Hw1|vmP%<1={I&wAXG3-w`<a-vdVO*t=S9|@GMCq#LsKYRuvqSSIzhKe?<|E
zSE8iHS69@pY{)&DUO8MnB!r4TyW!5>t<-lHvIS4!SP13JkBnwV>vqehV*}_l(RBV|
z*eJ-@NftTVF!6FKeTG@a&f)MU11yU1?~gimqr*bRbF{uSLb70<$7rEV^|mV_iKpt3
zC+~Hb?V{2!*Ap4y?QNtelq&{TFtN?<Rh;M;Z!_@HCE+fq-zkS~K;gt$Nc1RjZ_~ao
zL}CSD80g_~4jW5!2?y2a;3cWibfwOlmdfSB7-fTWE{>Te^nOh6c^BECzR#THGW1b~
z>zYss4`>%b7zi1##R9hLR7Y+05q_ihd;+}v?C1CfTF86`>1tc>kk}-bNOmnf@)LYP
z`5B0W(x^ZrVrdA<gwA3as#h`Jgtw6JWC*MYWhej}31BPV84@nw9VE)^6{Ch^B5x#|
z$=QjTKp=|ujr;kaWG6}nddVqx+-q22m=&y32{9$?YcgY#$*ztmVBtoQYK9a`f%D_1
zOp!|AoE5hBGbu!;kdzry*7rtmXPT<~IG2ziz&Z#*)HN8$F)DnHG%I(IpOhiMGYUfV
zbjUi*r)=ttunB`017V1NHp%T`SQy90vkVwxM~pex-;jXbR0oV5OUmG<?6*(^%3B&I
zr<;JxVN$l4)jGyY$+zb*-KqCs_9}h-_)dIAl*o}#*i%->sP#wg_fhb^4Uos;Wi!qW
z0fH{t*$<yGVp6^h<V42x!O@SNj==&$On~&GgnY!CEF|ng!83zrUTeTv@P9e4R<67a
zC<p3d_i$5uA$$kxq(p_N^_d=Nx5@vb7~dtM1wkkY!dEvGhz*Gg7qTvVnA|#qL5oO#
zx_T@zNWvbD6J4Ebd><g(L_%<1fm|q89XTBV)&=^2ewc8n9-*x6PZ@#&y)fa45o}}}
zu5!Kv!zh`cMt1#uEQVVAHiH96po+F~`y}kBd0a|r92P5C`Bx;NBH?S2Wt8QnXw%B~
zSjNPsJ5{pT)oIdWL1-wI&bH$W^KJ4rDoz_05H^>i;8IkQB2i5>SZ?vB9oN<{dV2;C
ze5Cym`PDKG4nV=H%HB{N_-C<P^!RQ@R=7P8Xqh`wViH4gd<N1(W5E@f5wb}_ZE9PZ
zH4jeG8a}3%2`OpSMwear{XIQ7p6r%pHf~cThRU7h&jgMwYBcU{9PhLOfADJ)=Ijtd
zc13Uz+;%kHe9U+k!#p{asZF5MB?O0!9+_i`*DW6)D&H$03~B@??CV2%hZv1eY0c62
z98j&^U&8g$(eH`p7|+;0o+#gRsat9KIHMMSqSE%GgQ~``#C@*GzS`iE)2GLBVJ(-3
z!HsX$fP%R)D?LuIQU!cwr3#S{h<{9&!4$}34*w!PqnNNLE<8{v!sl2O9oy{>g1n$f
zlr8V3UCve_AumDXqPP2rLPIbSk<l>_3cP$+y+=L{BqgMBY4gnq+_a)Z_*ogiBr&sN
zWOsPcWg3enE-tF5uoh|KuvvABqn@Zezw{b1adLZZh9xxxH2DVzldOfqB|Pa^Kll9z
z-!r>PVcYBEplJg*4*Kd+Mot~}bL5i^%w67k?u(*U$Znaw)U8e@oD{HGiz13?1$!v>
z7vq714Pj&F&<S~s@XdEmA*eG?0Y^qg!gJ~gkmDiF@PNN&x8&cdL`g)Yof{J^JKma|
zH+rI`@(SNj5+%d^c<FGs5*fF%9UbVs^94bLO}(Qs@+xxi&|k<loxQHatUIG&x5_4k
zd=Pjg&Giti-z&Ul=+Iixvb(^N)p;e~zsLI_Xyog8I&I~98K)(9>+s;{HrM_3B3;<@
zT6<nc;MV@cq!smO`MCD{w(;zx+KAJs8p!Z=@~*50*@l~XBWWTwZk>M<>@l8!bKnV+
z0Fl^9#%0c%%(rtGWds-TPK7oAo^Z$;6kRd}rKA&rz1tgDo3XTR1J`Owo{|;h>W~VB
z6GZH=9|en@I8|TG3N0mP-V0ha6L#)12V<ws2zzQqw{k%*`)6U59m_d32u0jEsGdSU
z7c=C-Ir@U`G^T~IZY7d2^Iiw=I))2&gjH+W_n8<>JUG_u)&?Xx^SX)CKvE~EL0Beu
z8k3fN<#4;)s@O$};o~kT)d=VIfo#D0b6s*wRcQJqeGaZTi_c=*poLER5|`8S%lY@&
zHJ`1IKa;6O`iEiUs3l4GZC`SKYDc8I4igqriTLj7o-|dUVoao;8eGkOx%|!CxLRo@
z57IbsP9=z8T05#rHff4gjrY`PoR1rU`fgFWPHEe{1$11-3Qn_1@tY%=eSQ=pEZd!V
z{l@CqG-GLcp0P}jlC<#2tkdFbsG<{%(nRIaTH%>u@Yp69EDBpYVG%~`>JH7WJZ&3&
z1b;M!h8JB-7?P@NHVoEjE~))p{zJL^z=g3&&~?s8af}_kTvCh%N9pStgh}_MlCpPe
z?U0TnxdXT=cDqBZ^Gw2xMRZc7gkaJMo#AZp!=)HA=DZ>ixA^l{@a=mMcH|9rNE@|p
zb{}R<V}q?$t8_!E%=3u?La#O1&4;$~YS_cX-NLtG%XU~9F%Ra3u1S}jVn*Z;8==&|
zEAtX{!LsI))jSe!!QxBHC37W5AUGJmBx*9vzhDXGYaZ(1eDhhA7HbYq)>u5cSo?8$
z-ISx6*HB+TdN%4~Ai+{p=|YiDqa;UNt@C;B8jOkjv8KpOA%sECUpdrWN-Ji7%~T;K
zp>b|!B3OZ=lr(Z)$In<Od4_%=Yq9XEFeZ=x)~=|*?2CFU^9|{unzcprP(U8fk-|9C
zZjgTPmwf4VFyADq58rQly){(E(*v+emy1PHue2D3<{YsVvG5aFq*P;D?PpDeK1Zu+
z<Y7sf3eMq=B8#k=2E1156DA*zYk||;JG65#E=1Y+1!Lr=a}lk0SL{q}n9e^t7ds*k
zube16F+b*%$yOyKV0#>|VrELE^Uy2U9TwHAs0_zs4pupqYk^m#PLG%8QJc3xoSvWx
zVsob`4yjW4JmrWk=JS?@>Bq&rE|y5G676bYyH;zjhmttL7pfIs#U9qG63M4VRS!m4
ziluX&@CKKXYu!BcdNK{K%p0rBu0yRMPcUz&o^llPm-1FO6;<AxeW#VzVsPOKxRh#^
zeuhJ~9!9bc&sxaqbG_w=Vwt9TG5nl~ks5+th*12Ub0po5-1N&vD~y^g-G{Lr>_VAY
zHiw>qioVk!oc^K~R^ZFM;vYs+r2WrWX3dqtQ*(-Hbz&bq;ijT&p)2#q<gqA_W*Z^E
zw;#+-sCo@E{YVTODh#>nh}28&z@yi(QuOzB6P3c4iAsAsR`a#B3Ju7^GzSfIi?I&J
zD}=5i?j!`Iht36pf{{4;Nem;S6pjeZ3CYBhsACEgYW2?D+~V*w{kGEc?U-{@v+YQ`
ztmWhRh`e@EEDlY!M4u@ugmS`j+^)=yX35c(uc?;fpHE2c4;lv+OAk=DjdF{k_O?cV
za*h3}G&A)X3e5{mZcX%EhH`5nmF$a>2tQ<U5?RueH^xy;S!$ZR)<ws~Z);MPdpOUn
z=2EhJsg<y7P$-sr6boag(u$+tda~tG64c4;8j{EO@_?TsB6x1L%WG+u#JTq*s&f_H
zb{pS|Is1fb+LN<|oo?ye)b86<<&stv6jR*Qzj)mH8q->v7?Z)Lk|ndK`DIe3EUQeR
zH#zh>lpw)_=Pr^{4teQdQIb8=nXGe}pgzFO48znw`8FCW^M<wjbv7|?tX9c=%|P~E
z?qHQE>3Oz7mo6LS(P$xd-{V=Vs2<}}OI{S$-e5Ky+s*1DSP_rQsg)C4ySqj%PG;GY
zz5PtbZ4|drD`Uf4yWp;!6WrSD#kxBGt?I^v#D*4`{SQD#pVtc++5GcSDavih8|Ydb
zt7%LNEIG8u(xE#g(}KEjG!yru>o}-nTPU&qc}-Q?qOKUZx$6GD?OOput%bb0gsby9
z`mKpVK4I+ZRXNV?TYTga<I`-$8rG5Qq`YM-=Ar8q`PHzW&+$(B_R*6-g399SHNwsA
z^!g{@!M&Ovt0j9(uu-GSr6cm?J0^$xj*07PpbYq}$GLB3vX;AzZI<jK^`Gs1+G35c
zt%TO=ky>R9$`W7CMNDQ~EN`Rj{c7P_J}C!fO9CsGXtpMo)Mw54tQl21(1Z^a+_s^=
zda&q!a?52bf+;UH3?R~N4bJKA4=P;!-1uyD9(Cd~M|Q`}{iD{jmS11;TywQ~sfA)t
zw-uTBKKt(ddUB1*IXxGYaQ)3xsiWn~ahRuY9(#)t(;UWO&0V*@hv5KzvBldtyW~;(
zq4Cu4trnZ)(Oczo&RbB?x$}gvbx{rNC_~&#{^!wnb#aCB%=|=6^0yOX`;2Ib^=kRb
zEZSvM;p<orXOI2L_1BgnoV^xy89OI7B#M_22Ia!8vd<ht(c2PQ;ppKEv1M!5#iG-p
zt_;LtFD0gFpK}fO@#gs`@~&b|pG+>-ZPMnRQ}e;Ofrp*lPmeW(HCsH19TRF=T092_
zU1*yO&d7V*OO;Y>+1;I+&&N)+nC_IaU+wd!Y385402!cAD>YEPT)4QtctT#>IKQ2&
zo`O9~_AZtm-74W2oM~=8yZ5Ty&J{kbl)UF~Qw(bKta&1jRy3(uJFl54%P~K796OeA
zlN?o2W_X%Da%Zu$IY#S9@;Q}xk@hz?X=iCHk+aHIOgxx7Z`rdXm9&y;pSE&spPq?y
z+nC)IxnuF<HL4zylN{=*N|UHRsCypMZ)J`|F_O`xdTCk1bdOq!;Xj&?QiQ)3(=>E$
zQJh6SY(JzaM=z?2w&bOcNx@bsO`SYHm1hr|;~jKxYHBG;j)1lBN$^Ztn}dD-c+cCP
zr6M2ABobM=QV}i4|JC_x8#6U7vj}tz`*@D|5}!FrIwb<rEPB%0g)FU&;mp`YD!&cz
z&bl3__IbsR0Bx!r)%98f8Xxa7rz#K4*vRI#qU2d*yYSf3Q!m_PnT@bhsVM9ZAZ4Ik
zw)68!Nvl`Um<|}0FeNaQ6ZF7aK(%;e%Y&WvVC@yjd|sKUPVoiP85tX%@9Iv=bpE}M
zgO2TgV#4^M#3wE(M)5P%;2!~4f8VVHa9_{^s5XEWV_IAmfEVLW^Tp3cfAL<h(6jx;
zd-3x%|I(-g*!Q>d0?_>QPv->_Bg>!83ji1AS3kpFTR`Y(|GUl$1_0CMpA8OwabDQV
z4O?{4er$QBr~@Zt^-U0dkB1bLpQOYo@bQ#sA$f+pl7NKm;=UI4ePLE4mOf|oaiTp2
zP4hsqjXhuad|pLApJGxbi&j6IF(XQ_BKnYx&p_n78J&Ns5@i=c(zO-5m~34rYs`td
zb&2A*AcKUTo5%^<1m{Nly>p6t$$D#j!i$ruM6qoNmk9-27ANNJ8$K-zr*=myjw~wQ
zAf4S(j9hhPj9VFfauM$9DEdR~ynslV<=uwfXp|Ic!`EHdMwa0bCo^k1GO-c9L{j?1
zZ<bOhf`v3=<{=Miqq5M6eAhud-NK4#>60$+#k}9K66cjB4ib-jbi-uZ@kid>Rk;gz
zwHkzPf^#erO59T3q2tFMMjGt#&3`YUv<C?x_8O2sUnFniCnV$9HV5+@R(m>K>rQ(5
z0XW2kh0(r%$MztWoe_(b@qo1ki;WQrbpEa^-Cifa`NJrAsiY2zHR`B80>XKGRPl;Z
z<E};ZYcl0h*#Hf_#<udmU!$!5$DaTHX0OrTUcUd{XhaX_uK(Fn^e>G@zptr(y~_TN
z^ZH9mJpC^hfr6#5sDPxl)t`=uzgiUjc7FfXeDw2%{iFHl|D8R7p7t;HgumjjSeSl7
zuNZzoo_=Dieq*tIS{3L4#)+TXU-4MKP*eXU8|#-n0pQX4U-Gejdbbz>Z~A{7kfnhO
zFiHG@)B1Ief6L1Hd4B)kWc_zoSwG2NzrbPuE*2ZWK*kK<ZT;i%KmVnp`zIfZ2{6tI
z=-vdlGXBZQV*16(qWfhU`3)$e1FWe4*T^qe89mcaUf54_ng-z8`1w5ZFZTxhZzBj`
zt^7$b``J$Rd%k~Z3;SzJ-~Yzd!NkJyx9wu3Dv;_Drf)pcc4O#-hY6ABW1**8Sy815
zgj5(3ei!&lwZ>Y$q~x;-a|*syf+BSd@utQM3Izpf2$Q=8jc2WI!{?!fP3O-NjmynH
zc+%hAuBGgy44u=~UtiqQ4V`?;s5`eWL$Tu7**RFl(%ZjA8j9=?P6t7|&kKON3K^xd
z*>pF;i*dzaidl6#VHJ0aKBh>Zx@Y#cXXF>t2|M~nvGLp2M|4AY$8)^Hu}=ZBiE|;i
z@I+u~m0iNDz7;|Q!uA$MgYOQk_w~;oJ|4#G!2E&JlZ2j}B$bGWfr1KO(Z?fZU|_=T
z?0sj1SIrbR$kpn@r%T`iq?d-rP2X5)=2a;Hj}dYXHqhtqKhWa#!sYX#m+4W}+cvB>
z8QBIOS;YjOIjJ4F9jYJl0*tR;nF-HJ?-q6edv_apa~mlb@bDlskoCh|Kg=QoWf+T(
zKTGc#xW%0x-(4S(WJoWvt&RSzpr8Vsm{@kwG?brUF5zL*5ns~VRV$crGA#LEqePk0
zx@{WEwHM!YyTD$SzI_5BG2Al7@UvCUCC~E1axFfiPO0IAS-K33775OJM8UTLRv=Hp
z9+{IO_#P1*O1iLzC^TsaTw;ViUr4$jS^{Fk>A*u@Hf`P!T*{!t9?CBN1jz*iTL_M>
zG%H-Jphn-61mO`9l#ftdepW;%2qA&JzL=H1%0iSRN*{|MKXgT3;I}~Hbj4o~{0Mr7
z9|3{y9aJj-0WanYzbOC$FDd{(DF6d6UP7=Hc-sSppSB=C_z@Ij3;ojnD(u=?;<Ep{
z)wIx;cq?Kt=S6lmY|B)TR3A5ZPd1*o_K>&IGse}gt58#+b|lvWjx$C_?^<J|GRmjF
z-`sn@daHONb?SExbF;gH-3OgzA6*xgTd#}1+&od8#4nV~Sr()PXhyE>zV^K!KCd6Y
z$?~Yb^?qvuHUo2G*of|Q0`~NA1*QTM`ap^9q`M*9ZSBA4pPoP_0b%RS>mvd72y=uq
z<aaNqE1;bf@52Q)59eW-V_E|xbzCe7rPGn^Z5z0YY>o41!Vf<rKRh^yG;|yn9bRDB
z>Ek=Bz;K1z&s+9UUuIrAXC1<gw1C>9u2><0SHJH(nDvBkVm}K}f%b?vLE+kh$=eV*
zJ(<*7{6@O&miS5$vrDOpBXwX3bA3lM+ReSwwQj~oNGlJo&$en|`yq7SStIdF(E^hl
zIU3|V83E8-r8QpLA!HPCF`QOWfN}E>eeToI%>l-M98uBN=AjF@;Y8K>CFk0<TAWX`
z`(`s`qFt^xF7>g8mr;s%&(L#XT3ls7uf_+}8{cGm2tDI-K++qzJtB|i(vl`?k7_r}
z`%9D9t*@_`hmFN0^eJ_tamO9UG=~j{$mC;n^j(hTN-3kI2ubqI#6?0i+NU^-i*W}$
zqtnBkrt0T#)|k{scm|u++ZrE>2=}XO*3HR|woH6ODYX}ic7p{e877EJG*#NwR_rPZ
z9`-DrA(+1)67^hzj%g(z5LaFUevJ+-!0%s$T-{frC&tb@oVKvP-UhZp0%3jB2gUm0
zRn>K2&Vg3nZe(q$TV{>ZDfr+f)_vi2x4W5FWN=Bbh{)(I7_@3Lke>j7c025No<p!v
zII%beUrCAZ4FQI15AQ;Y(mLuB?<rsMU|HbTNdZ>4Y%-z!ZVta(h4kTWI*xoX`OZ1z
zZW2jms)(tsfd<k*ByTT*kBBM;iR0cqyJG<M(KJ<l;!#|A!2yB(>pkH)$C{6_2^C+~
z;U}o{4qz?8ijPtZ1YsvUYI4+t7I&Y0a-OUp^$WQV3!HeEt~POp5Q-$Y4_oB$uC-XE
zJUKz2coaAC)0y#YdyBQ~C6H>`l0*wO*-f16Q{B!Ek+rXUA{HX--v{v{c>kCNCbWr<
zI@XlCKROVW+{)|9D=%6+MlBXMAtDKpuTP0cTal}|R90U@%&ukYT;$~d<?!*L_aTfn
zbQ;#kfj;q(_}<%>FLBOVadpIejv5se6?(!j#o!wJA%*kX^a4>P*xKC1*cM2d8*hSJ
z_>JZsO1p$kTfwD<_+jz~mN7F8dFs(ZqVWch6Iqc6JFuhhCw!myKyYBo*!|eGE(27G
z<;;fE`?}N*wwY=R78X+V*=r0Qtn^@D9v%sJZaNkoF=3y92SZ}omZViWYn3k6?5w&u
z9*%-0h<u4efTAN|L+&<}&-1?$!}vjNkf)OKF$m*h3|dJvDkX?`op@FdJGAmr$#se;
z+DI#poA{Va_k?a^379m(J|!6>QPZf@%uCbHxey{`b*_hbxHq|smqH!NVdOMj9$NF_
znVBi{%vlWrEthw`3uUKP5=SqmAj$TIx;)Vsav6~2<)bQx4%lldc#l4n%#v>zhpRQ<
z<$!DwAn>&?CFKSSa-?{c^28yhMpMzto(-+YiAXY|WXIOvBROaE+Kz2<>vrx!vyk^g
zw7;%-=vwHz*5+8@;4$%N)4|;9)H*>?%6+C5@z|~~8$3C=(Pv8!d9p@X_^(kRv7+DS
z_Grw?wlqw9QqA4Qu=k*}mWua28E{Ca6gHEcCC&Lm<d|w(RPBVy8$xi9@Ns(vqg`sE
z1H-<sue*KC=dZRJqrYck_@e20v9UBOT>2$$XpwS;@P|!jkuzM`mu3%RPRHUSuX|FZ
z^x|$KXDv{bdsj1_l04s9jAn~|sfG``<O*=h+O(^>4crQ-Jyu!dSfFd+l5S4rd34<R
z=5TALB#Es37NK@iX7j=rvWO%=spJBDJWrFY8vStC(#uZ%r?p3wP?RQ4=HEBI6E$|4
z;K8Ikb(31If8WHTbn1M?xZ&fUn?$`5AO9lS*0$?+GXaxRzC!=S`c)~A(F(rtVhgyO
z8S=*U4H(UY%TF;qcdJCl*=idv9ok!kZ&rD1C`O2bNht>iE<>f1Hl$7wG7{518(1$Q
zLAJ0Lz@URdZkbcU@9!48d`kIvhsOhoh6l~QW~(?=^rrmXP|9-IP0WRTv!_XS@;MC+
zbRT>V>;(P<?qd8ZUZV10-o$O|Mg6LR?oFXoNoz+DQj2k|4CGC$JL`$*c0<{-APv2z
z*P7oC(oxiP-adIw>{=J-&RINfX^if6Si_c79b+r^Mif5a9C9#RJCQ})how0n#N5LQ
z%pL85|F#}iFis4|fxgx__G)kK^x5$=(t?r|TZ{a$>{WJu_F&4_ed_d3PkZu?APN{&
zK2}bT10Y(`&-m!yJyCJ3ta)r(886Of&P|;&`dZcUXMhf~j#f~-vWu%M<G+NFgUB6z
zTb`jmIdni)=-DECZ=P0Qy_%Q_H3N!luzDqB6hJ*Awawm3)^#^yl#JpMUr*a^6^7p+
zHE-?~bc0#~aR;T7nFsD@s)qZ$h+}E(2XQxbpkO>*sNFX;WJ?;Xjc!OQM5RpNGF3CO
zK-uA1??+l`3gqzFfKuf!amvsGMzbSEOC5!;<Xh@yjMR*}OJl3Z4oQjH`rg@!N0F>-
zLmUq&@S%t{@p&d=)5o#fvRdQ0;A5Z_m&`kR1dj9Eu|HOnK?%Vc`3I`~U8fk%6Z0OV
zVCWoZ*98a2%o!IMt98%1Vj3m}PP~nng?RlnU&tR5d?#C7_l4qP&~u6HIgAZT64CoO
zoy&(T(=KP_;K?^fGT((o_aQws`xCRdsMh~L@0p?n@#jtQ$GvTQSbHyxX^j#x!N&@v
z^p#&KYr<rQ-==uaZWyP->dBfTA&d0Kle7N!Cd+W~2J<X-3~Vui_~K`saT%_(NvT<X
ziWvpyFjdFGw>8&mcj)?Zzw)(!bAn(Y{F|Y5J0f}!Wj4Z8F|q~qrb~pRo<wv|P7?S<
zDoa^J^qH{?=s?F2_&yVVG_IVAAs$Uo;^&^rKJ0)1y`1P(khInul7p5y6y?kpY`rpv
zdH)}4{^v5D`?;AG9%i-;FJriaDY2TST{+)v8O5y*4xjs%Ke@N_pM~(hNypnscilQw
zfPO5K?toHryq-R6q=m_~J{Hx>Pz+D!UFBCrEhF(9<}5`Rz6|8Cv6!;yFC8@NHq7)K
ztp6lKFN>_y*+||kE*&G?k>#TjaCztVQ7YTRF@Mpu^!D0tpCgaWIdC$AJ?D8(YBWeq
zpYj@BQZj*US`6W1!}AV?@dX0?mW1&LHYAMyE3V(i!x@9j-m66=3jPoAeHD}Y)8fND
z@eI@`8V1r><m63w&hg>?)Z;M*gk*@(Jx@mk2dZcMu;p3Nwx>JARGX<Ee@qx91W#HL
zV|e8e;Ou<Rjf9i{7e|AQ?g(dECJ_*~Mx78wI^qBV>(WU8XYT;k_~<4;Xfwk{?caD$
zzBY(5!TkCT0@2%PM0n-{>;uAKM(4mme&q!C&Dt9T{v5kn;1he(h(m(Bm*l~^TWMXJ
zHQ7;ds-Q|Qk>czoJ(TILtx^^_ql{$F6gAZK46-rMmnXZv*%3=9kR&(82}N;hS~>M1
z@8PAe58oSHYi%IA!gaMw(be-%$f&)DK*4Y{z02(U2Ta-!vs4(dsJpG8?9M3ij9s~s
zuz{}?CKS6}v&?UPXkO*9;MnzZH1|C;RL`3=!K&c~!I8CLxu{*;^>UyE8!wY`xMHSP
z@b~L3_P7s?0P9%gi-)D~s1qBvaSelvbAa^rx-Fp%o0@NSGf0do0o<SvnJVg$pzhRs
zT6g#m<d*5&R5CKWru*b8lkdFhDUBOS(U`hP8F!9M?!wu=0;lc3>1}QrA;G9vCo0N(
z`SQ+Z*}d9Hmi%bjFLmD!1UeNM7?k{IX3({${Oh-ku6_-iPr<GO3N$JRw-+S#GXn|f
zrVZh&;OrZ+{kvgrT+ny%20QVU*abX~41P~u=#VXjxfyREPDLVZ=$FJQ6-w<91~cJC
zgoQB62;PNQUcjaA3{!AiaNo6lV6{Z4@VvZP)RgAPAmY2F#S~^|v372()f<_q6!A_7
zAXrma;dDETLdu;|xll*Tmk?j+aS3=X)gnmTUAcOZ{(g_cO<H%|m&zBqtyS*5lJ-!)
z_23vXVPiz($?aU8Dft6<i%Xb5C1!={{F@7Hb_eKebma<M<|-%Z3Ej!CI+~zY#u#Wd
zx6^l17eUH%$iAxn6nFzngTs(`@MdVuCDS-YUfh?q%1aI6x^06SU{*zN*JCH-QY*Op
zLuM=FIWVp7++Afqh^@j0wdY%4m*+nr5}v|vS*xhKAKw+{e*yLpUvv*2oDM<VU)CAo
z&|gsw=NyMln-VXkwk!+tM2SvSMZ#mPX``LsKdD(5oH3Pi=(cm%hOxN}>A!R&`g(ID
zJkcrIOSs~fUhJHY`IQ$Khmlr9`OJ}+D&&?7a`~~I%e&WgwYog$oRQpCdUXXwW>6=}
zU9WKdOg{>OQ9THZh6moxF13cW$x6-Hd&5j_c8^F0oMN{RJw(iQYhg5O$6tOo(>w}z
zcSv4;YK!0zb*QJ^`qXoGE2;g$f|`lL|GCWz95GD?Y_PSX#G(qDvPXd$*mmI+2_qyg
zIWyKM`hg6TMVk<$uH`{u+C$L`oj4sk!!3&_9i9(u$5))jXif<e6k6U^4W~USg(DV8
z3%@k)1J$)$pZJ88Uf`SegbJ+;XS0UG!;te~>@j$|39?!sP|R7RrrE*M#W%mGO-${P
zLj~hBdIo|3ce`Ed=h-ZXc78K2t-Lq#$~c@`lrNwx3FB<V{JyUBC!&GtaR~`SS!G)0
zgE5jrNV%-q<8iId%hlD@$_Gz+;%5VjXEHh;Yx}T$v`iu9*<({j7-}Pv!X;k?;kGoR
z+mGi?6^`XX#Yt`xM;&yXP6~jC<L((Wi@7+cS7apne8J7(q8@tFTgdz<(KP)cgRDP~
zAkF6=d^}Ate9L%!nq0{~Nf%|$Uw*_CY~NPyVAi5+^!&1-t;yn$S6bgn;yU-egh$p@
ziKx-q{FKu#J_KJgDJ3nhMjV9{!w47Z`<uZjw`HD>;^52d6o{eCR`-`6RKkq6ty3D+
zu7Z-p`FHLZ?@VCWUJ+N4S1y(WUY*UmvZaujSdi0Moi?6weZWD!l&v&9O2FKN-VZ>?
zhLq$BRk0mA5*p?i<mp=ug`v8+78LB}<{USDzpe3u)=0@^I29pAM#lxV`6Ptfd;oR3
z@X6&Krn|%OBh3zHDU&86eG=fec-I!Si1OI5#I|TQO~waJRI$%6glDYTwX{((W8b?%
z%LNyKT&PCs8=h^b=L$mWRCajAsSlOWp{0%YTupDhlWd&?pGW;)^ye3ll?x#0EjNwB
zyW%VmO`8mlg_iHGXZ+o2C8$tU<&7X*T04iq8qKZy3H6UbKTIszfQ;^;V4}-gm8eOz
zbv2$Xc7r)Pr(mb64KzZSlTz#;R<k<wDN$96a$~;YUOvP}=;}$n-wWJ<7JG|cH55(n
zyWo-RjoOpRJJ4)cgwl(xf(e%;UbpkMKVdz<H<XE%F~1pU&txMDxrjDL?OHSEYDJ*@
zZtxkBE!KNhIf(g)tA)l%8Eo5O4K0e-T+3Cj>k{&_sSqYtzn#^b;q45;gNb}e8Wq!r
zN0Jg14(48d4sC0J{U<Rt5VS>k=tyyI&v#k%#`YqO7!W)DSGlyyMkt`&6<eh<*cU_3
zMF#;Ta+E|2jO%#w+jv8?U0^S>_V{*j(_VT;1pV>@pu3qw5yD|PUs+7}{V=evU_m=&
zVBrgreyqBRjroi_2Tq6ucB|tD)?`GWso<heir7*~foBqthPm-+gKYt6;iFzTCd&9@
z&N-d*0017mbY!ArYS9S(q{Jl@N(N;-^dQ@w=xpBJ&`D&3<Y(Wa6P%s#$Hnw_hs7+c
zO7P^L*o+b)m~$HQg(0m*&Q!d0mOrte8R-djB77fC63sDva0z9?^sd`4i9L&nCi4?4
z^h4;&&0*Zn4{1>TtkS>Q>eET!Gp-nX)HiM9iGeO)%~$5rQV^u}j)KQ$T}F&xhLeP=
z0wZjQV?uqDV(aOukw(CAB4KSRNvM<8NjyzBUH$VCN0A?2zy!xsaXgczy!Go6EXVQu
z(rRvf!>gKK>2lk>nQn3mRv2Y;!$Lr!vna5+{0LRA(z3bqQq?vMV}PRAr?u9HjhGuf
zbB9&oO8c-0Q8pe6Zc~glJ^l#53fb7FOr~3rN&li0l8|Wi#hIf$g{$HEAevcpxKhQs
zIR_k-Y@fP=_VGb_*h=}<Q$46#bW%%m7U}I7w7IF}OR(ZIeJoyFwJJrhb6MC<Lz>5Q
zQD@kxinZjvfK#ErD8VS&Ov<WbO$IL8=DXp6vuTOtvxA(GuY+c&u9OEzC;|kKgt{7z
zy#i+8Q4;aXo|hyV&nAYG*~jIEIIEtF=G%~=l=-f)2j#(UC17)z@9V4IGJr>0dQEOf
zWfNi|#@F&0q6zMNZ^*wsCF4HbJHJ_N&bMP`V!Ic}n0B>#@?c1r2y~4@7xOR&ruTo-
zFVP;Srn|msZ;>29bMMZ*)5#<%%z7n6w)>=;h#r3YNw}U?<ITJhU$5g9mAkGc2Tv%6
z>&1Aycz!MIJB``7r|y0T-O^y<+n^YepXgQdUHpTg3>xb#t^UC4HDo+~P7$q@`80i-
zLZ(=?B=1|@*`Z>E;br&CE6*365u`EU`w!o1w35)R5L#*c;X&vWpQ)6;Y1`rLM|}{7
zzDJcSakH)n(!|Nt<WXFY^FY1R-;g#yc^#dbE*=>j8Jg1E7H@rhd-p)73H$Hj5<MW>
z;!ij9Pa)~Iw)j&P`k7qzuM|4rFT&D7!hcfe{)l$}dl8h15s-E92ZatmS^ED+q5G#2
z3UHkNQwjYK3_5xM#pz$o@brMxkw4Gwzw&ys076gxP6>67RkKX~{Lydy5qTK^YyhUK
zs5lmgBU%rnWJ2wgUH$@OiV7f|J#@ac4X5cg7bOXa#e}CWvKexIvS@3u_@N@}iKev%
zx;1KdTRHl@-DwEpe9u5-zu%|u$xs(cK1I42{BmjTvF~Qfp;EJ~xa!TsXI`<52W1cI
z_B^9;F?N<ouYkZ}eN_7GvVU9r>A=%<Nly_K>@|52`(EW3oPW-B@!b684cLzhIXwl+
zany0zL0^L_@PZlw==rN6a>x@c>q!f|V(_HOeO*a#oLp}|?g})4Y>czCjpwzG`D{cB
zn|&)=NwN8<B(aVjMZCs(TFO!R+CeUrgqjkE^>dIAW;Gl0=o*ZRz=mXg7Ok3V?dhbg
z1Fa6@f)?ss&53mS?Pk)ohe>E4R=~%UdB4LXN)~ga(@>U5Nr*h1sHOBm!=kP7>vt+R
zY8kMsZyaIc!jI)8*wj$_cUicB-G$kH1cIvU!7<sSe0m^JfrQ_&iEJmhj3ATnq`hUN
zBW0w=K9|(LD~&+)k=yao3#<kK(GLdEw}Fuh^aU@37xgw7<Sd9Q`8<muFiS|T3rY^$
z`<}DIRU-8s_lS_!5axU?kAr}Sm7hn~NruoVa55L|6H-u-3IO(UVBZ3iIUfunj4*`Y
z6e$0ZlN5ndL_!4U19K-lrx60Dz=1f>2Y8Cl_P$4u3M4IvMNE*ZlC*|ULyV9^2qRJ6
z^paah8Gwg`+&rNA2=W;PFt11?=<*q~h)E!+QOFtca?nl5U*nLPibW4Bx}fZQkCMVf
z>_Ache@~d63KvZgu!s)I{0StxS_l&;VLbuAc>78641YMNx8b?G_D2uz9PQqcrI(aR
z(+252sBX%qL{9W0)t9MTpuq*>TO>}HXOKc@gm<H-<k$pL?bXdyT==iLFP1-6nxER(
zL$8@nXgkKmHnspKHKA9nl)E~oXz2mB=B!yV!1N6N`!WDf%Kp4h{#*tA=RAx5bcOgk
zo%wHo7QgS<f7O})*FcN^bTa=zRQ|mP{t}gc3%~zwh!#KFJAQx4YRvSkKUo$(zlfix
z4tfA{?vIf_(w6}7J&XWt7a%qNB;qjxcq7b!9+2NmyWc5HOn^NA8$izdiNK%-jIsde
zdB6Ag&CdHZ3P67Xv`&CT{F9x-^v4FE>-~b<{gPGzL_EMAfEAhn@OpZ{8`A@h$Ot$d
zJs@m}h3;ok5g;82pt1gZ{vQcUY^(rb_9qtaH+t_EMejFy4{+o^-{!aK3K(SsD6c;^
z0Fmz}qvPjv0TS~scn32ecL|`{G69mF7=E2H9iV>&Fr$CY^>_Xf3t$!ufPH=v`hMa5
z0OKrx|NTDWKQfpAGX{(UMt;Uh{e%S40mSQ{pC91E`aMSfk7EPI0L?#u{Q!UelD$L^
zke~m&75+CmFyJKq0{{6F8)6Q~UP2O{JJ&uqlNI7aOYW#S{B-UELHUU{5I+Db8zLAS
z62d$58!j$PK$bU&FtCmQAPcN-105+FL9nv`f{$40GqqsQmy6GmjKt;r-Y0Wk+g4xh
z>NT#956jB#n~vUI&kN!{=H>Yc;6Hv{%Wqrbt~T3$0QK)DkxpjSLX1T~P1AGRp@Ksd
zhOm*s5(>l-6BrZN$;}!<a>tENi%Py5<>P@NC=N)pz~>=^5WRGLFym|gfOiG|Fq9?F
zJ+$!h;n9fCfKMdwL!h<*%_T~4E|JaP#M_IKFb8<R;3vqS;%J?$tXgtm@BmR9^zmth
zGY4|}hpGkNnw<qSMJKl$G4uSZYWYz~k?vNyB5TOYA_z%4rxhE;E}K~($p~^HmFSp~
zU@<WfQHf8M`KlkWqf5lZMS>MK-v}jBh&fdd3&n!MX9x>xCNW~~yE;J0%YB_nO{7NX
zCEcoon5?tJVoN3qvTBzE3|(E##bPx0<YHcIIEfdz&sH<^T(egMXx<v3iN`K)6PQ@6
z;!>G4idEAha)M`!x^=grPG3!fJXgmgMO4v0?V}`q%)I<Q(_^YzQw-@=ftI`31EE7m
z=T$*OjNstQwJ9P&#7&$MXzUBUDK<j50^!)@b3teTuj$LO$zKq(=!*mYd4ymO5jxPX
z2OmEWf*=k7pzCJh#UcLEb$j3`f&}n`MDXN-W(9nZl0k6fg9?SV6LfBKqqGrtuLcNv
zaz=K%X9S)SZ(?rh3S$q+r*-wno&C)So5{$S{KRtM<WFZTYeF?eJ%!sNABTeVn0zP!
z;_DKlQGZ4{_y_b9?uLHKxs$*}*v3Rog;&Q$3kW(wH<8H*YFnIZtB0kP4cR!T&f6^M
zq#oKOs95xw94^11r`a9zB4BUPf=qT<S~f5-G=<Mzpg%;DLNKfCIyBxd6A$i61)jtO
zk11Kjc^s~>OJTPV(<LmQi7L&huekfQs=^oTbsIA>U`A$+4Mpfuf4JizEwI>Pw1X>1
zP{1Q2k5x;q!RPo6tLk!L2Te&$xjPDLv9-_so<(%koVF2mus&doVd6tVuYyonL~GFe
zm|6`{uplBlBqeJXOoi|Lw^ns*ROP;NHTZgSCJvJ+)*jrw&4ziZuXFE?($p8irY6jG
z0<mOBq29ahK)iacH2yd>enlkJKk?z*A5+WmygT7|sXeYuyGv>8)F=o5W?xXXeB3gu
z@U%M5cm9Bc3QG_1^%EOo@tqY-yGfa)BDc^~vk_r{uLl!W7Lq9K=6!!FXPPn#!}v7#
zB^CyVnl@=}!}XE#0A(B2sy-(PI4)Vh$>zJzTqNT5x2-owOX^R35!_Rqd5ptn)bSPr
zr%Hsun0Sd^cx|}Y=?9heku2a$)GsEgna5>$bFp*&wUzCwZ^&j&BosXoD$I@!KOB42
z*aGv19nv{v`<d<hv+wERva4@&@-?sc%N2g?VpZpw>n{!6Bv2;9(oHtXmMpbrF{gZJ
zjGJ1@Gvk?1)##s1GcTB^><`~H;~c_Sc4EzTyG;w2vu@ULEF+CF57&^b@U~CEIOK>=
z-^Sjil{NmdIkD6GsM(JC7|7*ZmGJ=Lnf36{ru(e7w-rUn#W^J|rQ^_`M6aegfZQ3P
z^js|utd?e<=5i%GeL0eV(N>el`3bFl*hi2c51);mA4NI>q$H`Gg2OJ1CBV`@-+QZV
zXV;zVVVB`PPl7BTG19vvYXlaZuXRC{#`#qM>sr9X$|oNNBy;nEWoInY#w^qZYDJsX
z4dmL1gJQ=khnx|{hq}%i8x$)67vb@hXF(c+*YQxU1s=E*#spPC@(cLb8vJlrA~z7M
zpv3+FYD?C&(e(Zo%!gh^7#2xXo9a1h*&sS}G@)>li0%nj!=y1Bt=c2#wFVb#iQ<4x
z_Q4=#QsWM^&p@`6<pfn;rcmjv;@$X(tx(6c9&|u!?N;Vrc9}ee!}ShQzDZ7&OrOm4
zvvJ1EFcbtnd5rfCOPb|+jG|>PJuj$uPVhQZ@1WQ1tN8dDpQ+89uR`;5+_T?s?I=xD
z;KefrVColU=@Wn)5|%gT;9}%RKx-tRH}iK(S=r(YOD7%9uNf{9se0GI%i?BR9ywr9
zzp5(R`ZJ)4%h`qBZWQ<ox(suq@{zbr6gyD}eMSRD38M)pf0Ix~>EeaJn#hiKuS&LC
zzC0GxsxZXZiXfuD54_Pjd71q#!cG27VJ82iaTAIV`n>%Kloq`gffL|;I!Gb*dQy|4
z!&XOIRi1PP`Y!gABFcm%K+WR?&85+wAvklZ6*kufbpJyYlQIw=!xv6NSi{E?k<$3e
zvVwD}uH;l*O@4W;?i;for}zj;Y2pZ`UQ_tT<G%P34Na7;rRMRJm>T!TO%ZM`<L)64
z{jM?J)x#8>?vY*!`gmf}Rjr5j-DFvaUe3;~*iJj@AN!W&kNN$?pX<$OmO9J16(SWo
zxh>2+l`A)``tZ1$MMyu!1ZT1dqWVo?$MW65!#yt0kMS<)i}QuoSO@FFIhwR{y1%No
zV9xuzcKSB<H%d5i$x%TxNu7}?g4NJ=u$f`*zS}=T)8MMXw6(}Yo|!N==E=mnMkV$a
zvc7oHDuZL2`LL^j-42m<iB*&}nu!X8Nkn8N0+o*&VJr~v!)pJqO2LPHR0N8)Y93`c
ziJ#g-`ZLBwYussTm`3n1?W`D#)gg2SGF*5MSl;|x95d>C&H+rzB$jl}mI&H-pyBCL
zTjo5(-GG^_)nlcTMJhD9y?p&H=AHdMa)fq1r*GHoR_XiJaTkqH7hqsCS*t^eJogt{
zdW0ykyL>BErbHQAZE?0@%a$JrUqTrY67II{mzH$qVO?1kQMu&02`TK9?7aD{0<olT
zBi-|IjE*MC&%Ep~V!i@{7U1M3w>Sdbv(xI9Oz1p78FWcp5kUE6L&(8cpMI(Df_Rcb
zmEaR~a!Q>rqs4y5Pg!dMmo4i+T}UT`hGg;%DZ8wBfg%s%WhHP%t=LOy_=}@ov(>}d
zxFhm7EXgH5VSXe6m)dUiN9)A&#T{A9F-kAIL(~XZ&RR|PFo7$NTa8MZbh(pNWINdJ
zG@>2WyS}&cIMLXBXx9dgyY^O;rFDbn$KAtan6%A1VUt@AEHi@<H#FIvSO`e*Bd6;r
zORPyRgieEERO|HFmhjBnFF1W~<!VoF2{%%?H$1G28Q#x0<VLnW5mS`h^WsM?y&0N?
z+MhTL0mb0}vH{roOCzXYzDU~l`TdC-VB6L%-*?O$8Aw0-t#w~6O8Ci}zZ(IYUia!6
zRq&hXi}3P!(M{bkVoB*k6!j4R<3<bVl)(*|4E)SI@HEPb282@X2}&_$OyZ+g4ItcJ
z7{5aRtJRuWZ$bAp%Eri)9Wk2GW;Zz(71SygVpMQT7Atl)^D~+Y=T@h1bM`xb#)7?N
zp#IdD+hvBmGB4iry;ALoT5D+k+o3N=eJONn12pwZ+-gN)SXqf^W6$f;1yMzd{3l(G
zVO<w9g~MJaxUB>C+}K!qGuP;Sfs@2+8JDBwP)3*nq_j*RgH2?;_xzi}M2y=*!q$)E
z`FRr?*3*#`s-*&$&5_U%IM#1*J`$7*Kz=Z0c0i~C+}+zWH1D9X4m=gI4D`H^pn=)2
zL4CB=fW9b~_K-<EP_(P8rqhEa_rnQ`w8A;esBwD&v3>9bhf3`1g7WL4lDpiS-AraV
zC)Xlal<rH)F3@@bxuH$;eiV_;3t2|HRp+_+6wucNr>^=<0_VW%<(5SvYJ*Xf#oz~l
zZ0AZ2*Ogl~K9C-lg?mU7uD^%v{d1aI3xvB^@`G(i&e<n7pU(9lALs}NbB23sUAQRT
zW^XAi`8kcoIooPtrW1tv{r8Wv`JLd{R%@@z4V9vT!{l)n`!DvU<7N+Jvrj*ol4yBy
z&)s;04d-=hZ5u)dcUDljl8aOxE)FRobW$2iFe4Te)(H!Z0vb={B7ywbD$fLNWU2QC
zhE6@2^<>jMRn<#NEa4sMm5dp}tQw?PPwMs5K#B!*LG`fR^z=nhJWT{2m0nyHFRAQR
zc*Sa~E4aVDzGjOz-k#~1b>EJtD&4E0CtLzQ!h|uM&3Tz;=M;fMeG=(KP9XcjGTWoR
zmlP*>4y(;%p_1moNh*Fr^Y-=r%U<s6mE=v^#(*~x%6C2`5kH+tJhZLbo2Pp4vD7^p
zxBK45<)J~PNWPw;+?Aml-=5wQ6ROZ3>;Z4hG7(x`_UZ=>wnaJ`B5d_MMYs+@QwbeP
z`E44>#D$2$V!a>vv8p+)ufL-&sw6IAx!mr<&>QMnIMmskmW}qh*2SJI^>r54GYoWX
zd7CfC?D;<~_<AynZ;;<Dg!k|y?s>S(*<*A;4|rtFf8L--xfaCW*en&2$r1H9?)gDw
zIL8=iT%5(Qy)bY~LVZ~QFT|83n_<NtEb4JDLx(NH-g>(%-WyO3*O0c>tif4Sj|9%s
zlWXX!$N`nXz;I2kqQ~cj7)r9fGOyuL2C{Vch0*LGrEz<!0RX0Xj?;Q-igdS|bxrC%
zaR$`avLGJuK1^ZCJ6nsjYS+-gb&oHwy`Q7gX^_WRgC!+-c0bvffnZKS(_XHR>D%1r
zLWWDgf&oHA8$}<ka*>U_JMSO{W6i9t8BP)K3?J1}vYt5g{WK*{k)>9)y5w4y1l_lZ
z>w&v0+CxkC?a@~?SmNp`>vS50CcVPGkurX0HTk+12yGVz<JcI<=?*kPy1=nP2i*re
zE|J%mD=%|v^tFo9WOz~GC8=05Lhn?lH|$<u>4dF;CErF-+9jKZvd$s%E349^BvVIM
zRh*B8pCa?40<Vj;UZ;N)EuLe45geiNy$o<Ng>Ib}hI)1r#>|KSlqA1O8o|tXB*jid
zU{*$xC4EpBmzQoIQ^p0@<!d|2X7CJj#ORy~+kUr};in*{+C2O+!8<xoJoop**n=I4
zhh6N<1TJcaVjj(?iC>f(8@;&{h`y<ea!iG=r|w9woODC<byDpu?gil0H&RgiAa`4o
z=9<nAxWY@{fw`CSAjdsQVPGlD(_O7UsZ9oNE`E?N75vZ?Bh`#%*(qMU*HVMGB!aRC
z@$srgv|!ae=<PCuVST3ZD*>(+M@v5|EPX-LMuEIvh+G7nj9*dcVBf=L7oXm)E_cuJ
zGFcxD{5X|7T;O|G?M*Lta8jvO+%ZjVk%;6&u=^y%Q?d~n(vk6Tnx#g0YRiSX>WT_n
zo_1VM*G#top7&#S9BThw*P;ic3jFzN{?EeJ-<P*#RYZkF1pg>+v;Ef=A1X#bWd;Bw
z21K7Q0xB4Q>h{m#$X_y6nEvYX`41T@fLHy!&j(Nu`A@6czX}uoQW*mzv;28-|5bIH
z2@o&wx7F<!weTeXju?=zvI0(chbL~~5Qog?{f=sCttq3(9i}ofvLndGd1HP4F7ZSR
z`BbdrA)#_S5q&wE#*)Jmuh{2m`}tdIaotfe-{YdKrfc?O5&2Wd)M|*F+#uc(cU65J
zxkHiewpsJiQB^a;pqfUCYoUM1vZ%N{`K@8_toC;6)1l{E?aPz-DG%4V&8pOkqy5pi
z^ESA{WZTIY@ngI52*~0ou5I!U8A&yzr%1nh-j4}5bE*cjy>as{h*bl`K8L!6`uyCB
zolSRLfrC{Cfh%uYI!}72=Vj*k)N1Y}EsLR&p*D*SqqEL!*Tm)D6(kF#&D@r=L>F)E
z-IeCxpj#eOMC}<ll3!bPP<x5%?n)UatD0Z4A=Ab0=2a^PtRb5JnXw|bHd2H8*Nl~7
zIN_feD_2es$nDpC7D&{iLc9^aw!(=7?}PbzfMh^|Khx0%p#=8go{$A9g=N9i2pIq^
z;UXgO^IAoG&U=>|2+zeDOlAv}r6Ixv8rkKOP53PyIq)diTO<|YmIzwL8x3CA4ANVg
z7bG3BNOD8|qinoLsvAP4xXx){w_(-$&U{?&aWsZ4t2iAYtTB9r;Y~_DK`c#txzAdx
zdn6h02)yCC9zZT+@JKo%yb(Ytkfc^4ykE)5;ETSCzTXc8jMe66#I$_IBeIuZ@M9Mi
zB%6`PX9qnHFxdaN$RBcjKGxhyemCZ+K9XgLVN0>vy>XV-YM-lwV=1gL>k#|gAZ+OL
zLfLB3!h=X_GVt%WAD}q+=LP!b-S<B)EB|MA=HFD60r%FwyO)05DgS!!{KMe$GsNR>
zr}h{35<q?WGuGulo6CPJD*sw~{wON{VOjzl^8a>_3O%6M`TH|gV`gNd2Be|<v#1Ox
z-TpE1yQs_ps6YRVSz!gJVt}&oKcZIX0gKMhjpe7G=~wK@U!zxkIh_D61eA<_=BNCu
zw*G&#ods|lOO~!}fh}fcW@ctai&?UmnVBuLn3-iUGcz+;%*@QvYx~a3o!Pmw8@mxN
zLLE`ns;bPY&aAHU^?%O)>-GCszp3vZ`1tpJfcqc-?E)NGei35^K<*9z5C2NRVW$6u
zSbvvlX8WC)^Ur=Qzj^H6j>^LNE0u{Ij|uQ0{(AkL76W<*pvAwA{>#4wu=tB1GXdc5
zFBts0h2LGF2Q>Rf7k|C~&z3H~f$`tA`Axom_xDeGmtUvD^oK3XAAtDxw*Pb~|9K$)
zpP(>3;Ntw_7W_|8SOr`aNu|C#r>J;7Dnd|$+DTU_2p!srih^GNZsn&Czpx@4Xto-N
z>7i9~AE?l_$VYh1PU7w?!X7d_t6;qx7=aHOa`-?rNk(AXXoy>qcE$FahQ;*fV;<Mi
zo^uV&NAs`eMei^BY3{b8^C&&G9xFLf@sNq*nHbeBt`;d)cQ=Mdz2UQ}Wp$K%E;=?_
zr>b4>h$I_8=7cr~{Dxs7i{fS{7C{Nw?h(DL5%ruQ;Ar1)-6odWpRfl{(t5qmMHXMX
zp0Xya2uSL>Y%H+$g4?>i;C0DYU~r$;jng-`_E^l@jT8!4*y+V9Qa1Nk>{IG)<;U#h
zKnMEsO~RYw43sR?TPc1D*8%>7^pP@2O4683i*Jwee$Il#lkg*#Yv&J77S(Ym<tWEi
ziL)*WBWJ{wDxfnjA;r7Q6)VhDaX7Y85K?QGDILbn7W*KN^eDZhFtt$-vCE_vD3lb)
zsuVex)lrC7tlvbJl<*Kv3Fk^XNW-m9H5Igbitn3Yv##<wh(Fv!Wj#VBQt2Gzte2_~
zo75J&)bX&eEA3z5W<D&RB7Aq!w>VmH8FO>|(#j#7Svt|%>kqG%;v?NxpgTj4I*Od?
z>lr-O-&z?-XFZCV8Z<V5^7ZW}fdzd%lEydcf$l0~Rgij!u&#OiOI4tR@cgfs^wb4C
zH>w3fd0Jd)^u{H?7Q$J%4+31M6`j`002PqFg5eBtGP!v4&7$%W44hLh7D4I{65DD2
z6_E_$jzXXx1V^CEWZ1N0iEoZbEM{*%Lglz~Jw>f?v|b|!(_+<><I(Y|#laZ1P-sDm
zD?tzy#1!eVq__xM@MC;LsxgHTynT9@C<t_^pb#0d$p%2A5Tz`7ta*K*jY8ml_(;ye
z@|=$2%c)@|)&yP$^JbX;IMF>iVtuf+MY3D-C&_OnQ&D-m(K=*z*d!?_xbsD{+?k1m
zvM`eHID;YKx^u9^d_dA3+c!us|K_w=H`G4vamXK`X1&L`o2gpuG=n`mADrZwiYBP}
zwkUQM7Dn4)=@SVo%h8KB*RA#jZwoN?ore<JwdUwx;&m%H0r%e8ZT2WO+l@xf-z>Ez
zi!KdOe5TOKnE1H-Eu&{4vx>4=<4&2L)M^L&l`_Z+Ipr`f91WFs-o{zBM=k#ax2%pC
z6(b_h@+&NiWgaUC^;%-r=;EZjci~kethHTTdn9*xznGzbnGEMm{WdcJbeQ?UDz?z@
z5pr@E`T}zLs}|}rpjxw&`|!Tz5Md4jQ(JEkzXI1}i&&^ZMHDm}j|mN@XJBlXV>_<Q
z%{K3iBp#Mm!Sm{Bk=M-~;PD5F?yg~+>&g1;3|SEaLsFC2Fu~s=c5*79!;8AHBSBTU
zTM|P#TYMt|NY%lM4Q)IY4h&HjgKC##f64?rMw%anDuvvkQwo8klPn+?PKIPQ498Q~
zPd5CH*@%qE{chMBX$vXgtE{4|mIjh$Qa_t3Cq8@3JzLx6juy2uMX^~^JW_a-R)vyv
zs2<9=^#}5ov%cMD)RKCl_mG-tJ)j9AJn2R_AdF_OiM9ukyX)S`*l4I2X_XLF{SFo1
zp7cFL@5i*|*Qj6(oEYSuQ3yA-%1HldBaHYc+2Pb;D>_GKXc|)CPF&&a6sgMnk8h`0
zLMHVaH~!fHn?C$>{6XN3o!T2ewzgLEL`W3Tq<~3mSbarXJkJrMpsxI{@bR(dr|9}`
zISmkV*9uk#1(+1;2}dRmV2<XC%nQpI64jY<)V7TZ%OKPhSCdAI1Vf0S)4{pJJ#g4s
zA3qx8AS34>uaR^u2lx2ngjB+|n$DbqS{p;e!?+^y+rYSbL$A#akYJP%Q&#xW#_cgn
z_R%3KF|31l!d|%hccazxUaf&*L=b4fu@R_Hm}}jTPDyQS*=JU4=V07orw>;Q<7%Nk
z7Mrs>M;4VYrfT&m8NM4{XWkBU*oyIFI51si-F(#(;0M~~VB|4@N5zUdV9{c4>p!*n
zfH^x~i1W1>2bcH0(~LIM(ieQSLF21dnCI^Ht=0lggQMr2Dh~=`C@5&UXXYT-sPy&j
z-9ADgLTu=C($Mvu3lTDDf_dYqM!o6m@Wb|!;9}{^T*O5j_7q3{d1Q8$kL~-+3)-6K
z^&{GR;0i{Mx172I8*S#ukMg%2sO#q3<Kr66_cYgU%Nm3G*Tb@@kHA#kz7>ljIm9z2
znRCr(%G%*&!6nTLp4+qW8mO_rP2O3$9R1e0q?E8Ln};f{yC2qX;@pK3dQ7~cl`;%J
zb?FLaOt^E8!C9m17Af-q&y{)mtz8FeThdie?y+M@4V|ax(wY_nswei)O6gA-(tAPl
zP3el*VAXna9ZXZcBk(AL>1GJ(#;u&53lMU%nIaG)+Ckq_MdYZQcO-zF(IunM$l?^|
zL$!X6p$;x0XRKas4ZhJRD5D;gm9_R&3}3`Ry5@T@1G!)3m!-5aK%>6M5d2&e+<1C)
zdzc?&uc`%?UDnZDlVIB6ZtoMz3E8m(6M8L^A~YPZZTLyD68`)c#W9a>$_C@5Ya+#q
zHH%Fk_zoTB3<UXh3*7~u8o#C$4gaIJ)vMPf7Uy!z{X3FNdAQV6m9brI1JBolz+=LB
zTkJJKN@0p0-nm=6Rz>gIx72w%>uA`%U*91E*a`#`s}}=QQnI@&pg%xE!^rsHgUYjo
zEE-_NyfeX@e0g~Q9$ZutH4&mwW!cw%=j8h^<(hk45)m)f9ZKg>618ZCfWE#&3r~Pt
zX3}~g`3V{8sE*a4gDIdn%9-$8q{!jm0|u_Z5Y?LA%3OUXv3a>a>bh8fPN2op{=H6i
z2o(h`3{HT$DqrY~{Q8H!Ch}Gwlw3Kgrw=q;pNZfi1-Q#71ni;V^!$uZA0W*WV!?v=
zME$kk#L^a?49m2JwePX&S$nJ~j*Pa)DfSBA_-!_f$kSedJnBuaN2V9Q@482(D|S;9
z=e%?u9*@S(H!%upI>(tl7^cFXGfQreD)OC=WL|M$tmHMHofxl^Z*o4sk;G+u%^jqE
ze#BnHsupDli{cGc&dL8yhK3atH%2kEFH86C=eJ5QuRxZid&mGL0)uBLS$%h{kwjMg
z^}V6yI6CzvPPQts2Ej9kM+h-o<LB{Ch{u~4{pY){Gb9k?>g@7LYs%1l%K}#=2WMi0
zZ#gPJ>yY{fGQ*6{Lr;xhhn;EJ@oiC3ALqRbX=NS8XVademgawcQ*jn-B$EYL(oWpz
z1eX~ijuWE@W5;~M4&hNOv^d#_WS%!HT>mPGgMj?81QG$AI%PQb()=w>J7!FYRBO$V
zQPj<HzbuW+aL~f*IcM@Jl+CFHw<`Z=M0z5*#8KFI+<u=j^td=K2QLCGprXYl=~kY=
z0zM;b(Q^n(m9*fdRqxY@I1{vVh94lI=!zIzr3)P4Nk`rA22)qsFI7maDsPgQJ*cMs
zY;1h3hv24TtvAEMi?w2uBTd2XWoBy&m?}%BH3YcN1D{rgoux3{I;dMY3iI@%JHa{>
zZ)SEwEqpVycng!%NrdDrmVv)LrQN4TEFx#fL&3l}57CVf&`6wB8eVmFf^DaF8XJqC
z3@3Npll9%yKnG#e(5&hj`x<nim<<R-9?mtJx1GyNN7U4II!fOQ4}Y(BugU6=p_4kg
zoQBq?^5jEHNc&9eCK3m9W2SOdb2^FL=jrKJ;F(%^Z}Yf4Ch)XN9XX1CF##2_^AD%f
zO)V0~YA!60w{na$79KW5=Nq?1Eb35+(1hf$4MGsQwXp2St4r|-vDLD?Jc%ow<lvRB
z!uI$xU5QfH7lKb(*Lh7l<w5rNp2xRZg%2;>q<PvzQWyx|2{rJyR0VfSfE~B5udbj2
z(S?^T4ZXBXSn2s9sIbc(M-YDSv4UWd1r&mGi`N2S3zTfa`L5w-BXJAQ39^7}5SeVE
zpiNQRz9!wybLh4J5%ZG>@j=kX3dxwTq=40cJ0r|zA=Tm-B6t(_<x8TC^6B=9DWR%M
zHzxJVgfUt8TRm~W=FW5JKdX$fqx6;DE}m7Tgk7ygUcl|H$ga{nSL4n(9;GCmTd$jz
z?G)ppK-H2N$4~ONW?I2>>eQGWe<EQynkA$*eqm^+6J+o7R}Jrq7UeH59c^~TAf$7O
z?S+Qh(_Mu!J%Kh9_(`v-<&+a{;NPR5bt-b{be<%Vx^6l=f_g{}=lxmY<OF9Q*&`)W
zg>HJO?OApN%Z2*i<8cOnWaqB}^1sGml2R(da)STx=lUPSo=pGn=K_S7|2wfKAcXx7
zIP5>Wb1?vfJ^w}Q32^`U^W^?p9LD$`#GY}oVOxA~edmuzZJ)c?4Nizi&_9ElC`P4w
za}J+r>o+?9FM+26tVY()ywcIpR*;dr=SzxJg!`r%8qz!9Zhc=4GHRU7j<w@=_9uk5
zzl)SKpU@Jupc3%w1=mNONYCv?aIlOwl$4b=b5gGjU#&o5C^iU7awuIX=si_lW_UfX
zXL^-+@4U@imq(rYzsnT9X_958=6TMg*FAJ_6!@PwMJ*EWiq|Po<Eq(AurKhI<_{i+
zp;z3Jj1#0n6-34(4L=%u-d=(fAStf4H{cx!TZ<K6(^`(8(QqwodrXsd*e^bvopbwz
z!X%=KcS62fYs|)Ituq-jIHBEFhl6t*4ZHEu_67dz9c@SEP<LGLrrKmQ^tKEROaT5#
zZO(705SItxN?={C(#goOexR}!jij@Mr4~f^1@c?Plm%G4H-sSssH9f#wP}>vFg0<6
zazb~nJeVPZZ&o-8Ug0Re`D~gW%OJ2YKG`SYFFArYJ}ERTz)h{7apZ%pwmTeLswvGd
zj9&t{9%d6PM&X-DC8J83fOK;VEXiT`FyT0ae#(c#(0|SrJR<AY-k=icfE(A7!psru
zjz;yH#P2IdoDHY5+$!#r^rO!IDes3)paRO6_tW^uEJxx<vq$YnlhZDBAKhwTEtH+{
zy43^Aagj2gc8N#v&T6brIff^rrNBxm!E^bu)5AX9P4nd&(5Sai@4wF@0Lu7tr2Qp4
z`t>G&8vRce!T)KB{T0{xgCYHERi58N{$IJ)-<5>_Yp(U5`tomF>krE@fI%?BZ^HEd
zNh}EvkN+#Z0`PX`-yXn>zsL?VpoGldA=WQ-;~z}x7b0Q>hzS4o-(RSQ1t6DX2C#Dg
z!1~Ry0E>*jOUV2ofdsHFfUEFt3FPl1vi!zI0Q|!M;9Y;`SAbP!KzW(Jz4dDgHUKR9
zv!OqfkH6Xl9PZbP6;MCuZ~Dexi;MuTVirKdzbeoCLQ}u>iGcHA0xbPn|Ba`9?Z)!=
zuK%HD{8gido{o_nAnp7^*!Ua4{gzSw;dS<3i`W2;|L4H`KXEEHz<vBLIMsJma7Dz0
zwM-_1PsmdU17@F$^~|+0AkNe+J>x_B%5uKMLl6l31eX*50z<$HAYu)u{PqEY|6?{#
zd97bd*~El$M}6z|RQ*80Jk8Si?c&lMZ|lh6BMHdTl}v}>_z=VW+G9n>`@8dbFc>VX
z01?7ei$DXY6?9UEp}AO`e4Wcwx6kYj+CpzhYg%SVtLj?fhH<dV9eD=gwF1|L1?{CL
zZDh9RgC8yX#&LGsC?1&H5YfW@YoV^+LQOYsCR2r4xi6>o1mdhhKyVf~`eoEZ;DZto
ziu42HcgYtNYV|t_Gj%5s0nJwRc7v&$d`Qi8+Cku#oz$G1Soyl*JP_A729fq5+4ecv
zH}zs=H0h`bX{F(N%B|cO{oSIft1o?4t_2d&*)ao#c}4|Adyhkc+lY#GmS!R`nh#@N
zh$wRuL(yCd7(}A+RR$<`D%x=nDAuEj6-OF~gec=E0zEbbnl$sj3~tXI)y}=W9gVVe
z%!aeQtLlCq=3Q(P+dX%)+Wcl&5lFLAeHq`mAOQ3_t7hhEVfKwnBs)SvO!@J<t<8;*
zmei3~Uq4!)e0tV6NkACBN63EgV=qv5<OL}TqGT{~uY?}-rHURlJf8)gHnCsuSTBkm
zCA^dcsS_N&5K&cRFu@(-r4UfG5D_A|5Jt3&0i2ltA&MTgrhQ0L`44wE^{vlxc?Ww6
zGno?JN9ML!55+-|qD4gfc9IN2XG9eW1ZMD#Q}*ACGxp1Dr)@*F%fIVzuhHqQ1xRYg
zB^qrthD;&Sx=6OsO1s(0ukX~^thm#)i34AMegG>ou@R)S6wLpojxul=NzC0Ok!P3?
zqfWe(II`K8J%G@|Z@@4(9YOGUSsLevuMm5M98NMT{UnVxq(Q~xIf*iFo6Kyt5<|~4
zP2f(@V^yP$l(MBzLo5{{nPM(utyN(GHqts|yi-vXxLr?sQB)8SRPq8yQq@IO+zhUS
zoC(d5!Sm5c5?;fi3Y`KiRjIMD!JUTf-M>jb6K%9l;j(c&U>`(V<KA#T{Rq~LojP98
zb_2@Yk4sH!xleaJ)3gJ05G{V*^9r|9sa8$Z`b~@f+TUw%Hq^?8_QZzVIsn)K*AJWR
zA+R#oPXOiW2kY(a^P(yPjSlJvG<VLH*_CW%;m|iHhvDwP^p@ZwbTK|SH~vZ8f`x3*
zpM(_KKLiH^unj9Lt$7BxMBZra*E}-La8^1<IX2}HZGFqh*^P@0%L<PC5D+QIzS`S&
z-vbPtf^a(Si%x5B*Q~jtrpFFX@_&RE(Ueit*yP90Em%}jc&f<|AFss9V-Ie)dig#e
zx4LZ@CapiQO72^hmyPQcCAjuDf=qN;xrp3=auI$aei2yDS07DqaH3AOVkxvYFLDTL
z3T%_CSEZuHdh6ZxPQAl95dePK{5Z8h2kG@qZL5pPMabLZqC<d3ufr4FeUX*EQP;vH
z%NJyOpAOT>-eq2qYHD-e1}Mz*jZO+5zX{@57xoPH@}7M&f;R(ZCZG3`*punqgy}5Q
z&;Jr@UY^^%hav_YE5zmi6CSJs#@c4O#n}C1->p$ql)o@4CR%7c-~LOs{0PqAm(;7|
z#e-B%=Sjf}wt4j|TZnNfVLrYc!JjumHYO2C?*)-gUp&p?2KBrVGc05HkY*LIp^*z^
zu}V>x7JBBdiFPoFwZ?u#mN=v`GwJtkTCc!dL#`~fJxX*H1}1hjtQP{CFb2PW)H~GF
z2CZY7T>f;yrEJzSqOBQ@x}g*~-?<fg{xf2ID~%ahxuvw*WmI#>DK3dRFdPz*!^-)B
z#S?*W)sOfQc8$@_CeDJ#%r&)V3ql<uTwl!e3fg=-oi<7KsJ_DhInMt}TOv;PIQ)VY
zluV8&bOzh}Mwpd*Hk94_hB-~ZZB!P$1cz3;TaP~pG;bZQ(GBU%GgRjdvVhah%Zxw4
z6P9DIh4_!J`7I;y7;z(Dbc0>qoAX~%T#!%~Lej9_QeuvRl+y9?`K=7QznI{fo{Oxm
z+LC21LlabkrhC1Q^>>$wC{xfDSEg7<Q5Tn^Y997IKkH%zfAMzXwK%>%^d$$%DEQ$O
zg|pB3;OdaP8C;aYE$=5$le|;uQRjCIwysnk8PY8Y?}gt?a$-P`h!Um;Rb6`3TfoH9
z;G%M8q-M8To+`oen!dcJy~h!N=YoQ=nGD9bg2O-qyLjCgQ=Qg~q86AxD?cuAUU;u9
zDm|*E8oCka%-8~9-uR6skFzIvm_kaqU|LzbYcp=q;Rihl-K5Lak<xJyEN?A^z#^$2
z5_W)i91^uC?wwPX5ek=h(ZNiOt1Zn6q-pa)<hST|){l?0eiaVf)<{p<oW$Jk`a8+{
z`$86CWBOy05^t{ewgZgmPg|>uiBH9IEov)!R?dV9YZq#<MmDj^H{X4L$)4IWoFd~L
zY#m(VqYm5DFMx*kAx69G@Ml=R7Ji+o(0{^xi^mu`ys)1ZMt)Ixy7+OhDL<AIS6_<5
zf3<iec<NVUX(h<xW2@KxQ=Cvy3Ga)<2LoSRF4C$6)w1|%66#_7hS)_lv0N06PYL21
zoDE;bRNY~lZh5kgupJ$5VUj`@WJWCDAc77Fv1L9wt-WyHB)v-<)A&U2G>4k8;C}yV
zYZ?;3`&>xx<NbyYMx$We=}<LtzwSPnbW7Oo4hu2YKha8UXLvz;Z#-PwwYEvsY}d%c
zWUG($yy}ZQPewusL439wGq*(KSzh?)$5{eSXvh7=U81wf=GtB%QZo}j_r4!DnYY^S
z4IS$OFNFYm_OR;9<ZSdU%?P65@-${91<7~hwZD1p2&I>k8mSh#PBJHTNsjRP)I<lf
z@5I^%8UyrXnZ`JC^?9Mz|K$B}@jZB<t9(@Uq2)f{=fox{L=om5S}>T8V<Sw%=WV!F
zbZd9q&jO6IEB^U#eOv6eHwXdSHgszde3!S0K%WKaEJR%f^n91yLF+vyX^-GarS0gg
zunndw&Bi`xhb)w8`6X`01AnYUL-C%1Em1u&FVqnV+ezZn@w7wbegFlly&+CSnw`jf
z+c?HG@dUw0$8Q@cbeLJK5*zq*wH^~RRV8K_Le85x#*-DCZhkc?vp~7s{Dqp7dw$KZ
zgsUHVu3ZS*p-yN4FOA4tCd;a85e`2|8rQ$4ugE^@Bikk3nDHc?_uBK%$i8`AYI!k#
zP$ViZ7f2)9q*RQ&<#0o)yHB5`2sV?KEPV<6))AkEeIohh#Ws$`JH=6<#w`jj)eoXq
zO5Gh0NJj9W3bWd%=3}Qb`%R88rxaoJAe1*}5aM$`u_qrJ=mVSWUwH5&z>K?N%j#V!
z4h9qvxv&UZAa!n%EBft3x7fI3<hmD$C`PAaeDMjfnSW`=L=xBMr<WJ03~{@;EJ`&z
z!IvaAiw4I^04HdF6y^Up^94*c?9m=h#($uAVvnijIz2{ed4A8?Ak$lW$@R3mi+g`@
z#VxEUoV-EpxKwi$HD+2u(JWpmg&MuiV&UoQvXv3|-NJ_Dv8)i*ydA3ZEy~JByVInp
zjTv_dWC|0+V)t%;Z~R99(W<)hK%~$Oh3?lROl*<TqYxm;mAWPxH_~Hns7O*rdl>0i
z-k@!4QSY2r!3~k|fF0>Ty|zmd*O5FNE;r6?*&fNI;1(>&rIZ~S$Ex6RY~duV38en<
zcCV%RDic&5UKS!#oyX{21+}@qW>ZjG+*qEc@Z{`}F#v@MNocr9|CBTswPe0oI`*{|
zg%Px%TaVFrcw}7;Z=^j}z+!xk@rQ2{7>^-cBIa82btVsd7rV+M2(;U)$ZO)NFKx~p
z7QZExkSej@s(w_TvjUt|TrIY3G!L(eb43AefaNHo`cxvF(rDyyepx~27D@foR|;AK
zC?k7YdX+rMk#If5Cg=vXN$xB!!dv3|)kPU6O0E0?hbo}Amgscz-rb#(Gf!N_w!kCW
zD(HG=SnPx*i4)ZJmrNeIk2_6xvLBmf2T|{)oQiR<=M}3ApNO(m+RPF*kcp)H3yG(P
z9P3CqkFYYPA4*ipwY~6w&-3{<*e^Mu=-qz`cnWDcO$Lt5gbtSxBTFfn2Cfc^`}3Nd
za6RJ?GOxMYznZ5a^Dnd53y3a~l9TZjV5v=k!FsSRg;^v6`)}6?fRFJDxQ?Tytgdz$
zMoC|ftipfCIV_|savhfbiRn(F^B%pc^1V9jCJFOkgbeF*$-O_47}!BT+$Q1xw3n|k
z)3jh;5B1Fgy@2h<I(n~xrzpJvLzMCFGEGDGAjICRyW+>92iMvxo}-ujsjM)FjXwfm
z;jhU(KCCAA+JWAMPwy6>*x6ssHQ6Xl=PhvF(@`+60_S5y4<!tA1>lDe4Nv*P(0e;$
zZ;o>W;A<09v1=OcyjBoe$v-B#+TeRqN#mJ97fB08Kdy}8vsgRp>gnz{Fq)bO=NHq`
zcTJ=^Nfg$RCx6c~a+#8UG40pXa*>NU^r<X8h&dWbOC8xfpK-g2CkqqDyPl3hz!yyZ
zIzX|>g|0M1Vo`4uEp2&_mi`=Kd{nMjt8@=;Ec|?T6)u?Sg_goLR#y;x=v>CGLC8gt
zCN+XJ_|0I=zv&DN3w?zOC)`Q6GwFLw@`cBCIhPglncBxh;&DAsdyv=?kt42hI#z^=
zuz~G)zWds7AgwVk5lJJ{fKD4;Sm|h?+4`(pDa=I;HRtAU=I!g$ii^ZyhU$+I4Hq#>
zHWQ5)8uOrgyh9mg?#8$0o4noUNHCQ$&DD8QI(=Knm53QZ&@pD56g5L>>iERZgC`zw
zsRQUxsR_G35`pQ<$VH9i7^8Q(q~eND`80aFJg@>1H?(5(JVUa@5)P4Mesb?z${*X3
zvVKtT>Z~DB57G*r)5yy;b(zucmCL+#uFbLX%B6g*m@G3rz)Ka!85BCw95+$@J_F64
zAy7Pi{_eEyoWZJa>!>Tmk(}r|Em`^`5BzyKC;{0toI}6sruQYSt%5PcOf)c}215#k
zP@jxp4>jJpGQs9iNkyvL)%nU5%{wo~KfYC7uS7VXCAr}^e%c`8GKJe8FBrMpS;#F@
zinOEQBf?TCO3W%!%E>u1jH~nkE^Ip@4K59BgKDW^f?Y}9$gz8iRwLqB8B57R@B61%
z@*0VM=RgbqvB;nC_Maf=uX(D9qO7pAz@ORk0@fA=|0#7IkZwuMz|M+C&kCRt^z?v&
z!~ZNU1g!p35%GT&7yjDouW9NpdCZ^cLO>y7+Fxh~07?x3d7r<@!M~6sfcg2K=l0*?
zR6vE@e;|W@sv_<gr(?dwht#)zO>Tq>@*KtSs1pWarU*B$UTpD<!KlKT6-yTa>-C(T
z%xIcD2ocIVk#0YsX@&Lbw){w&!ybjpZFeW5Wv_(%*jHfDHf1uUxBxly7B3csHhY%b
zR#dPgH__kcR+ZrRo)xxG&7lJdt#lu@=TN-s9CFyx@BVbdd$+u6boRb>BC!a%lF4ek
zX>V_!b8oEmlu1!Q>?Er<zX4lRZneV*iNn7C6qJu7Y)_AR+!_KoKq9!JPaS9nzSy;M
z>lYQ$e1yF?ag;-90CR4oC7W2s#j<A6Pjaq);^tW6irYG*3@%^CRdw0s)BM%ha+xui
zF5KO990a?rB4fnd6`zH^K^wAc{`K_iwgdLWn!0Mz&gX{`;SckXP(^atow1u+*Ks4t
z*J19Mt)Ow+BO8z-zfE%{?7SoEqHsC71(CD?eygI&IKOT%B)>1oB6x}-v{}TOdXo_G
zS!6y5J_05Y@+RbD{(FPBBYfl6L-HqFg?RgiU>S#4AQO!rxcTjOIH>p<<yt|gvwRU?
zEI|eCXW^Sglhpi@4}lY+IJ!U-8Vw>gV!mdOYchxqCV%yN(o?n%n2omQKLWXyisk?|
zEXUnTZt_Yp-g{yX<xjHu25vNi)uQj*SDNgq+^*c*Op$-T$n(X?daQpwrahyn-gJ!o
zxgPfT!dus*<?8KYSsO3zzt0^&I_{st>(8n4zt;x-^|bm6g39o#MfqP7Nzq!#%2eOl
zzz{I*|Bwsv@cioG?_BKPoBdk)UlY{7?#X}Ur@!wRf9L{#GtU2C{PgdP8X(F0odXSE
zr%Zr6>%T)&06J#)O*{eI5g?`mtp1IVG6VQ3VDXp4js>tMfR6s6j=wGx6F@WgHzdmR
zyAJwqNgwMUY2m-xU;^ZO12lWTBz;T(<sV@2clP>kVEKo94{*kRlkYJCZgIcZ=PyS4
z$9{jv@mQGX0qp|%%J_?w{wcu&bmh<S@_#{_^uHkLKkie1O7I+&Jij0sW2{S8w^a2%
z_@?HTS}9LlJ49W`6$j1B0||j;7FG!;Mew1BfrTW30M`@5X9v^U#Iy59m#EeWG;Phw
z+m_Z7bom5dWjhFsjM?6rO=ugRMY^+G_opr&+`Ju(a<^Q!KXX0v=ydES-*?d`L-p8l
zacQOIj4%rs8CfzerBz|()L-StE9Vm4Scg>#s2<8yb~8w+MnIMdeUSu}`Sz9-wWc`e
zXSSmuVn7wk3^zg#Wvzjx9)#M%)hBTo7OGeZ0bWk&6*xqU77?gFQdOu5m$*s033;jW
zg4<x5l9cpd@SI^5)Koj#*VotET1`_+VxtaS$cmD<ygF{jXk44j*7j{=+Gf$IfsEPS
z{Dh@En3hpplFc35c;mWJg7n<RE+58y7iV7`+L(-ylf7Y8Zak#Lzu`)R)Ow4;l4Px`
zY1qYT<j(SwjB!I#qm(_pwf&W*BGLf6YX+r#)15_lgp*Q&*-6Co0m5ZC<I>Ys6eC9-
zYW~O&i;!UfX`zv|<XmI1re?@$SF%gul!c?2SH`xiNBydtZY@djiLgg=NJG<TxtSEv
zCyj5yddHP)7ToplGkQZwFe9DSmFs%Tl>((;C7oDFi1y;bQo=d(90BD}l6~WP@4frm
zhAv@3G$AoaVy!V9Vu=#h`PlNARN6-IkB0B~TiK%McrvGfewYF&n5gY4+24FJT;heG
zCvJL~*kM3H(g;Yue1wM_FsY4f1zM4w#5qmG#g9bsjIw8tEk|;iebnt^{u*>qq7cPM
zjjTv<ke`wFz5mV1SgXm`lQ;c<?fUUid{4cBTyji-?5$1b+>1KrV6@*_hU>=bOkeVc
z2v}hW_jst8)A~8}_vd7fYthVb!O^KRLbLqJb}Nu}oGy2vkx}91Hx^y;(TcTki}9B0
zRfCCb+|SMCWr=eG-$HU*c-baL>(#?E)ISxm5CTPuT`nlmS)WMkFUrnIu$?9Q&M-bS
zFv`1NzX?xRED4Qep}G%-;j+NIEJr#^`yAIFK%6+pZkJtE{OqbOI}IGA)lm!Up3E62
z@jb#dIc_Vq;&E8&af7*K5>0Z#FbhqFT6K)VmiIUJ5mC0wI-O2rO+wuMPRZGe#DhZL
zejHRF=V7HxF$+rO#)S4rWr=du4a(d6VJs_8T?4j(VfcfRvt&|h;Ir|j2K}b*go%m3
zqgv$$(;pZLa-Q(9mJd5&@{V0!=5`EvXEO_)JWt}+e6@M6ZJh-z6Pru!G;h)GFOzLF
z+=Vq<!&>{~*9>$K5W*N6O6(o$8<T>6UfT+Cuv|YmToUe$tGZ8A28UnIYV#j|8as)(
zT$H}=eyJ28Hi`99r=T(txtOWJWC;e_Z9Rpha^9YN7NgjxOh9=;@XM6vTy3gy4a9Xi
z)itjTFm$>2ihF%C4&B*T3Mq36eX#oQ>MoBj87Q9xoP+NhOeDuJOcKBvn{im^liN@7
z6_kQF&i7Rh4)W!p&2$i|P0|}NSb?9Hx&HiPOq*Qg$evM0F-yj{Zav`9ljgRPC5Z$2
zTuv~31~HNwodref0iE)j<}Gh5Izf^gzCGc&hkr0sZl8s*Hv|3E=1zK3f#x+Nw&W@u
z7h3;>v|!7RSZEZCgz{(jA-H_CK$P4{?^0m$C|NHbVGBLT?_h2C)XeT+Hc_%St612}
z-wjITT_{jmCq5@yZFswbsNjkPN(xXztIi_k>B)^TzEa)rLxG3lwgZPeG0!DvRS<%P
zDcXQ55)D?E*6GO=^lof(3SM+Y;T2jdpa<CxOF9o4%lj4GcS><LaWd>1#>16jUWT{w
z0&!Z=9OB%E^J2q71`SjOzjd!EIQrG4aZkqC$hWUQ&sV(#cfn6wPKBL0k5`rD_KG&O
zv>{6kyA7B*st%b&DLT#7B*JRDv+@!!E-ENku+67V;D`}+%V>Q1yw*0_*?soxucuUU
z<bi{`;!NurVWt`nR+1$xv|p5WimWkn%*m^-%LTt$*bQUJGqC=8y+Ex}QS%tbz{h(|
z&&3C>&Jpc_2*|LTkB7Ey^GLEdSz<MP=YfssaWvqOaOhPbW=09S`j$=b^_UY9kHUiy
zyf{m_HKsj)m8N?{T3>r)D2_K_Rphy6e{kPf^~Vgzmuwh*Aooptcx<y8mbg!ry%-2E
zIYB?z$?)G~x@j)>>961=@OOoU1gaqneSlefLxM@+0|}-)f{C`o!~{t41Yz_d0{Wr|
zFl6!LvgI1Zz9N<MxqfTINU;o&T(d&0E&FPEZb7Qn)`d})v@bV$-@(M~y%DV9A+Teq
zLgqm|hOdp}(iRH(4yq(B<@N6mS`2{4^FO~i|9sN=>mWs3KuF|^;@^W5*8d?1iV>j9
z2bB0?#-nEh<UavaAHRhD|C9s;a5oavwGlHkH8yd;``0+-*Is`ervQqz{28VIetP|v
zFy-G=9{?keKhN&J3saZ?J^aVl|38N*vHaFsfF!8%H{>;71nxXr1hFLuVIcWDv1Jry
zlR5Ls&+`bs{f^ET?uZoCslz~q2U1<Va5!slpD9!!Jp=Gu4Q>!QxY1y`?#ir&ER;=n
zn6jRnKYY}aZ$WKlO;MSgd90_BAb$R=p-JI|4kD%WPP5dhR(y`o0F#cpK}p8iXXOp%
z)vi=`ltc9paBWj0-Kn|2e}}5^`YvFm<R+TGwuhP|V|FW$E83s*x~&vaX0-yti4Z|<
z%C^fRPCSYhkE3OS(5EQ27P;f?sxP{mQewU!s8AeBQi2S9I$V7`Hdp9{w;2TgHIHNF
zF+IG#0AuM6hz8b>Xrjm`wyvJhtpYPEd%eA~_@v@~?_$-yv7S$V0{+vUI^V)PmQ>j|
zpRyAeRbff`>kdfVgQW3KM=ty&zeD<V@fT;yG`ZR;k#Opqm6>^kP^8=h60sk|=Cb19
zO9<q`iAbj3<6PCl<0r(x%?_$N+a3Iw%2n`?#VRrq;HP20wGu`ZhPe#s5y}&FkC3z&
zMw7o|kc@~VlIDS$$+F9g$__#=ti~>@U3H<kN4+sT&e-h}A+MZilxQsMcD&`t(0<R2
z#C`;>+=Tz}@53Ho#`5R5_;b|%@8gUAI3xZl#P}<b#_tjOuc7t-S)}nFJM(W*#-Az;
zz{B*fxQF)tNA?k5dj5Hpr3|f%9ZUecf)P+tiS~~*fcSzPP_^lAYrj4d+W&5nvakbs
z`;S@r=R^8;$Ind5ug~@Bb8$;ZmVIlzYi9H1-2N;v!D!+{RYh&cN~vbwbLw-Bs>~G3
zl)h8U$>*BQuDup+KXSkC-?#6#o)57P-P=!0IIb_;Lc<1CB?n_boq}})ifjt^3iCto
zK?^csvg3gfkwXB9e{gXbjdo#uz-9CqK#1tWpCSu<N6zvM6ubaQB>|BQW)#Zxam%pc
zvlY*34EJj+1p)(!`{29t4kM(G?2`gVO9W*?1~Ds`yTSG$@)O2(m@q-wWdMseFVJH)
z5TC89tJ<c{As@3d8wz+}md`i1R>qa|Nq7ib;FRoaF}AgP9pZ=hv`d$8zoevwh6b`;
zY|_wv0!5Gj-!-vT#z~|IfBm)%x((UR&Sf~~2Wx|z4>(y(j0pG3R^Obu-32Rc@VYj>
z1$4`?349}k_%r|nKIUuG{E3ec3hS2h=Z3Qn-Nl53&sNUXJNZ57voSu=kqQF{gm8;K
z9+?gi#775#<}P261}LFR9v59a5YT6IFx#GN@JBdWuxaQmF|Y?md$5ay6mU3!4-cPl
zU_yoUF4*L-rcX(Qsxl416y^oGm?qE8Y>;q1Ke+Nr6QRHmhp(!f+#6}hnOMiYXZj-|
zfZ9KS1MP*t*YM$Ub8R-Kcng4$bn>!*V8O<H8X6fC8v@?M;cJAQRXZ_uwby}qNmO_#
zO!mQ>#P#q0kPe2o;VVYUw)uG@(%+@a*QJz{Yx3Qj^UV_|I1E@<j11=hOcYPf$GaGD
z)TVIVxcjO~3>ObgXd{CS?&kVBlhhrk?W;*#>N)Rb#~r1SfoXY>74Z$`rMJh?kqK^A
zM=%i>838RdG!klJ?5E-3VPN2Qtvz1g2h11R5_*uweY-4gildygNxsK*G_am$%%BBa
z^^5oD@DCU3*jBmVn%{sQ-?Q7X#V|mKJwo2SAvaWp-@o7Umb`TQcwa|t-aXvA8yR~a
zdG~I_KNDkJ=luvm;l2f4O(%rd>-%*6h4R6g9J5<jO!of$?q~}i#Pvru97t$buw;l|
zp4W18qlOqR+bCK{5XU>o^m+;2jg4T=M!yNx)jnoW6e!SZVq_M_56aClKSHXLL`WI9
zxsGSE>@1U?vbW-5BOmZpsd`ZXc0*Eroj~5Uep0$NnL00@LPK&UT?8PaeCXj00{Vfx
z%9S()5Xipye#Hl)0@;3<`TU^b`GXWi^2<6c23*R!=Nqvj+sbDcP$}80?9fe`iJny0
zlDVI)!tJg2?+-=HBtr$9>KGH8kG<=Q8+r+MNkLCqs(o|54%<b>(Yc5pE&MQp1bLiJ
z+41d7IU+BG)K?*oGzeT@MLAEw9-z@ukm;#XaccNl9y@n9Znfyhlg_i}dQAf?f_Ch7
z<u}zRtrnHNpbucEmI^n;^n2D-Ml7`mBWgA`-U7TW^+jX(f<TR3QjXQ3GT{gC5n*Oi
zx78v%P&>hsD~yew?cADYh#RhNIZF&E2oBZIUT)azKF7GLq2cZvFmD9A%;n!RXXKOE
zR&zN<?xYj?IylSt?_%ndZQ)a~qn=yy{A`=E3G-KshhE`eE+7ShbXYm^;xTs^Vh)gL
zb>-%OP4VDrDQXzjk$n?Jo}`F{o1`^VO`2?Fq}?@eJ@u8S_{0pVgj2Hh+$EA(;DkDZ
z!VE&Ct`tD|LovIwV@puLX_T3_ZXB34Glkpk$n0sbF|}ac=6bN$LtVSRz&UE@=IW3o
zF@+!Hin2Y_y*{VG+4E?`WQHmiDvl24Y~e0z?UFLx22oyTe1blG;5Eja@kB@xJWs@T
znvACAYYU8RK|q<YRSFGo>6Y$u-|p!UHzjqeY<ue=dA&k}^-YT7kXOp&1@;~XdnLUM
zyL6N2$&7}~GKcSr;^DPJ0p@wCI1pzt_S$G+D9Gx|5$iO&Bm2uH%Y22I<;-?BHa-18
zID*plPWI%=jLQ+Lb1U9{F^lc>IdOmrwf3i2$N7g(_#Cqbp&Gdlb3^u4+plMcyxw10
zb$$Y4fIfM`Oe`GQc=ZsDjkjF3k>+ucB~Z<>i|1$vi(3hXHwx%kpWNzj$T~57GUK^h
z*R9}IH7xo5BHxPtw3MS0>;R_p!s17XDVs{?S=*T;R$3==4|UQM=#~<+RF7plPneK6
zoDIPhMl?>;MRZMKe(h}^sTDJpOadI#Y|lnk*jrM{f2{X0EqgdEalNDG21G@M`gnzn
zyCtLz+@UcMDbek-X5aL*NGf{4VdCl;=%5w7M#I;J3P|B8b%<+vRe`5#KHc^7SjZOQ
zQ{>b%^S1kO#6?|7>ifu*nmqlN&|IzWK_;q@`A9tZbTeloFrvdJA@TuM#Ur(JDdD0u
zQ0X$U-^kKkJeuca5JwVi_}+)O4_qr#OKHKnw4RxT<OXtNZwAXUm7_?ZiI_>^sA2XL
z?(6Xr1jXc+I9IR>u_Owiy~ZSbr<#UNF_l}4^{`fmha%Vlb?tjPh|$--L=7>#M9(?N
z@oc4wBoezv^@}cNfkgdayekeXXX0Lx=6<kM$0bbWjdJ6?(M^1Inb+gvv3IRKG0Xhi
zyN>$Fn{#W!<13k`&vZZkapP#1PB4!FImFM8HF@Y0X93i#YG$-Inpu!z3+}QSxjSuG
zSJ=dvga>+P;e?=+T1{?sBWpMvp_F(-jF((j-g&61Pf=UiI9iW4gb^`@Qj2zty5vo@
zeVDSk1o3DF`=E3Ly#jNtUodFQ7T922vb=>1<Si)t+bo{(6OL=0zm8Sx|7^8S6eO3J
zOdQhoq&N?p82a#|_eq2?<a2}NajM{*<^fHgl&!dy?pC3(cA|q{O)n^7$Z~9@&UnsV
z_oD!BG+XVU6qE)uR9*gbnZn{?qjv*++v98ld6*z38T9s$WWj3zZ_Pg2Wv)xb8+(#a
zkgBd&YTX&m<XdeRTO)Pwk8U60<0~t2ru)0)rw2od<tTv;cnRAJ-<(-JuzubAw>y%(
zyjejlL^07ZR%Cm&o<V0`vomQYxH88pxdqRKW}j|iO9J`WS)A?Wjk$Z{IW2Vy`6)}7
zkXdTV@44u<*TcL(E=HE?5G&u6e4YA{J{cU8)q<_qUPo+ws_|YNt?~FOqv1$q&%G3a
zb5X7F*t0E~S4ad2=@*jO@r8gM`4b=uDDiM?QaBQc)?;LF{xxnywTAMu+8(E-u5_%!
zt|)a@-z@(;qmXWT_A4sR`!U$ld<<0T0M$8xBz{b$c=chKoi@yk*UCJ=@>@RXzKKzE
z!r^C4(nAsaykjITOeY4z2aUEj@c5-}g=+We1mNxMRnwy)p3g|x`mcH4QV<w&7i3T%
z)jUh4l4+6c2*0uAY?-n%qxyv2oM|p|s$qKwNhd2~d@$nF!8*T%U+Cgnp2lN7O%G(-
z6G?;>#SdFY*(aQlz8CyiLeZezY<)?><GrTefua;p4FTD<1WB#xK{2-MA^h$7WMIdC
z&U0*N_!aTXRM)J@_p$o{hZSyxSH}3J<rFE3wV8rK{I;%oxu_01^!7nV>RQZg2_qyB
zm~Hk<lcCn=O}E2UIX{zj=cEyAHbVR2Fn!8wyYmD_?|j^~K53dLF5Y?ges8Zuqp(qp
zTYKY@JjTWmg8ag&T?7?_Wvd4s!x%!QnvZys!}uKDZ6S`kt3Q;4dRAqtW+(Bl-W=+h
zBEDE5oNuM7uomkl&6a<JXpGE^C!JJAbBtOjp%GN-kEtapAC)>VTJE6JI<Z7Nl00G>
zw2^O&qm5|kHJYXPUJHaVdKa5U&qlr|Ux;PADuExke;`zSahf`*=@{K`d~uiL@#2zj
zcRBdB)&6;flYMImfykawTgxHktRILp-^Y}+;h|&&v`Cgdl9Dh(xq#9^)ji`ob%GLY
zo7<s^ci8ok6e8)2Lc5%#cyt9?-Xec$u+83J)5X2|^Tn6?+5^@X$UKTI?S57l;uVOB
zJ}dnsB8TMi<ILTn>oFJepxa83Rm;wTg94gU2BHWbSr!RrD9yFY3X2DXSWbp+%9(IZ
zkyZBx)=buX%I;L$#qC|Y1U9P1T;<$ld=uybpIAC7wH;~HLlk4@5M6e{iD);~<QA2&
z^qX_yQTr`!S~Gn0<Oy^vQrd-$fJ|IILj?lQ!f>_2LYNo^{Q2uL5>t`LN9$hbG==@g
zlYyps_s*YE29a6X>mP|>lbp#<QeSmQRLH2I6+ksbODjtcH6mB|62B76lj=hj`}b#i
zEzZ0-q$}^leGM$kF6hH5a(isa!qJ0L<idRtYa5Zz?2REz?QYG{AEGX$?A{ZG|4u`H
zDy~_(99GfpM~{%W){Zs@`S>9+fVpDGBz*LeRAmVV%bOv}+g&t1p|4+XR~#-(7WL4v
zYGpS5#UU5gj$J9Ab&y@Dng{p%Q(|HYdk7aR#9nU4u=$`xtfku%hLNMXTFXFJ!n382
zqmG4t1Yyd9cH{teIl{43w6g4}==b}93tU`lf{s{-213`6$$7`C^*}rdHqJ}WSh24j
z+SAL+HOD&n%>>^zity5(CW0PL+8?cLhb*1Z6C6L}g*=BU^*DR)D8`AJ@guU!Oq&k#
z@<h=WXr-=s_m~VSC6f-MxLW@##@1N43w?vz9NH<$20g|qU!mK_m75x7Ly1?@v~N%9
zWFlqPu=3dhx@Kmes^wghRn%_!S|A^q3^Aup7T(5B8UNzd`UsPVr+PWHze}CY%1~1n
z96CdAz1@xJ8<u-4*X}=Lb-gfYD;pad$1}iO3wJn<I!b1ok`?qVCj+!oqBerF{j!f1
zj3jZH@C*i|n1}vTH1Vgzlpw7<sWP%?L85ec$`AJlA}GhC8DS>l!+W^j+wjYi=l5H>
zFCDkeoNSh(b=~4t?F4<X+?Zc@N>ra+Yhv#*u6G2?9pt6U#ci~KkhKWNw;w-u2*l6$
zq6%(uSPrQLK7Umtd+Z@WC|S~;P7d{RXM>)qlA1FM+up+vRC(%h7LbMG#Ld0aGE0Ob
z5yB9<MT<L?id!8(=rH@tbZO&%^CL$1*m!R$R4yM=u>j@~C4jfJ=HreGXngyfoaeCl
z`;E!C(navRwo|A-W7Wk|5lt&zMq&hT-zc_Sy?&A2dE1=Zly8lmZv&&w3@xwIaGAqa
zIKZcA_iW4m-UA8(hCImRdFcnayAW?p!C?FxYh?SQc&BbNRp)_^x15BZCxYQAMdV}o
zt>z*aJFW6LB{fk)8ey*WE*$l?Qgi;d=hj{h`WO=B`Nf|qX>tQbM>Dv(C&kU#MK<T@
zJd1Ule94;rHV+P?E(CSpN*T>Mybb&}jFEhnVLu&ufr`(OE{Zk{49{~D9-F84d=^5g
zpY&&EY3kXc5c#m03`&JA7c}<@{jnxeH)~!wf%?Or?|><VBNk~XHNIr#bkON|r_j|q
z`<`waI6EKG#5Qs4owE7}fnn^MQ&5iOh;ud<S`#^I4%Jwh?EhqJn0{$*QDOrm>b2V?
z_so<)CGP1+L~?$W4J%2juR1gJCdq3<l2Y`^a-5gK-72b3Y7vfm`Qmq-4tL<jEjYeA
zpKB98+C3x)F|#X7Y<6C?H}Ek{yG^LTc<;E!=Nk9$=4@XYHQ8EA%6sLZph_a>)L{#+
z<&G!4U@vItER~iq^f<%CwuY5D%7OmeH>?na_b7ZZymILg{XG~-m?ePtnl&#YfCobs
zBZ$d(h+$UM+57oL?rI>!KzhxSnd2IKFrMNwz{{Srwjwa<-n>uls5d`nB>MEH*!7nS
zTZICo?T@c@oTHk9+l;iBx5x9Li0f7@J8q6G@)SZzw4+c+2s)2lNF`tO7EtK~>)>A3
z2&Agb=+;48@qH-lv}rlK`6U|~a0Ix5Oo&3z0>NshrY}oFPN_VnTCZ-5HY%wX^P4^6
z&zL4!IkU$YXyLmPQ7|x}sp(dMO9_~B&cg^-Aa_<EGcqLyDbx5^Ue@R1KA*b5S}z^!
zC{2jY5(kav?-GmKiL;^y<f3a*H<luN%f2BGr_L@Gfo%+_iX7sd<tjEiM&0bTxP$U4
zky@ZM{4h;zZu@0615Elgs(Q$Uca)UCaNp*-yHr8_g<{SZ5_G{oj)c|veb%_Y=SKO!
zts||KE=JU5je8Df6Nq9w;<W}%0;5^?gu@5<4pkYoWc?<PIDfw-@8{5@KLXvkKv-Uy
z$tUm%Wm$E{YwqRiuQgw%;0B_szcU{fe{f8;#chJybsL{)-PtMH==$KwPCnAX(EvSs
zHh@Ip`lUwH1t*>9WHp+On-{w=%pye&WIWSHkrOWH1+S|-+GIXwuKuNPplZ|j*@=yk
zJcurA0U}lfK^To;&M&JfI(^R6JkxGY^<3^amSmlT(H|DQOaL7x&<|9JQMqRrbPxEr
z73-y@JO&7&nRCewmw6>H{E14Tq;r!6-kd#J@$HmLBtiW%WVASC61OfBq6CQ>U;gbd
ziZmX&DSB>CgeWC-|5aadAM>XVTIcxzYdnl)(Y}yBRgf@s#wO#5;-aV3+*VtFP^SlP
zVGXwtb{&P?jzCvsS{kp%GW#+G&4Q9I%Qe@XO<kPwR_CB~p#*1&212fO8NReTj3^05
z<05aj6^Rmvkq`$(7>Z&Lrmr1Mg3jmpC2p+g*5o5BPfVGY7LzT=t9H$3EVH)AEanBP
zE0EeF!>yCU#475Uu*Q{4)U-EE?AM?|!4O$13f+7T8H^jM?%XqzalyR7XV==WBE&h2
z7u(MstPkOzam42al`BSn_U~omPf|Yga4^x)swM(>9e&bV&t$$VpnR)*>9Vj}A$dgg
zcs}ddyI*0TUB~@t)nC*wxQsVI>fB3<38mF_DbwmXXEWdY18tY{jbrT%V}ug(POM$B
z+TD}Dp`w*e1XUV_?V}UidiA~FaoZgw>>TB%8D#6ew5lJpCn^nZjsr(p7VUAs_e(2t
z#%avQi^xY~M+I2c*@~m+RFjjtrk0Vut-uw2bbDcU+xKQII7eVG!@$!^C+<Vcdd;L)
z!qdDfXx>H~J3wB}bIiTBIQs0Op=hLwM=n>JpqD04n1&uNi&mq#WnPjQkju4(KDk5l
zoFXwhVeYYb*!*Lkkut-ab;5?fJljTZYl4Uxu{}KZ15s}Ir#3mH$hIN~IKKcP77re*
zp(q}6Nz{4NxZ|`+IS73_Z+~-BRE-lHw4@kts67Y%f<$dZgPPWkVR*F-Qoi+huhQLx
zRiA%+PBWK`{UNodtuRMxtj_g$9VWCYXKs@0EPgf#bd@WkeMS_Kn8w&F{Zb7@qeyxM
z6sx$~w;&CFHvY}BCj70toE!dAI)b!gs@rqIH)R9^MbsOVIfQhea!RB_r>P=F7zpij
zf|;}4=-n!lh%t!R%H>tWRha0^Sr2*8vm9e_|Mm?p$L~^N7bZc@KrR#RU!t}me)txi
zmJ~zkH(Li@_cOUCOQ74gGm8dMj(zMVsX68xwEG;bKQHV`C@H6et#>DS4g4{dr1AEX
zFP?+M*FC+A=*iF4?kS&)o(@Mcly+q$j(V>?N)+Sw*6(`O9Ge~Wq&BS0LT_E11df;q
z`XWhr>S!qJKqABJ|M2SElPHM*&pBkPS;<&YyiM?N3XN#1oMqL&&aT6${!!Gy1Vm@@
zw7JSW_CS9@3)bT{Z+_Dv{TjY7UXcVF7>PLmJ-R?<NW<~}TKnp_D7&U%1eBB#5TsF*
z?(S{`>4pUsSZaZ#8w5#_?vxH`L|PCel$1t7Qt1$B_^t(S^}gTxectDf@ArdicTdcj
zGc)JR*_|^pO+!Z|zKOoX<U^_7x`UKy+?#7pHXder`k+#fZcFFWe4CjukHUS#KzxhC
z@{{ZGzT#9e<;BOZ2hC5Oy)-+VczYpECKpz6)|{H=-zR7@TIGkKak^)PLr0x3BoPx(
zG$?Ace1~Cq&s{lz!YHx_x}-cze%p~M0cM`$J}pS`Mc&5AVJMVL>@ctL%$!8uL}j03
z=^h3F2g^k8Y5d{R6pK2^Oz7jfT{O@2PR8_h1I7B(8$0UZgaQT{#Ejp{0*#{;aZsM}
zO`lCSHgF!$cVY=Gx0Mi!4{@&IM{~t_wwv7u0@EYnD{M+cCXo4#dXI8_-5a^cAB^5$
zNMXsNMo!`_*!H=blrFxc)jDKTk>~XoxA9B5II)djW4*`FlUMEqElFv#`&Cv+&KQrT
zTJyJ#OJd38I7~~9i-g=25+il)?KF*T)kf&9^-_j>W=LJ^4?dJ(xoZ$_N@<Hatah?B
z=ipT<!@m8#>6p0Sd!0RP-$=rgmPY}D=7GUohxaYVJ)N78)iw8TLPv2CN5*3gf+Zyg
zoDyX=l@=HXj@w)2-}7R;DP}OJ^XiQ=6fDY2=4{x%NS`kq$IzZR8D4+R(;QSnmFdZ2
zqK#bNDOUBM4hi_iUUa9(@q?oFCg6T_v9Levy1^th&4Qk<TSLIpWV`@A_HST$exJ?@
z@njEI#_e#GtmOGjb3P#jqv2@l9v)W~Z85(w=|Jpj4PUp>P&a65ah`}3`gv_eSgL|A
zp^IEP(Va{cb-hbUWaT-2Pa0BJADZz+dp!A|+=g!U#OU&^vhD{$p)s~umBKcd#?R#>
zT>F^K8pr{CCddULQu@rSKWF`=4s#l-i~?)n)c7oleTnvLtT6J1+I&w|`aOnPi%Ctq
zpqQ5i9SO2OiZT&Osi&Z&DKEuSAHVHd_1=2whGAd8ngiAEmUW{cp?I@w2IO_ka>!4~
zeK?;}Td5h>YRWd%A;Z3*;kaGe*Yf`0-Py8juPvAucF6LSsH^VVGyZIMVcgLh(;fuO
z&kyIyY>X^FywZP|qkBV(EMCoHE6$*Cl~uF5QZfDLiT@h6<>BYf1apNAk2!D@=zT8P
ztuE!4Rr2o;GAP){IZK*Gn~ZXAYbtPCF~m$Ae|cw1fWoQYMBpBBr0)@%58GRa6tywc
zZ923sViR~uY4qeyCvQ~A&|CkPuLBERSE4SG8*}T#k9l#8reBBZ4tNK;ua9*-{1#no
z9T9IhCH^u<#7UQJ<r!uQ36|d0!^)V~&m2fu+|X(DLcY|<xR-aOJfS*^&K#qbw&jqz
zMP&Z?6JcATyk&mfLxqmQLQ~wbp30$y%rW{>9kEGwVIb!uU2Iu`!9pIVC1zTHSoO|9
zF5Z5O7g$OuzdtD(s^D?wzF_uu;gN1!*z~~EMJ<!%r`h{SnTt{quX&#<tyiFrI+0C2
zorLUIj$k-!Pvv2luQwW;IS;V2J=hP6Bo>;qTkzfw33sj7cwaY;SQk7QWF%R~hL&B;
z7L``55?*2>D40_EfI;+el2<KqhO_Cbxg!Dh;dk#p7+>CC-<8R2AA8U%v|VIAi^<Uy
zZX-e#_EnRCQXxj3TKLrwg*v;f;r}#svUB|9EB*H?i~nxnSJqUQ6xU*qP*T%@UjeoO
zS->cP`^0bqqPRHJ4Y-a6SoA5G_^u54K;ju5ASmB&qdpICUGPtvK5)a2<=PPHM9IVk
zxIq8Srq2Wq?V#oaw$-Er{EGi>)t9gYIjVu3Xen9jL5?5`N07ZG-8CBUE6J4XSEj|k
znN0zY>Hpi<$PHX_zBWDn7sf^ay?-+{I^9%Px1BDFqokrScp8}1F-XTmGb4+|frN?4
z@$TawRS4zG^Whn)09m;Zzwad>v`o*P(Dz>K`fOkNtd=_WPg!p^X9{Gl*O#w<`NGLc
zsQ!$SENe{`SyM$4l@>!~l<Rfe4FrD^anUSVI_wu{m>9tqY_WSuVnakkcvPpYi0VTs
zID{5zNFWD?23cg$X(TKhT%-`$-Z0r7M1O=QlqxIRq7QTlvu^DLiXk+JBiKGb45M0p
zBK6X1koA!TV_Dx^D}peqh^VBrwEVmlzv$%cuxI{asD8|->zK}fW9QR(1k(pZw>nSG
zN{};~I0(aI2nc`d?cpKTg94*0AKmo1;dd8v9g{V{B+F8n7J26mt0?x;t@G!KA#o_i
z3z+ZDUX$D>t2CtK#P{QP3o@ofeEMNWR5Z&Hp^WdXwgO7livVgLl=U+#9E3AI62FkB
zZr|SXo(mn+XGigvh)<*KXmQ$6p4lVvw$H2hwKPQHjFB)PAW%_?Ac@&y;w575A-By_
zqUgUZPk)NA*82<r2Wy_fjfnN$vv!j}R@Niskam#|^N)|o5nqNm!JblA5DjOZsy)rl
zHvU|Bih7w*%7FMSXy>%rK9GtsQbbHtP^CQ|HK`FrU13uhwO{h&MITz&O<4r+0~#7X
zghBy?wc9R)7do>YC^yfS@sA<zsZe($sGcCmkfK^5e8deqjXD8|W~d+#Zr-@Pb2=C4
zTP`aliGV{ygkkT;--eTWsos^G9oahm=1UG}7J-2>(vuoteqdF^mz&o^EDH6_0AOld
zVJOW8*}TkmSmjx`JW^MWnne&1dqV7&6%oiODk-UtM@d6-6ZOb7`I7YQ+`O;OZP5ym
zRHO^FCm3Hskh{;``w29)I=EeA-<+f9*+IIP5yK8$WXFGdh<)(NA5WBZi)QbV$<K)H
zl4tg^bno)zg)dZg1_|qOdF6EZQsgrw57FmUWrWCHa_uxmr)XXW$_s%J(i6Sj1{Ktg
z?>1YTb+vIGllP)NBJI2>)e+>k(7MSQ!T*s2oy`x_*54ZX{gBpPXr2{$1R>P_yve~{
zM|9rz*uEI4alwuNbAF&Si}%TS<F*ho5!!K)@DnsS1k~rZj-wDA9^8qKLO2SYH$(UL
zxLn{7)kgL2SWrdajktjz^C0E>s<xD5Cqg?d&b_6;TiI#8C$g6s6dV51izA2Ox8AJy
zmix+)p&vX8kSh?TY;$~JutrgrF+yMMhUP=oz%MXIjtiY6VfJ56|9Ai!qv{_aa@DE~
zR%gS?C-!2*AUT~Ma6Fyvsi-?xFKFW^O5vMeIP|!y;exz>qI_s1AT~)QYhHj`R1+e#
zwuvmkE$_)7UZN+rLeVdg*U+I^oQT0P#m0bb#*JX-w{V}nU!4$0NaAvO%O{TJ(V?Kr
z+u;=Ygj=@pTcDR<en)?bo8dy5kmu(72Rb!jeMCol%bTBgnjt2h$A^v94@%lClkJYI
zVCL8lf3OYF2#(sd>wnRR=2c}<2yN|AUm|=jAYshHweL-R*;f)M%=SY1k<h3y`}tc}
z7QuV-(K<2WH)QklrG+CO29$+9)4SKRT$GV5<msf8PqMofz?FDj3_}_dtNIG-&6C`B
zdYCaIwuPCVM&#okKy;q<!oXh~Hy9QfxG2Mial5)bE0&&RQnTRQ@h5jYq22V8j@O?T
zdV7L}s#0u}qE-1jS<vv*CUaBug%dV>PnjpY7h$|P9vjbkRn-vZwmkT&%QGXq<?DM1
zhEuf;LXfScXumE>+bCL~JbPq#q&T6vI4bFUp!%urG1s>UM!Mt#Z1U08g+bo}#o9w9
z{f|sSotOQITHaX0g4E0dUK2Fu6jfMULGz<yTN+j$6Q1ldn5ZrgS3SVMZc%a#FZJ|8
ze$LgS{DHv&Md0<xcM9QcTUeUV&V*a?9y1H_6OwfK{HMj(2N_ST*Q*$uM>tPugmw(r
zT2<fbJ(O8nwvKz{NxDxCrFa0ioZ-ih*}jO+pf1F64h!WKIWPY5X}Yq@Bd8@u$c?UU
zzLQvy1qHn7>E*@|#8A+d#9__p-8j<n9UUqimqlGYj*$Mz?{MpE!+Xpd>ic5e-(YM5
z|NZkJjfBFU;gLy$G7@5B(?jB%BWBSsB=B7gTh8*0a)zps`aRCF{jrWlN#7RS>GFAx
zLA!b7Rt*8Zi;gwZ#a+!+RF3B;`aX4<+fj^L(#NaCRv%c!WBKuyjsvprmxif|xZEDR
zl=VFhBWTQb_IvgI=$Wb?#fu6dw7$_^M&6P>RYF&yf~;A5iMu~a=z25ThK%z2brF!5
zs)y-SXf#jOL~jbaoTjwjhvDyNG5hPK?~uG}t7vlF>nkq7G7lHR?hh)6*Lxqk1ra39
z+IjP6ut;cd2sKyv$)LYDf!&5+aG<_&gA~2!a?2tWm#r)PH>yZ^c~ZS*6ISV_aa#d>
z8;$`1v@n8`l!Ex$Y>I*tJkCzqGJ`lXmBay%-elG7L(=H8l7*m`+}<IMXk9I;I_j~z
zH-pQ%^)-okRgM{WAqyR>kFXI7mxm%}<#Mv)Ws5VTOr(O#>~F7bedJZ{=7w={B!u~Q
znJ-xN$(N~VYp5wZwpLE^CZBl$o88BWp&QFNo_w{I7u;)UjP7aWq7oxEMfDfl0cKq_
zX><bIa-J0eDV67UY?pUB<3EzHa{Z{<twniucjkr7czynDrTz{L#$<G>fSinE?l$>N
zwA2S{e2^U@)CVqWCoQ}86ll_motiZ7jhyd-^uo)0X}H<97RQ4SUp1a8AYN=FkbV(E
zmPY%gmmSeJhWXL!2YsI)tY7ICB7dqrGL6|w{n*8t(KJ`@s?&@fe#3sGo5+@*rgLa)
zQe1lwVNq+&I}0f4Tk+ljsJG^JRFm(qG$_7xM&5rRo`SNgrGDZSwtZ2U!PiuL_Y?Q(
z{IoO0%dX@53ghZqp*?p=#y(Z1;cGp-q3|R_T6(ZH6wJCHhY{oH=^e6Y<#RM%RKE4W
znYjLmR50+vL>CKRD#neEVy@?ysZ(E#3hCQ0^7udI<h~34xB!(|hk`y=t*#~%dlNg7
zpO8L6!G9jxeQYTzkY_N8L?DRk$>t6fYL$JM^?CUo#gS>ocl+teys<o4=V~L~NjWI%
z7WU-FTW)P39IhAay4LRRY>3}H%MTlRsck&TfWB9g#)q9SaX?(nJd>KfKer*CAF9*n
z#9WywE9m1v`Jhyj)mWWBAY<-31nNn`5h5bsxj$pS^N|>mQ`B$yiUtO)F4IM}Fa#n{
zp7^DXI}7;s5En&7Sq1gEdcLU9(ww`cX{Vzdm=se~XfNoa^V!bWXLIQSV^eiMLIq8!
z42g-jprIvIhxXg>i=mblqo<a&((>3pMpMZ!bA&s1m6O-LKN<*3E69v1`XG9)l_X3o
zXc{Y%WlVE;;`vE*pL!{=213@|M!)Vkp-j&gmrE~SKFya1jTW)4m)Q?$%9mctqRE+m
zmG^?1x3T)c8SzjK;^ys(M*GF@=)s>%z>6JX4NkK{tl4b_A25`XbmAA~x==SiPtC|M
z7a9<Umo3HZPS1o<ddYl6x<_&*SJ~29V`3I}uL$$8w&5e<cxwUYsfLA_u8q;y+0U$K
zfy$d;b8^9Buayq360unU-tZJX3Ko&v9J$j^@~=5XBw;a1ioD^a`r>-bwc%Uj1SBim
z2H|C7?*kma=Iw>O&U%I1Y>WT_2OZml(2g0NA=NN=Pk*?<=bN7*ro+crD$EvwMTeRu
z6tR7d^Z0?IXj!kb={IxzT%${YjsE<ESR^)n_GJ&dN!h&V!_~o-hk}Vadfv0@lV(qx
z3Xah8-i@7Ue6H?yb#T3dc;p2VBY0lc|0YE7W`WDN``tIM%pYKp4|`gF@m!&+*0pOY
z)NIzgLy3uGf`0PI`QGOay2hG`#9nZ=X8J2){k0L3_j+c5bJ88D{We2+s+F(Aw!ZK4
z+!jXk!uQ2=eci22={J*$H+=9a7Un4tY<IH<x*lPvq?BperF=jA1gsZvTL&8a#B!}(
zULkrnL=D{H9BrVAK$xPEk)OV?`2-uq?6u_9UDY(?_|u&rP)}Itd);6!<PT_R&D3&&
z(jiS5k736xcTu~r615pf_enJ)oQ6}D^OO6f8p{%kE!sq!khdT26HIe8esCoYy?Ip3
zMc*^!{$RZ90i$$Px}#?esp81ZEma3(b2P)M{qa`sH^kh6X6sBnU#A)sP7rR^Ju#p`
zQ<S^prvFAWyKLGY<Mw)jNV2jfP^zi7Up9dFtndTuY3h=nUrtbvn@T~!E+fB!O1`Yc
z04$=s2x=GhCALi(8l_+}ICJN&P0{4Se*fo)WGM$*JWkvJ1RrNTeluR1cteFC_1gT`
zMSJMCb4@=ik$m9GuG>kZl%G)+IJffLh^D35jm>a9;>%mgA<VZcuN5HW=p#IweoBF;
zF3>anp$wO;Ci?;QySdatySU((?HO#~O5NeWG`K>$-tb{B4=YOp3zJAa#p^e{Uu31a
z+L@f!y1pEh61{YXy>FMI5~XHvsdC9&OJILP<$TIW4dK6oRR^_VT=F1_ki-ZfD+^5)
z#N#x3?c1_rWi6a<veNPKsK+ge+rd84k@P0ERd{$z;P-=(qI9YL9M6wA`i6=TBnys2
z-1%S7xB_^eSIC^>h6{&6QOG$aC_7~zSGr{y<p!4t$7vgTYZkavvuwFgO?IilnD<{d
z%IQA+2%W|wAy$Fa7Y?DVE^GvP&dFOPK;-1Q33>(x$BjPkN(czxG8Q#Qlm3_<hlRnK
z(5vVLQN-(r@{$@|v67RJg?;I(dpYTG1U@E(Kg3)KeB&y_b=*=hVlyALbut=zdFCKY
zOzt8IaeGB=*)z6LX!|xm!xF6{E4$K=E;IbS0OkIGmK}Cwwc#3<zm>YOGS;ik7o~C5
z>h8D>j0F7kHJ>zc#tOIj4q&U0(M-y49(r#XAhtBBtLaJOR=}g<(Und&cWfd`C+x4h
zpZe|{AeHNuaSaYXhT*4Y)M=<qXg<mR@a=U4zs#JMeIahehLUi?`v7{?oDU=YQDg0o
zI2pdUoYx5Z%SbCt894Kfz2l`*!0#C>eyq_VFARO||CV^4iMQW;g`hxxZBSE2N}xh3
zAvI|!ayo2!v+rVMN`580v|jC)K<ZfiUS+XBT#(a|oKQY#>Akdb^bE5->k%I;XvSR4
z_P9>HU3a2y@20r>NcsJZW()OG*4#nEZ3kmx;4)=W!I#R#9EQ6mIis;?b=U(8GOEGe
z4YG<xbjFi6J9MKsPM+@~4G~#BaTs5ubli!NI1jKfe&OlxKG~01#e<8NIx_opCa<NI
zgZQaZ|Bn)!*A(;bYDiXR`pQONZ;>ME+9%SqS__}*HwsA%V}D!2HY9(bNgP1qV}{Kg
z-NU8P!GSF76kXq`I9HFmOR*!fRTb>{Zn{N^wWY!ED5%5I^Yfdsu22UL7V_8-PNl@m
ze8Qn5^H0&PN!FRs4<|59=3Z3Qv!8`j-&=@zs)1N8J%+KV&8T0hOy7L|HKtI$bH0K5
ziN!Ik9|yF>#K{T2Ws3FeTq?T>B`5H~r8jblOP0)*Rqr<=3klurRFO}SRrM)z6;Dpt
zq=br(4>nl0f-AUtMwdft5Y1$HrLp8@nI<HtwhwFH!!R$0q`A?h2DNwzC)cRxG^W2Y
z+9XA>SKQ$*ksW-6TwdOPq}ZLM$KNvfg}nTpp(KpV1ZpG4r)HPU?CUV6Rfvg7oZk0A
zOh-4Dyn4qzlq{paY79rep}5R|s@5-DEpUIM$I$8ZVW^LDie3j`*n6?6Z7@q#d2<@;
z7D`_xZsR2vQ`xqvvIu+dlE+t1WDgKT^Lxy#^;%u8b7WAL{Xr}iC4FX=GD%$5tHdbd
z?NWKfk1h;vIt>jZBbAuYKC6jkd^h1zzD3;o$*h#EFNP2$H|eZ+15q#ry60uBBNvM6
zZ;i#mK;4vccSp7V90O7at|z1`_X@E7YMVi8Zf?f)ytnjNtfgA9EUdqk=`5o6^x}jp
zGk0O1A~lemkPCD_f;1Jj1WM}e6o>2++a9yo?0x}$$A}P(>=U$PWUkh?bDF+Ci^fSF
zYnk}y`w*84(OgFnt4{qRBIO9pZ)W`F49@r1-R%;W`5Sq&hGlUnDn&hbm7Iz@))aB<
zEE!#8{LeR;=oi*_B;J-ZkjG#@I@cD99Y@tUEhiv!AH*`6-_<0wn>y)f=zgDicS+E7
zD<*zgJDu0a@T`Q&bSyhC2}@4R_1*`4Tg+M6_gD(&I06%z`cBenk+^4X5V;p7>L0w!
zGb)(=Va*$wW|6xpgpO#foDy2NK30k2J*p^F-Ut)DBT6(?9Kv#-&vsucgSd7l+MiD+
zY<+j}VS@sr)x}p$jdw9i`jciBTT$!`bXI#(yH9d*3A--Ts)V0wH=v!T`rvW5gv}}K
zFOqGru#`deyaxqC$=={@Rg&A2rw}B@>gOAsGSO$8)RsP*7$(bs)P&WeWjqgFKh_S+
z`O3}Ie5Z~rC^LTl(b^%4fww>>d5;&1pOi0sDX$u@>f<S7tL-cig?v)cvnE4GtET&1
zwZcKdkM$*>S;<Lu2ECSzUE|HM8#)P|6YKil55l!_IP-1pd^Ylu-seQ8i8mV|TrB67
zZ%{jE&8FJ(S^YA)axrqJ>D!NN&NSs?lHHD}o^P<5OvQJx4IdJIPT*Jaz+|rgeh58v
zxIINUOkjS`bYrtt2&p^EG>Of-T=3!7hOxfv3bz_9#cVLGcGZc~8Ggj%6QvlPXLmz7
zM2zO;_*WGmx5upWhN`(zZ8W?gRSXx;T2UM@rzah;3eKg63;d$c_J-sqRN92I<J%HB
zIkfoZzE_}H_yD`tK9{7vzWQWDsYIJ(%E=m6k@@i{#a@+EtJ!TSgCXtO3~UaI+(<I@
z9<ozf52G>%Y@Hxq0jmh=?UUhibN9iKWF@7`_aWuG?B02nOY781xdCcCF#_6ov*EHk
zk%7abhr23^@@S`cK{;Vgss%4@sI0be3UU0H)O%?#q}tBD20FoT8)$b&K78c;IOC);
z)gYCAurz@A5vz%3n(zxR9@IINK4$Aa3*PxWo=&W1Kk$?}LpgDB>xN%*e85}beU6RE
zfNsF*LsIRKym;4HRbLM(2379=E+vZ+^L|&@dJXRji6@2fQCR7!uw|^}^Nn<imt!-Y
zEq$Tbk=wnWKTq#8Ignaf8*Wz;vVAtKCsuq-eB0EsFNW6kL^j4H&eW-SU&+u$tZ(!V
zF8RR;OG`O+V9WvWC~bd|UDIb`PoMH8Yx5`*1)YgpbFQfmH$MAqKMrRO`p}bxQ97R*
z@FJYpBoEff%t_8~*byfZyD!zL!MdBP#?(^PAl-I?`2KLby>abU_XixignXU>gHSE2
zoN_W@Myw#g$CUyLpFki4<vZSQy!5x~J;tKIk&GVhLH>is_dJn9o&>)|95v-Szld9S
zyYnW48|jjjVZ6KbrLc&-acj~(a@jCMTMpecD>h@M_vpe<#`ZN|_Dy<3-umiSDW8N_
zpcOKq3*T<uV@o(_eG`KEWiYiH*Ws2QcI$5S!|*yE3zK^eTK$FhSDsUQnpd>vf66~H
zISkyqP?~t9z;N0sDbDlm(|rk>S9q=|K3?+XvPrS7;zVOx$Nr3^gZKn`bw3!qPS||V
z?#m@_9;`XNXXv)J%$Lnf9#F~iyitlfC2?1ZB7f0nlviKKv<WTaa0vgYS%!-WbMdH;
zY_`5d4*R~`I&K4A_eyW^z;~f$ttE%zSDN!BRMa=S^n!4UrNd^EPAh9<$==i<k@FnN
z>+?9P+{9H)d)%w~!!l^wgA<gTZcfIEDA&$`U)GI+@8x@6NjyGkK+i$krod|S-sY^)
z<8qel<~xvvNP?Rf7oDf_A;V7+zwX(IXN<eHBq5#~vJuv1Fb&0<gy99@6yMe|mbZOH
zf3VAcd{{S}-e&YZkDMm5M%>Cx0r}N3{fZSOiPSWqBd%*4Q~0|_({fFWl@ilCqbRHp
zLF?U;XYDkuGy)0wBD)Es-Y>yc+K6ptU5d9UgifWoqnSq22Jj4mL`0CM3Qe$PNQq?<
zZm^SJevQy=%y}Ojoe@2x@o>yQ>iK5bO67}Ptkr3rCt@V|8V~DZ{lWMWjbfl^5<SO}
z8;AGR#wPq2T$qV;-LhlNJ~xx=p2a9XIw2#DNKQ_Fjr8^<ikgS5jxJg7;THSBcGsK|
zCpArTMOy|A1+3jub6BO=Kk@D;BX-G~)6+ZWR9y)Jis@lrky_?VnpWiFd!tZ6BqFp%
zgdg`OrucT&<7xaIV(g*E+dneZ42+aXA3l57G2TD{t#!nA;E?o2A9D5{QQrw|`#$Dq
z2x;4dF$d($+oX8WGz@4Te=Q%9>6tIKB&g##Eju78R&E?GI2CE}ifjB_K{+mSDJ15S
zRDDQCp1J<OzX3fSFQ|Dv`8a(N;=m)my9i;KNCi1`Xtun0qp>_CvpRd1TgWMA4`n)V
z_JG7+{)po?TF60q)^`OSs<B?N$w*Z?UEK2$)P!Pry_GM~)a}je;+GbF(5~E@dIM7n
zJ-R`|;DlX}>{|W7{$!x*cSbY&i`-iqc{hyoKlF|i)5LajqO~w)vG*pqM@Bdo*{^aa
zVTw}6gcAxMM!r~DN9?q+aF#}p(6wIG&Q!2LfAY+QM$Ty0r{SggrjrT-O!l<`p=_5g
z7pv6b=7zPTu7llX<G8DG-91tv3Xqja|BH;kUG459JGI$@t*6quKlC8odSi(vXQ&3o
zO&Mrz)X{C6Iz*1oSd^t<W)Dd<TVoiBO@dd-1u;LhkvwM<cijufyk*!eo7!8}O6dg?
zdt4kl4aH2vFmPm<u_`T@%fT-(%`8;d3p9`r1h2=q$ZZ~cl=tO$V^{IobJ4FI-Ai{k
z>tR`0jstrM!|@rnbBb+3dTievpGKyZ9f2D9%pP50Xl_u|$rGfWL+%;WXo^*%*XwVk
z-L^N3Vi+dgATLkeE9^DQ+fvv|ID0aS*f@YO8WQh|@ruYJt?5M2Yi(hWi0buwcHIb9
z1_hOv_|mGe^wnn+`wFi%_n7>JcT!*4H*DMyYe^nwWWpE1E8NQAd0(+U8p)3m_pujI
zZmHok{n5T1>sgxC?pY#V#&UaA*WMi}mi=uZue+*aUt|O`D2>%QHhXi@T!LxDs@WPv
zmCxFkW(qh)r5y@Z>^g2RH5s1aj^{0krlnjWt>u{S|C@#J&%naJy($0Q!YCu7rY<RS
zWnl#T0RPp($j%41F!BN}!+>3o2gnEZ|HHxvm?E`*`Wkb>{f7Tf3nSb@_@{-D=ifc0
z|KT$VK>nQ#?jP1vz=-+>O|A{Ifa|3+#1ZBMM~9LPh@--y2>SDr7R1cS5(Y%mzK&1x
zI}{DipRUcniT<C^`3LF$i4L5L|1CN|O0>UI2k^`OZGOE*M+Ia7hEW2PhC16h!L7&&
z5HlEHK?VX;!TZ9aSYN#Yv0koT0aG&k)c~-%!iDTA-p*CqzsK8Ac6PFX*a48Q2}gO2
z$rafsf0~7V4hawB{}-gc`J$>roB$&-5JLt&9*Y<dApH8B3dj)*pvwLeLlzZBunUl@
z?yB-PsJ}UtUxNZ9_9v(-e|iAuzk|A-Im=_Po6}#o{s9+W`)jUODE{T0D+Ywv`A;bR
z#QzHJ|HSui7XYr|!aZ}Zrv5WX{J-OSl@S7Hb`AHR68QIg5m#`3=3M|t`cq#27q~ws
za}D<z`QLQ-TjtVh`u@z@2#5QhSoKpBuE74p{Wn;6<=?6Mi)lZ7p#RCUYdF`q{)WR2
zgh2WY=`qL_5Jf#oT1luWlbRFA(TR={2<Q%{E-lE^+79Y!12(g;1;h$Q2Z;RjkhB2*
z!H#wy8zyHMut?w)*P#n&&7JMcAa)i^FeeajF#_lX)cp(uMQaZ<Wo84wUo-FeGdsNV
zHD|61IDi7+n!^<WS3Nj^0wC(rRe>YeoJkF2;{tMmfF0Rc*|>o^Aa2vIIv}>883bPL
z1h#`g9nEc^u5|F+ioYA!LSgKz?Cd}_*w(}b44)LRN&l)hu>mG+1#$!00GI*+Ab!_C
z9ic7|Czv(F9bRb*%+Jvd=>2o8(ZW$P1v%OSbG|~22Ot5w9*EHY3x*jOY;Oy(GXnsy
z0)qep{3IK^vl$oy1KT-5oh*R~u-QXwOrb!7KPG2pZVMmT31a7NVGkeqT9mFvHiMeM
zq5gede*iZJK^#rsu;I9|vs-|iVK4~D&fXDf3WmY1$N)s*|AnYI$O`Jr&I*F#|Hmj6
zAV)L1tJ(2d*t<JH9Zi8-Eo@i8s{nC<PXZ_gu!pnk=Q6%3wFJm*3$g=hIQ|+Lj{hHX
zyCOLouLZ=((%A%<_D_j`gR}+N*Z?X83iRf&0fQXv0QrLV{)0CFGk}Rf?b!IZIqV%F
zcBT;fE7tP;AvQo8pph%UeKsBo`)kQy`?=d)A!!RWwRVQF@$vA2?CotJrhtq=0R{ru
zg@2I+SbkTc!TJ}TY(F>bD**PES5xFOh1%LfVIV-d;AsBQ-5y~575&-%paKxL>lgfA
z4cK8$j$kl+@c{A1;MKqhLJ?vMh~Gb!>h-{X!Tr_k?>gvD#s6!)!%6E1g4qIs00ub2
z%KaNBAXn%wl>TVK#`?Pn8<1t@R~r~WSUZ3}?Cd6v&JZ}O;S>L3RsC+@1U4~ocDkC_
zAG~n^*;#^|KsIn5b6D8JoSe<g;UKQ}!>c*ELTs!7&`<zXb~Y1`I~ZmMc6EX?i0!BH
zf;R&L`VSCjc;RpGS08_}0R}dOvz7j8AG)5wucE81=pSY9-RP&F{b~o7?w@PrR~Z)_
zpmqPz7ucNsS{lE$Lnf#>lN}grMhA}`cSS&$v#l-2(H+<pf5LQy07Ak9gPJ?J0_vOz
zYzl?B!+^~c=yEM8f0Kp(z((5t9n#<B1J267G|SI3&b5+I1txXP#=njEhpPCm$M{ow
z0f$IHklBBV(4U$KPJ+KMxjuwRT*-w6{6GRfja?nn0FD0Z5cTV<b1m5HKV=5qz~0$O
z4%mzV%>^u)-<JpftN~7&lq@ppYDOxO(qh2T7mnF=y^15$9DY;pn&dyhNk9SZ2A3Z>
zGXN~a98f}(v@lf}ao}#BoSixJuMfbrJAhztjFo^mj4T>3@L%|{D1n_oX26^Pop!x~
zSd;;kA$BE|K<<_+nfQYuSG(^oionk(wE7%8+`vDNJ}WD$4E&)2f5^ihAo#-#{@{l{
z^jSH9e^yqpKR;Y`xoYSPFNXbkK;ZALD(wDje+AtUUIK+bOyLhO{PDZzU*kx_i?5nq
zfwBG7)D`~js{2)~HT>gWovw!c+lW_Fxx!hSjs*yYXa%fUS~?a9M=%IbIkbu(b+9(2
zE1)STErCQtj=&0_15Ub7GiOsUQ0okm6$ctHQfh)7VF0Tr*_hclnAs_5;k^_gF2LnW
zxNZkHb_Eb<Aql8*;QYf6-@sUT*tywRIXGC^b%7QD_x|@vb>LE`Js2M37Gwhh1MF6p
z(q@)+wy_Zh*#Sxt7|;>|qXho`dz5}No{js@1#`U=lufMEoNR&9EIeEF&o2+0KzgA6
z7;^?V9{rII?03wWP3>u``!YD*<61yYS9g1(;*)z3ZH**Q?hQOsnkX|iv4pNFe0OO}
z>G1PIKAIk^)L>R?rMLcq5ht26-VF^W>hb!<;CKwf`i|bttO<ddr|7S<BIqYW*o)HG
z<-NF#DPP7C#YD&LYx)c>dK1S7R!M3EFotsxTL+tHh!GR#VD=<|(OV>?2xRjf?6alF
ztr&-=Aw=vFBuY;p#fQ+)(p6yzs7Xs9urVOgt0U6;t2@wQBxDtaAQMsEv8@y9eDtxO
z2(dUi{A(!9hxcJsS#MixvTva9J&ZvYW2!=O>Yxg*YFY>^Viq%%Qcuuve{`=6FQex<
z>5^B4W`DkyqGN8B7Z&wKmJy<=INb+h6ML*w4i6Xk1Nx|&%r6=#{gGyFzUB;IM)D*E
zsY?z_h`0?sv{0`~YbwN%+<#MxQ$<2X!La}7U9sv)O<C&WWUaoF23_BpH$I+_7M|^o
z)|`Hus>$J2Q$;TZ@xP&xA>;Fg&(yvgzmaMf8Al8)<1dBG4y=oOOx|1he&p5|&X?}8
zpU-W&zuzfv+|shypmF+&ZM;vPXFb_La}kR-S!#Ms8OwU$)#=KTS?-PyO6dzR6n)<6
z@t%i#Esw>Rnb%N*gzs$I%+A=AeXptK80T@#PaVgu9;M<nc+D?>_hD+`slnRPhdWz|
z0kN;famOUs))9UDxV|y6U>;$-`GVo#Bc+*z_|0lYv1HA|a6s3)PJey0!AcM5PPaL?
zjxgzrB4OXN4aqqdR|Q29@yW6WN|b~vsCgJq@tKqtiF+^f#}}v9R?H2V%0-%ZpaT@Q
z+SbCb9v=nO+TRtx2m2jL_g5y@j@A@eXyizY=?Dxiij_0_-=3Ec#iewO!e)K?nK}jk
zS)0vM9ZpKLb#GJ@9j7tOcLb_;T2PizS261S$1V&Cp-VnFevK+CiAKJ6luI~dxi}2(
z*wxQ6Ca&#KbS!ELqPe|^ZCSt~Mba8n6KsDV*koqp(97S>@B<&47+O)s8cXOpr}g>P
zrUbP4=qsng_EV20+5YCy?oKPTn3J&6-0b7o1?F#o-MIz4Qe!j+V*+gqK@%hq47ojr
zx1Lr&PLWT;Jl~{UggIg9e?QdISaGhl8Cx7R)1{kL;?Hz4s4XMdOqrW;g?7#4mWbXR
z&i<&Q<|_Y#UqUwFP%oWoMB-y3)Ng-q7Q3Ri@uUZh^4zT!r#<?3d+Tk`H^^7gGN?Y<
z-Q;~93*(BlmOI%!r5@+wH(Z-sA|Kx{a@ClrT_AI5G*Gv#)=@5TA(^&q6n>lf*pTZ`
zTPqMJ9U_jO()_7nN<qi^xfD+O_EKHkq@C$JSyi-UjSY!-=rsGJ_qQL*rXt%4tt{_+
zW=(xR*HLtRYH5yb^vX!x6}vC0S~*X1QZT{H*R%*Sd14*=C1-v!T8VP?{>eG3#s1em
z)BfqZ;zI%T#E%dlc+HaCX*l2NUz#vds_jN-R@ZP)M~P8?()si;As$jj7<(!fr;zn{
z2kq$+YQ3^q;A7)}ck$tzOxheea|0rf2U(~DQFl(C)+k^(pxE5awxsU4Y)ZIKiTB)%
z<DIy)Flntb%_k}VU`mzSIA-v(2i*H(I328|B<;`CQ2RgkH5y>OeNg!7ovSu;H$vFi
z+`Fuxam1HC!6PI-#JgKK6|oBD8XD`JLp7tp%?*CE^xc^#l&A|w9k#;J+Or7dTfu#f
z!tMf-HJe_l;TQpBC1Vkd)5Uh*P53$#zLRqXzs=0E{q~KMg~Yp}l$hU~sNrn+;3)A1
zFA47QJQQ!XxBW$f*oE-K<?VM^`CCW3s4Td9mwYCdq-^mly|^NG=N@@4aUI-!F)zQy
za=UuJsE)G=?dWUE!i`DEg<(DYCLdo2;i^<~r`T=ogRyXBOog82&fQyY>R7YP`Y*pV
zttdw?Qlb9KFO)#o;C~*Bv{k|8@Kc8w81B`;O34m<(F6`1Y7h@V!vXJs6n1))Ksq?!
zA1GFZLgBv)T^}*8&noOczuB?KNy5)ZY!aM&5>jjuK&(TchLeYngOiV2oQ<1Jl3QGg
zOPp2szi9%D^dAGD0pEsxxg`7%rI75CoUDcee%J2hWphoHMe<Zui-lFnip`{8GA!rE
z$6m$iDdbwzFzR;nsik7fV5QW69lt)=Xt8@EFAbu_!rx0Vy#%Q(i5Kk)rDCu8&NjdB
z95^;uO*>sh3YvLbf8#}(EA!;dv~cHHSHs%kL6rLvn(nBgl|QCR?*`Hh7B&;DcZxDf
zdb4_c@K6KY+ZZvdKFF6K-PmiZkz4Qciz45iW+OLmXH2>s8im}$in7u!n1oFig$}`a
zhRW!N*&WEJjXD{H6mCox_l{i~i@}(tG+?yi)>u3Lb`}3-GAw(f@r{U!KIXCTj)S_)
zS%ZW-o{MnOo?c0t)hkt&3Fn)_gZuA_ohW!G!5X-6N1~i2U4rQfFE!9+S0AeM;LUsI
z;o$idg(_7A^-4F9(;MGc7bj4cxIH2+WE0R=ar1SXA?<$4a+#1Al8qn##0|Y{v=k7}
z#;$*gL!+s{#9|dO2--dy_zfL{U{)*4-#{K!yAbCF9nudfNzsXxiu+U~?Y+6k^8WUr
z%Xzn_Kq91m5@%NX_eaFTg77gW391+p(lobUe4lnsn8D)hjvf$Y969x3Jxno(!3tDB
zO~hfQ(ieXdXpc6DgGZGs*FNj&UG0hC=Y)P7E7%<za@cHf%<D@Z-%)TBp{kKR@`KiS
zQ(gi{KK7FHA>B~0{3rM2qr1g*y`i9G^;?SciLLYnzH~DKakJs_M);BRNCWW~tzr6)
z3Mk@Ik#^LH$<^$F<GK#J_mRL$X?yH?#AK?5e8a+J5zlG24QZjwblb;?%CCAmviA6b
zp;U$?Jq)^eWSZ1><W~Ue8**2cau06a*T_?By+?_!{^vRazNR@kxd8`Cbaqx=4pt6y
L8X75OY4raGG=vt6

literal 0
HcmV?d00001

diff --git a/setup.py b/setup.py
index 13af1caa..080c3e41 100644
--- a/setup.py
+++ b/setup.py
@@ -17,7 +17,6 @@ def parse_requirements_file(filename):
     "optax",
     "chex",
     "numpyro",
-    "tensorflow-probability>=0.16.0",
     "tqdm>=4.0.0",
     "ml-collections==0.1.0",
     "jaxtyping>=0.0.2",
diff --git a/tests/test_config.py b/tests/test_config.py
index b2198a1c..2dcf5d6d 100644
--- a/tests/test_config.py
+++ b/tests/test_config.py
@@ -1,26 +1,25 @@
 from ml_collections import ConfigDict
-from tensorflow_probability.substrates.jax import bijectors as tfb
-
+import numpyro.distributions as npd
 from gpjax.config import add_parameter, get_defaults
 
 
 def test_add_parameter():
-    add_parameter("test_parameter", tfb.Identity())
+    add_parameter("test_parameter", npd.transforms.IdentityTransform())
     config = get_defaults()
     assert "test_parameter" in config.transformations
     assert "test_parameter_transform" in config.transformations
     assert config.transformations["test_parameter"] == "test_parameter_transform"
-    assert isinstance(config.transformations["test_parameter_transform"], tfb.Bijector)
+    assert isinstance(config.transformations["test_parameter_transform"], npd.transforms.Transform)
 
 
 def test_add_parameter():
     config = get_defaults()
-    add_parameter("test_parameter", tfb.Identity())
+    add_parameter("test_parameter", npd.transforms.IdentityTransform())
     config = get_defaults()
     assert "test_parameter" in config.transformations
     assert "test_parameter_transform" in config.transformations
     assert config.transformations["test_parameter"] == "test_parameter_transform"
-    assert isinstance(config.transformations["test_parameter_transform"], tfb.Bijector)
+    assert isinstance(config.transformations["test_parameter_transform"], npd.transforms.Transform)
 
 
 def test_get_defaults():
diff --git a/tests/test_gp.py b/tests/test_gp.py
index 5a940f53..c99242e8 100644
--- a/tests/test_gp.py
+++ b/tests/test_gp.py
@@ -1,9 +1,9 @@
 import typing as tp
 
-import distrax as dx
 import jax.numpy as jnp
 import jax.random as jr
 import pytest
+import numpyro.distributions as npd
 
 from gpjax import Dataset, initialise
 from gpjax.gps import (
@@ -32,9 +32,9 @@ def test_prior(num_datapoints):
 
     x = jnp.linspace(-3.0, 3.0, num_datapoints).reshape(-1, 1)
     predictive_dist = prior_rv_fn(x)
-    assert isinstance(predictive_dist, dx.Distribution)
-    mu = predictive_dist.mean()
-    sigma = predictive_dist.covariance()
+    assert isinstance(predictive_dist, npd.Distribution)
+    mu = predictive_dist.mean
+    sigma = predictive_dist.covariance_matrix
     assert mu.shape == (num_datapoints,)
     assert sigma.shape == (num_datapoints, num_datapoints)
 
@@ -75,10 +75,10 @@ def test_conjugate_posterior(num_datapoints):
 
     x = jnp.linspace(-3.0, 3.0, num_datapoints).reshape(-1, 1)
     predictive_dist = predictive_dist_fn(x)
-    assert isinstance(predictive_dist, dx.Distribution)
+    assert isinstance(predictive_dist, npd.Distribution)
 
-    mu = predictive_dist.mean()
-    sigma = predictive_dist.covariance()
+    mu = predictive_dist.mean
+    sigma = predictive_dist.covariance_matrix
     assert mu.shape == (num_datapoints,)
     assert sigma.shape == (num_datapoints, num_datapoints)
 
@@ -117,10 +117,10 @@ def test_nonconjugate_posterior(num_datapoints, likel):
 
     x = jnp.linspace(-3.0, 3.0, num_datapoints).reshape(-1, 1)
     predictive_dist = predictive_dist_fn(x)
-    assert isinstance(predictive_dist, dx.Distribution)
+    assert isinstance(predictive_dist, npd.Distribution)
 
-    mu = predictive_dist.mean()
-    sigma = predictive_dist.covariance()
+    mu = predictive_dist.mean
+    sigma = predictive_dist.covariance_matrix
     assert mu.shape == (num_datapoints,)
     assert sigma.shape == (num_datapoints, num_datapoints)
 
diff --git a/tests/test_likelihoods.py b/tests/test_likelihoods.py
index 57f0ea09..e9b48f32 100644
--- a/tests/test_likelihoods.py
+++ b/tests/test_likelihoods.py
@@ -1,8 +1,8 @@
 import typing as tp
 
-import distrax as dx
 import jax.numpy as jnp
 import jax.random as jr
+import numpyro.distributions as npd
 import pytest
 
 from gpjax.likelihoods import (
@@ -39,8 +39,8 @@ def test_predictive_moment(n):
     pred_mom_fn = lhood.predictive_moment_fn
     params, _, _ = initialise(lhood, key).unpack()
     rv = pred_mom_fn(fmean, fvar, params)
-    mu = rv.mean()
-    sigma = rv.variance()
+    mu = rv.mean
+    sigma = rv.variance
     assert isinstance(lhood.predictive_moment_fn, tp.Callable)
     assert mu.shape == (n,)
     assert sigma.shape == (n,)
@@ -57,7 +57,7 @@ def test_link_fns(lik: AbstractLikelihood, n: int):
     x = jnp.linspace(-3.0, 3.0).reshape(-1, 1)
     l_eval = link_fn(x, params)
 
-    assert isinstance(l_eval, dx.Distribution)
+    assert isinstance(l_eval, npd.Distribution)
 
 
 @pytest.mark.parametrize("noise", [0.1, 0.5, 1.0])
@@ -65,33 +65,33 @@ def test_call_gaussian(noise):
     key = jr.PRNGKey(123)
     n = 10
     lhood = Gaussian(num_datapoints=n)
-    dist = dx.MultivariateNormalFullCovariance(jnp.zeros(n), jnp.eye(n))
+    dist = npd.MultivariateNormal(jnp.zeros(n), covariance_matrix=jnp.eye(n))
     params = {"likelihood": {"obs_noise": noise}}
 
     l_dist = lhood(dist, params)
-    assert (l_dist.mean() == jnp.zeros(n)).all()
+    assert (l_dist.mean == jnp.zeros(n)).all()
     noise_mat = jnp.diag(jnp.repeat(noise, n))
-    assert (l_dist.covariance() == jnp.eye(n) + noise_mat).all()
+    assert (l_dist.covariance_matrix == jnp.eye(n) + noise_mat).all()
 
     l_dist = lhood.predict(dist, params)
-    assert (l_dist.mean() == jnp.zeros(n)).all()
+    assert (l_dist.mean == jnp.zeros(n)).all()
     noise_mat = jnp.diag(jnp.repeat(noise, n))
-    assert (l_dist.covariance() == jnp.eye(n) + noise_mat).all()
+    assert (l_dist.covariance_matrix == jnp.eye(n) + noise_mat).all()
 
 
 def test_call_bernoulli():
     n = 10
     lhood = Bernoulli(num_datapoints=n)
-    dist = dx.MultivariateNormalFullCovariance(jnp.zeros(n), jnp.eye(n))
+    dist = npd.MultivariateNormal(jnp.zeros(n), covariance_matrix=jnp.eye(n))
     params = {"likelihood": {}}
 
     l_dist = lhood(dist, params)
-    assert (l_dist.mean() == 0.5 * jnp.ones(n)).all()
-    assert (l_dist.variance() == 0.25 * jnp.ones(n)).all()
+    assert (l_dist.mean == 0.5 * jnp.ones(n)).all()
+    assert (l_dist.variance == 0.25 * jnp.ones(n)).all()
 
     l_dist = lhood.predict(dist, params)
-    assert (l_dist.mean() == 0.5 * jnp.ones(n)).all()
-    assert (l_dist.variance() == 0.25 * jnp.ones(n)).all()
+    assert (l_dist.mean == 0.5 * jnp.ones(n)).all()
+    assert (l_dist.variance == 0.25 * jnp.ones(n)).all()
 
 
 @pytest.mark.parametrize("lik", [Gaussian, Bernoulli])
diff --git a/tests/test_parameters.py b/tests/test_parameters.py
index f384d046..4679fdf4 100644
--- a/tests/test_parameters.py
+++ b/tests/test_parameters.py
@@ -1,7 +1,7 @@
 import typing as tp
 
-import distrax as dx
 import jax.numpy as jnp
+import numpyro.distributions as npd
 import jax.random as jr
 import pytest
 from tensorflow_probability.substrates.jax import distributions as tfd
@@ -164,7 +164,7 @@ def test_checks(num_datapoints):
     priors = prior_checks(incomplete_priors)
     assert "latent" in priors.keys()
     assert "variance" not in priors.keys()
-    assert isinstance(priors["latent"], dx.Normal)
+    assert isinstance(priors["latent"], npd.Normal)
 
 
 def test_structure_priors():
@@ -189,7 +189,7 @@ def recursive_fn(d1, d2, fn: tp.Callable[[tp.Any], tp.Any]):
         assert v
 
 
-@pytest.mark.parametrize("latent_prior", [dx.Laplace(0.0, 1.0), tfd.Laplace(0.0, 1.0)])
+@pytest.mark.parametrize("latent_prior", [npd.Laplace(0.0, 1.0), npd.Laplace(0.0, 1.0)])
 def test_prior_checks(latent_prior):
     priors = {
         "kernel": {"lengthscale": None, "variance": None},
@@ -199,7 +199,7 @@ def test_prior_checks(latent_prior):
     }
     new_priors = prior_checks(priors)
     assert "latent" in new_priors.keys()
-    assert new_priors["latent"].name == "Normal"
+    assert isinstance(new_priors["latent"], npd.Normal)
 
     priors = {
         "kernel": {"lengthscale": None, "variance": None},
@@ -208,7 +208,7 @@ def test_prior_checks(latent_prior):
     }
     new_priors = prior_checks(priors)
     assert "latent" in new_priors.keys()
-    assert new_priors["latent"].name == "Normal"
+    assert isinstance(new_priors["latent"], npd.Normal)
 
     priors = {
         "kernel": {"lengthscale": None, "variance": None},
@@ -219,7 +219,7 @@ def test_prior_checks(latent_prior):
     with pytest.warns(UserWarning):
         new_priors = prior_checks(priors)
     assert "latent" in new_priors.keys()
-    assert new_priors["latent"].name == "Laplace"
+    assert isinstance(new_priors["latent"], npd.Laplace)
 
 
 #########################
@@ -233,8 +233,8 @@ def test_output(num_datapoints, likelihood):
 
     assert isinstance(bijectors, dict)
     for k, v1, v2 in recursive_items(bijectors, bijectors):
-        assert isinstance(v1.forward, tp.Callable)
-        assert isinstance(v2.inverse, tp.Callable)
+        assert isinstance(v1.__call__, tp.Callable)
+        assert isinstance(v2.inv, tp.Callable)
 
     unconstrained_params = unconstrain(params, bijectors)
     assert (
@@ -252,5 +252,5 @@ def test_output(num_datapoints, likelihood):
     a_bijectors = build_bijectors(augmented_params)
 
     assert "test_param" in list(a_bijectors.keys())
-    assert a_bijectors["test_param"].forward(jnp.array([1.0])) == 1.0
-    assert a_bijectors["test_param"].inverse(jnp.array([1.0])) == 1.0
+    assert a_bijectors["test_param"](jnp.array([1.0])) == 1.0
+    assert a_bijectors["test_param"].inv(jnp.array([1.0])) == 1.0
diff --git a/tests/test_variational_families.py b/tests/test_variational_families.py
index e6f86827..5def2d16 100644
--- a/tests/test_variational_families.py
+++ b/tests/test_variational_families.py
@@ -1,7 +1,7 @@
 import typing as tp
 from mimetypes import init
 
-import distrax as dx
+import numpyro.distributions as npd
 import jax.numpy as jnp
 import jax.random as jr
 import pytest
@@ -128,10 +128,10 @@ def test_variational_gaussians(
     assert isinstance(predictive_dist_fn, tp.Callable)
 
     predictive_dist = predictive_dist_fn(test_inputs)
-    assert isinstance(predictive_dist, dx.Distribution)
+    assert isinstance(predictive_dist, npd.Distribution)
 
-    mu = predictive_dist.mean()
-    sigma = predictive_dist.covariance()
+    mu = predictive_dist.mean
+    sigma = predictive_dist.covariance_matrix
 
     assert isinstance(mu, jnp.ndarray)
     assert isinstance(sigma, jnp.ndarray)
@@ -194,10 +194,10 @@ def test_collapsed_variational_gaussian(n_test, n_inducing, n_datapoints, point_
     assert isinstance(predictive_dist_fn, tp.Callable)
 
     predictive_dist = predictive_dist_fn(test_inputs)
-    assert isinstance(predictive_dist, dx.Distribution)
+    assert isinstance(predictive_dist, npd.Distribution)
 
-    mu = predictive_dist.mean()
-    sigma = predictive_dist.covariance()
+    mu = predictive_dist.mean
+    sigma = predictive_dist.covariance_matrix
 
     assert isinstance(mu, jnp.ndarray)
     assert isinstance(sigma, jnp.ndarray)

From 171fdd964877fa6182a3049016444a376ecc86c0 Mon Sep 17 00:00:00 2001
From: Thomas Pinder <tompinder@live.co.uk>
Date: Sat, 15 Oct 2022 20:13:59 +0000
Subject: [PATCH 08/14] Update workflow

---
 .github/workflows/workflow-master.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/workflow-master.yml b/.github/workflows/workflow-master.yml
index eeab66cc..da8acf75 100644
--- a/.github/workflows/workflow-master.yml
+++ b/.github/workflows/workflow-master.yml
@@ -5,7 +5,7 @@ on:
       - "master"
   pull_request:
     branches:
-      - "master"
+      - "*"
 jobs:
   codecov:
     name: Codecov Workflow

From e96b918724bc9dce31a3af929a18b5770c8aca7a Mon Sep 17 00:00:00 2001
From: Thomas Pinder <tompinder@live.co.uk>
Date: Sat, 15 Oct 2022 20:16:10 +0000
Subject: [PATCH 09/14] Drop distrax ref

---
 gpjax/config.py | 2 --
 1 file changed, 2 deletions(-)

diff --git a/gpjax/config.py b/gpjax/config.py
index fa7be8fa..88592bdc 100644
--- a/gpjax/config.py
+++ b/gpjax/config.py
@@ -3,8 +3,6 @@
 from ml_collections import ConfigDict
 
 __config = None
-import distrax as dx
-import jax.numpy as jnp
 
 
 def get_defaults() -> ConfigDict:

From 1252d46768917ec60bc23e2da3d1f82ebca4d0a7 Mon Sep 17 00:00:00 2001
From: Thomas Pinder <tompinder@live.co.uk>
Date: Sat, 15 Oct 2022 20:20:01 +0000
Subject: [PATCH 10/14] Drop TFP dependency

---
 tests/test_parameters.py | 21 ++++++++++-----------
 1 file changed, 10 insertions(+), 11 deletions(-)

diff --git a/tests/test_parameters.py b/tests/test_parameters.py
index 4679fdf4..94c377bd 100644
--- a/tests/test_parameters.py
+++ b/tests/test_parameters.py
@@ -4,7 +4,6 @@
 import numpyro.distributions as npd
 import jax.random as jr
 import pytest
-from tensorflow_probability.substrates.jax import distributions as tfd
 
 from gpjax.gps import Prior
 from gpjax.kernels import RBF
@@ -57,7 +56,7 @@ def test_non_conjugate_initialise():
 @pytest.mark.parametrize("x", [-1.0, 0.0, 1.0])
 def test_lpd(x):
     val = jnp.array(x)
-    dist = tfd.Normal(loc=0.0, scale=1.0)
+    dist = npd.Normal(loc=0.0, scale=1.0)
     lpd = log_density(val, dist)
     assert lpd is not None
     assert log_density(val, None) == 0.0
@@ -81,7 +80,7 @@ def test_recursive_complete(lik):
     posterior = Prior(kernel=RBF()) * lik(num_datapoints=10)
     params, _, _ = initialise(posterior, jr.PRNGKey(123)).unpack()
     priors = {"kernel": {}}
-    priors["kernel"]["lengthscale"] = tfd.HalfNormal(scale=2.0)
+    priors["kernel"]["lengthscale"] = npd.HalfNormal(scale=2.0)
     container = copy_dict_structure(params)
     complete_priors = recursive_complete(container, priors)
     for (
@@ -90,7 +89,7 @@ def test_recursive_complete(lik):
         v2,
     ) in recursive_items(params, complete_priors):
         if k == "lengthscale":
-            assert isinstance(v2, tfd.HalfNormal)
+            assert isinstance(v2, npd.HalfNormal)
         else:
             assert v2 == None
 
@@ -109,10 +108,10 @@ def test_prior_evaluation():
     }
     priors = {
         "kernel": {
-            "lengthscale": tfd.Gamma(1.0, 1.0),
-            "variance": tfd.Gamma(2.0, 2.0),
+            "lengthscale": npd.Gamma(1.0, 1.0),
+            "variance": npd.Gamma(2.0, 2.0),
         },
-        "likelihood": {"obs_noise": tfd.Gamma(3.0, 3.0)},
+        "likelihood": {"obs_noise": npd.Gamma(3.0, 3.0)},
     }
     lpd = evaluate_priors(params, priors)
     assert pytest.approx(lpd) == -2.0110168
@@ -147,8 +146,8 @@ def test_incomplete_priors():
     }
     priors = {
         "kernel": {
-            "lengthscale": tfd.Gamma(1.0, 1.0),
-            "variance": tfd.Gamma(2.0, 2.0),
+            "lengthscale": npd.Gamma(1.0, 1.0),
+            "variance": npd.Gamma(2.0, 2.0),
         },
     }
     container = copy_dict_structure(params)
@@ -172,8 +171,8 @@ def test_structure_priors():
     params, _, _ = initialise(posterior, jr.PRNGKey(123)).unpack()
     priors = {
         "kernel": {
-            "lengthscale": tfd.Gamma(1.0, 1.0),
-            "variance": tfd.Gamma(2.0, 2.0),
+            "lengthscale": npd.Gamma(1.0, 1.0),
+            "variance": npd.Gamma(2.0, 2.0),
         },
     }
     structured_priors = structure_priors(params, priors)

From df530f6bdc1b56a95fe62bdb648bc4e2e7718b9e Mon Sep 17 00:00:00 2001
From: Thomas Pinder <tompinder@live.co.uk>
Date: Sat, 15 Oct 2022 20:36:15 +0000
Subject: [PATCH 11/14] Update docs

---
 docs/conf.py | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/docs/conf.py b/docs/conf.py
index d84d66c6..f7e060ea 100644
--- a/docs/conf.py
+++ b/docs/conf.py
@@ -103,10 +103,12 @@ def find_version(*file_paths):
 bibtex_bibfiles = ["refs.bib"]
 bibtex_style = "unsrt"
 bibtex_reference_style = "author_year"
+nb_execution_mode = "auto"
 nbsphinx_allow_errors = True
 nbsphinx_custom_formats = {
     ".pct.py": ["jupytext.reads", {"fmt": "py:percent"}],
 }
+jupyter_execute_notebooks = "cache"
 
 # Latex commands
 # mathjax_path = "https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"
@@ -183,6 +185,11 @@ def find_version(*file_paths):
     "logo_only": True,
     "show_toc_level": 2,
     "repository_url": "https://github.com/thomaspinder/GPJax/",
+    "launch_buttons": {
+        "binderhub_url": "https://mybinder.org",
+        "colab_url": "https://colab.research.google.com",
+        "notebook_interface": "jupyterlab",
+    },
     "use_repository_button": True,
     "use_sidenotes": True,  # Turns footnotes into sidenotes - https://sphinx-book-theme.readthedocs.io/en/stable/content-blocks.html
 }
@@ -199,10 +206,10 @@ def find_version(*file_paths):
 # }
 
 
+
 # Add any paths that contain custom static files (such as style sheets) here,
 # relative to this directory. They are copied after the builtin static files,
 # so a file named "default.css" will overwrite the builtin "default.css".
 # html_static_path = ["_static"]
-
 html_static_path = ["_static"]
 html_css_files = ["custom.css"]

From 0d73ef8c247b6b503a8eafae9d20dfc37751cc86 Mon Sep 17 00:00:00 2001
From: Thomas Pinder <tompinder@live.co.uk>
Date: Sat, 15 Oct 2022 21:28:48 +0000
Subject: [PATCH 12/14] Remove Distrax from notebooks

---
 examples/haiku.pct.py        |  1 -
 examples/intro_to_gps.pct.py | 57 +++++++++++++++++-------------------
 examples/kernels.pct.py      | 49 +++++++++++++++++++------------
 3 files changed, 57 insertions(+), 50 deletions(-)

diff --git a/examples/haiku.pct.py b/examples/haiku.pct.py
index 9e014412..42d8820b 100644
--- a/examples/haiku.pct.py
+++ b/examples/haiku.pct.py
@@ -22,7 +22,6 @@
 # %%
 import typing as tp
 
-import distrax as dx
 import haiku as hk
 import jax
 import jax.numpy as jnp
diff --git a/examples/intro_to_gps.pct.py b/examples/intro_to_gps.pct.py
index dc758e99..76f487e7 100644
--- a/examples/intro_to_gps.pct.py
+++ b/examples/intro_to_gps.pct.py
@@ -1,11 +1,12 @@
 # ---
 # jupyter:
 #   jupytext:
+#     custom_cell_magics: kql
 #     text_representation:
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.11.5
+#       jupytext_version: 1.11.2
 #   kernelspec:
 #     display_name: Python 3.9.7 ('gpjax')
 #     language: python
@@ -105,15 +106,19 @@
 # $$
 # We can plot three different parameterisations of this density.
 
-import distrax as dx
-
 # %%
+import numpyro.distributions as npd
 import jax.numpy as jnp
 import matplotlib.pyplot as plt
+from utils import confidence_ellipse
+import jax.random as jr
+import pandas as pd
+import seaborn as sns
+
 
-ud1 = dx.Normal(0.0, 1.0)
-ud2 = dx.Normal(-1.0, 0.5)
-ud3 = dx.Normal(0.25, 1.5)
+ud1 = npd.Normal(0.0, 1.0)
+ud2 = npd.Normal(-1.0, 0.5)
+ud3 = npd.Normal(0.25, 1.5)
 
 xs = jnp.linspace(-5.0, 5.0, 500)
 
@@ -121,10 +126,10 @@
 for d in [ud1, ud2, ud3]:
     ax.plot(
         xs,
-        d.prob(xs),
-        label=f"$\mathcal{{N}}({{{float(d.mean())}}},\  {{{float(d.stddev())}}}^2)$",
+        jnp.exp(d.log_prob(xs)),
+        label=f"$\mathcal{{N}}({{{float(d.mean)}}},\  {{{float(jnp.sqrt(d.variance))}}}^2)$",
     )
-    ax.fill_between(xs, jnp.zeros_like(xs), d.prob(xs), alpha=0.2)
+    ax.fill_between(xs, jnp.zeros_like(xs), jnp.exp(d.log_prob(xs)), alpha=0.2)
 ax.legend(loc="best")
 
 # %% [markdown]
@@ -151,19 +156,15 @@
 # $$
 # Three example parameterisations of this can be visualised below where $\rho$ determines the correlation of the multivariate Gaussian.
 
-import jax.random as jr
-
 # %%
-from utils import confidence_ellipse
-
 key = jr.PRNGKey(123)
 
-d1 = dx.MultivariateNormalDiag(jnp.zeros(2), scale_diag=jnp.array([1.0, 1.0]))
-d2 = dx.MultivariateNormalTri(
-    jnp.zeros(2), jnp.linalg.cholesky(jnp.array([[1.0, 0.9], [0.9, 1.0]]))
+d1 = npd.MultivariateNormal(jnp.zeros(2), jnp.eye(2))
+d2 = npd.MultivariateNormal(
+    jnp.zeros(2), scale_tril=jnp.linalg.cholesky(jnp.array([[1.0, 0.9], [0.9, 1.0]]))
 )
-d3 = dx.MultivariateNormalTri(
-    jnp.zeros(2), jnp.linalg.cholesky(jnp.array([[1.0, -0.5], [-0.5, 1.0]]))
+d3 = npd.MultivariateNormal(
+    jnp.zeros(2), scale_tril=jnp.linalg.cholesky(jnp.array([[1.0, -0.5], [-0.5, 1.0]]))
 )
 
 dists = [d1, d2, d3]
@@ -181,15 +182,15 @@
 titles = [r"$\rho = 0$", r"$\rho = 0.9$", r"$\rho = -0.5$"]
 
 for a, t, d in zip([ax0, ax1, ax2], titles, dists):
-    d_prob = d.prob(jnp.hstack([xx.reshape(-1, 1), yy.reshape(-1, 1)])).reshape(
+    d_prob = jnp.exp(d.log_prob(jnp.hstack([xx.reshape(-1, 1), yy.reshape(-1, 1)])).reshape(
         xx.shape
-    )
-    cntf = a.contourf(xx, yy, d_prob, levels=20, antialiased=True, cmap="Reds")
+    ))
+    cntf = a.contourf(xx, yy, jnp.exp(d_prob), levels=20, antialiased=True, cmap="Reds")
     for c in cntf.collections:
         c.set_edgecolor("face")
     a.set_xlim(-2.75, 2.75)
     a.set_ylim(-2.75, 2.75)
-    samples = d.sample(seed=key, sample_shape=(5000,))
+    samples = d.sample(key, sample_shape=(5000,))
     xsample, ysample = samples[:, 0], samples[:, 1]
     confidence_ellipse(
         xsample, ysample, a, edgecolor="#3f3f3f", n_std=1.0, linestyle="--", alpha=0.8
@@ -236,19 +237,15 @@
 # joint distribution $p(\mathbf{x}, \mathbf{y})$ quantifies the probability of two events, one
 # from $p(\mathbf{x})$ and another from $p(\mathbf{y})$, occurring at the same time. We visualise this idea below.
 
-import pandas as pd
-
 # %%
-import seaborn as sns
-
 n = 1000
-x = dx.Normal(loc=0.0, scale=1.0).sample(seed=key, sample_shape=(n,))
+x = npd.Normal(loc=0.0, scale=1.0).sample(key, sample_shape=(n,))
 key, subkey = jr.split(key)
-y = dx.Normal(loc=0.25, scale=0.5).sample(seed=subkey, sample_shape=(n,))
+y = npd.Normal(loc=0.25, scale=0.5).sample(subkey, sample_shape=(n,))
 key, subkey = jr.split(subkey)
-xfull = dx.Normal(loc=0.0, scale=1.0).sample(seed=subkey, sample_shape=(n * 10,))
+xfull = npd.Normal(loc=0.0, scale=1.0).sample(subkey, sample_shape=(n * 10,))
 key, subkey = jr.split(subkey)
-yfull = dx.Normal(loc=0.25, scale=0.5).sample(seed=subkey, sample_shape=(n * 10,))
+yfull = npd.Normal(loc=0.25, scale=0.5).sample(subkey, sample_shape=(n * 10,))
 key, subkey = jr.split(subkey)
 df = pd.DataFrame({"x": x, "y": y, "idx": jnp.ones(n)})
 
diff --git a/examples/kernels.pct.py b/examples/kernels.pct.py
index 062ca11c..ebb46ec5 100644
--- a/examples/kernels.pct.py
+++ b/examples/kernels.pct.py
@@ -20,15 +20,13 @@
 # In this guide, we introduce the kernels available in GPJax and demonstrate how to create custom ones.
 
 # %%
-import distrax as dx
-import jax
 import jax.numpy as jnp
 import jax.random as jr
 import matplotlib.pyplot as plt
-import tensorflow_probability.substrates.jax.bijectors as tfb
 from jax import jit
 from jaxtyping import Array, Float
 from optax import adam
+import numpyro.distributions as npd
 
 import gpjax as gpx
 
@@ -62,7 +60,7 @@
 
 for k, ax in zip(kernels, axes.ravel()):
     prior = gpx.Prior(kernel=k)
-    params, _, _, _ = gpx.initialise(prior, key).unpack()
+    params, _, _ = gpx.initialise(prior, key).unpack()
     rv = prior(params)(x)
     y = rv.sample(key, sample_shape=(10,))
 
@@ -207,13 +205,31 @@ def _initialise_params(self, key: jr.PRNGKey) -> dict:
 #
 # To define a bijector here we'll make use of the `Lambda` operator given in Distrax. This lets us convert any regular Jax function into a bijection. Given that we require $\tau$ to be strictly greater than $4.$, we'll apply a [softplus transformation](https://jax.readthedocs.io/en/latest/_autosummary/jax.nn.softplus.html) where the lower bound is shifted by $4.$.
 
-
-import numpyro
-
 # %%
 from gpjax.config import add_parameter
+from jax.nn import softplus
+
+
+class ShiftSoftplus(npd.transforms.Transform):
+    domain = npd.constraints.real
+    codomain = npd.constraints.real
+    
+    def __init__(self, low: Float[Array, "1"]) -> None:
+        super().__init__()
+        self.low = jnp.array(low)
+    
+    def __call__(self, x):
+        x -= self.low
+        return softplus(x) + self.low
+
+    def _inverse(self, y):
+        return jnp.log(-jnp.expm1(-y)) + y
 
-add_parameter("tau", numpyro.distributions.transforms.SoftplusTransform())
+    def log_abs_det_jacobian(self, x, y, intermediates=None):
+        return -softplus(-x)
+
+
+add_parameter("tau", ShiftSoftplus(4.))
 
 # %% [markdown]
 # ### Using our polar kernel
@@ -238,30 +254,25 @@ def _initialise_params(self, key: jr.PRNGKey) -> dict:
 circlular_posterior = gpx.Prior(kernel=PKern) * likelihood
 
 # Initialise parameters and corresponding transformations
-params, trainable, constrainer, unconstrainer = gpx.initialise(
-    circlular_posterior, key
-).unpack()
+parameter_state = gpx.initialise(circlular_posterior, key)
 
 # Optimise GP's marginal log-likelihood using Adam
-mll = jit(circlular_posterior.marginal_log_likelihood(D, constrainer, negative=True))
+mll = jit(circlular_posterior.marginal_log_likelihood(D, negative=True))
+
 learned_params, training_history = gpx.fit(
     mll,
-    params,
-    trainable,
-    adam(learning_rate=0.05),
+    parameter_state,
+    adam(learning_rate=0.01),
     n_iters=1000,
 ).unpack()
 
-# Untransform learned parameters
-final_params = gpx.transform(learned_params, constrainer)
-
 # %% [markdown]
 # ### Prediction
 #
 # We'll now query the GP's predictive posterior at linearly spaced novel inputs and illustrate the results.
 
 # %%
-posterior_rv = likelihood(circlular_posterior(D, final_params)(angles), final_params)
+posterior_rv = likelihood(circlular_posterior(D, learned_params)(angles), learned_params)
 mu = posterior_rv.mean
 one_sigma = jnp.sqrt(posterior_rv.variance)
 

From e84489ae1b09fbf119c7682db371a6d46e569842 Mon Sep 17 00:00:00 2001
From: Thomas Pinder <tompinder@live.co.uk>
Date: Sat, 15 Oct 2022 21:40:31 +0000
Subject: [PATCH 13/14] Fix missing dep

---
 docs/index.md     | 1 -
 examples/utils.py | 2 --
 2 files changed, 3 deletions(-)

diff --git a/docs/index.md b/docs/index.md
index a73af5aa..4c4fa24f 100644
--- a/docs/index.md
+++ b/docs/index.md
@@ -42,7 +42,6 @@ To learn more, checkout the [regression
 notebook](https://gpjax.readthedocs.io/en/latest/examples/regression.html).
 :::
 
-Are you rendering
 
 ---
 
diff --git a/examples/utils.py b/examples/utils.py
index aa287445..b8bdecd1 100644
--- a/examples/utils.py
+++ b/examples/utils.py
@@ -1,7 +1,5 @@
 import matplotlib.transforms as transforms
 import numpy as np
-import tensorflow as tf
-import utils as ut
 from matplotlib.patches import Ellipse
 
 

From 40ffc83b5976c90265eb4c7511673e648cca69d7 Mon Sep 17 00:00:00 2001
From: Thomas Pinder <tompinder@live.co.uk>
Date: Sun, 16 Oct 2022 14:27:43 +0000
Subject: [PATCH 14/14] Resolve review comments

---
 examples/classification.pct.py | 13 ++++++-------
 examples/collapsed_vi.pct.py   | 11 +++++------
 gpjax/gps.py                   |  6 +++---
 gpjax/likelihoods.py           |  4 ++--
 4 files changed, 16 insertions(+), 18 deletions(-)

diff --git a/examples/classification.pct.py b/examples/classification.pct.py
index 9e22f3e7..fc401e21 100644
--- a/examples/classification.pct.py
+++ b/examples/classification.pct.py
@@ -25,7 +25,7 @@
 import jax.random as jr
 import jax.scipy as jsp
 import matplotlib.pyplot as plt
-import numpyro
+import numpyro.distributions as npd
 import optax as ox
 from jaxtyping import Array, Float
 import blackjax
@@ -71,7 +71,6 @@
 # %%
 parameter_state = gpx.initialise(posterior)
 params, trainable, bijectors = parameter_state.unpack()
-params = gpx.unconstrain(params, bijectors)
 
 mll = jax.jit(posterior.marginal_log_likelihood(D, negative=True))
 
@@ -145,7 +144,7 @@
 L_inv = jsp.linalg.solve_triangular(L, I(D.n), lower=True)
 H_inv = jsp.linalg.solve_triangular(L.T, L_inv, lower=False)
 
-laplace_approximation = numpyro.distributions.MultivariateNormal(f_map_estimate, H_inv)
+laplace_approximation = npd.MultivariateNormal(f_map_estimate, H_inv)
 
 from gpjax.kernels import cross_covariance, gram
 
@@ -156,18 +155,18 @@
 
 
 def predict(
-    laplace_at_data: numpyro.distributions.Distribution,
+    laplace_at_data: npd.Distribution,
     train_data: Dataset,
     test_inputs: Float[Array, "N D"],
     jitter: int = 1e-6,
-) -> numpyro.distributions.Distribution:
+) -> npd.Distribution:
     """Compute the predictive distribution of the Laplace approximation at novel inputs.
 
     Args:
         laplace_at_data (dict): The Laplace approximation at the datapoints.
 
     Returns:
-        numpyro.distributions.Distribution: The Laplace approximation at novel inputs.
+        npd.Distribution: The Laplace approximation at novel inputs.
     """
     x, n = train_data.X, train_data.n
 
@@ -209,7 +208,7 @@ def predict(
     )
     covariance += I(n_test) * jitter
 
-    return numpyro.distributions.MultivariateNormal(
+    return npd.MultivariateNormal(
         jnp.atleast_1d(mean.squeeze()), covariance
     )
 
diff --git a/examples/collapsed_vi.pct.py b/examples/collapsed_vi.pct.py
index fe48d87b..e8b3ce15 100644
--- a/examples/collapsed_vi.pct.py
+++ b/examples/collapsed_vi.pct.py
@@ -87,14 +87,11 @@
 # We now train our model akin to a Gaussian process regression model via the `fit` abstraction. Unlike the regression example given in the [conjugate regression notebook](https://gpjax.readthedocs.io/en/latest/nbs/regression.html), the inducing locations that induce our variational posterior distribution are now part of the model's parameters. Using a gradient-based optimiser, we can then _optimise_ their location such that the evidence lower bound is maximised.
 # %%
 parameter_state = gpx.initialise(sgpr, key)
-params, trainables, bijectors = parameter_state.unpack()
 
 loss_fn = jit(sgpr.elbo(D, negative=True))
 
 optimiser = ox.adam(learning_rate=0.005)
 
-params = gpx.unconstrain(params, bijectors)
-
 learned_params, training_history = gpx.fit(
     objective=loss_fn,
     parameter_state=parameter_state,
@@ -102,6 +99,9 @@
     n_iters=2000,
 ).unpack()
 
+# %%
+parameter_state.bijectors
+
 # %% [markdown]
 # We show predictions of our model with the learned inducing points overlayed in grey.
 # %%
@@ -168,13 +168,12 @@
 fr_params, fr_trainables, fr_bijectors = gpx.initialise(
     full_rank_model, key
 ).unpack()
-fr_params = gpx.unconstrain(fr_params, fr_bijectors)
 mll = jit(full_rank_model.marginal_log_likelihood(D, negative=True))
-# %timeit mll(fr_params).block_until_ready()
+%timeit mll(fr_params).block_until_ready()
 
 # %%
 sparse_elbo = jit(sgpr.elbo(D, negative=True))
-# %timeit sparse_elbo(params).block_until_ready()
+%timeit sparse_elbo(params).block_until_ready()
 
 # %% [markdown]
 # As we can see, the sparse approximation given here is around 50 times faster when compared against a full-rank model.
diff --git a/gpjax/gps.py b/gpjax/gps.py
index 4d23a1c7..1e3642af 100644
--- a/gpjax/gps.py
+++ b/gpjax/gps.py
@@ -110,7 +110,7 @@ def __rmul__(self, other: AbstractLikelihood):
         """
         return self.__mul__(other)
 
-    def predict(self, params: dict) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
+    def predict(self, params: Dict) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Compute the GP's prior mean and variance.
 
         Args:
@@ -195,7 +195,7 @@ class ConjugatePosterior(AbstractPosterior):
     jitter: Optional[float] = DEFAULT_JITTER
 
     def predict(
-        self, train_data: Dataset, params: dict
+        self, train_data: Dataset, params: Dict
     ) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Conditional on a set of training data, compute the GP's posterior predictive distribution for a given set of parameters. The returned function can be evaluated at a set of test inputs to compute the corresponding predictive density.
 
@@ -310,7 +310,7 @@ def _initialise_params(self, key: PRNGKeyType) -> Dict:
         return parameters
 
     def predict(
-        self, train_data: Dataset, params: dict
+        self, train_data: Dataset, params: Dict
     ) -> Callable[[Float[Array, "N D"]], npd.Distribution]:
         """Conditional on a set of training data, compute the GP's posterior predictive distribution for a given set of parameters. The returned function can be evaluated at a set of test inputs to compute the corresponding predictive density. Note, to gain predictions on the scale of the original data, the returned distribution will need to be transformed through the likelihood function's inverse link function.
 
diff --git a/gpjax/likelihoods.py b/gpjax/likelihoods.py
index 0585ba96..e9bf3173 100644
--- a/gpjax/likelihoods.py
+++ b/gpjax/likelihoods.py
@@ -116,12 +116,12 @@ def link_function(self) -> Callable:
             Callable: A link function that maps the predictive distribution to the likelihood function.
         """
 
-        def link_fn(x, params: dict) -> npd.Distribution:
+        def link_fn(x, params: Dict) -> npd.Distribution:
             return npd.Normal(loc=x, scale=params["obs_noise"])
 
         return link_fn
 
-    def predict(self, dist: npd.Distribution, params: dict) -> npd.Distribution:
+    def predict(self, dist: npd.Distribution, params: Dict) -> npd.Distribution:
         """Evaluate the Gaussian likelihood function at a given predictive distribution. Computationally, this is equivalent to summing the observation noise term to the diagonal elements of the predictive distribution's covariance matrix.
 
         Args: