From 2d8f661924dde784f4b9905520524e77c1663cd4 Mon Sep 17 00:00:00 2001 From: Giorgio Gilestro Date: Mon, 7 Sep 2026 13:08:58 +0100 Subject: [PATCH] Paper-wide rename: the pop-gen construct is "the biological model"; "(exact)" dropped "Model" now means an AI model everywhere; the Wright-Fisher construct is "the biological model" throughout (19 occurrences): tier header, section title (now "The biological model, and where trained learners depart from it"), Table 1 support column ("Exact" -> "Closed form"; "Analytic model" -> "Biological model"), Results, Discussion, Methods ("Biological-model tier"), and all figure captions. "Exact" survives only in technical noun phrases (exact-match verifier, exact oracle, exact equilibrium, exact recovery); the abstract's "exactly Wright-Fisher" is now "literally Wright-Fisher". Co-Authored-By: Claude Fable 5 Claude-Session: https://claude.ai/code/session_01BkRLcc18rwT2Lysu6PbG7v --- paper/pnas/body.tex | 48 ++++++++++++++++----------------- paper/pnas/build.py | 14 +++++----- paper/pnas/figs/fig1a.pdf | Bin 61362 -> 61397 bytes paper/pnas/main.md | 54 ++++++++++++++++++++------------------ paper/pnas/main.pdf | Bin 394146 -> 394440 bytes paper/pnas/make_figs.py | 2 +- tasks/lessons.md | 12 +++++---- 7 files changed, 67 insertions(+), 63 deletions(-) diff --git a/paper/pnas/body.tex b/paper/pnas/body.tex index 0979667..68fdbb8 100644 --- a/paper/pnas/body.tex +++ b/paper/pnas/body.tex @@ -4,7 +4,7 @@ Artificial intelligence increasingly consists of populations of models rather th \section*{Abstract} -AI development increasingly resembles a population process. Models are specialised, retrained on model output, and recombined by weight merging, and the practice is described in evolutionary vocabulary with little use of evolutionary theory. I treat multigenerational model populations as systems whose inheritance, diversity, and compatibility must be managed, and transfer the quantitative framework of the evolution of sex. Its starting point, that training on model output is genetic drift and model collapse its signature, is by now established from several independent directions; I develop the structure that follows from it. In a minimal inheritance model that is exactly Wright--Fisher, and measurably Wright--Fisher plus estimator bias in trained networks, I derive and test remedies. Grounding acts as immigration: a real-data fraction far below one retained most equilibrium diversity, with a per-capability observation floor that makes the rarest knowledge expensive under unstratified sampling. Refitting a child to the mean of its parents' output distributions cancels the multi-parent gain to first order in the rare-item regime; union-preserving operators realise it. Merged language-model specialists exceeded every parent in replicated experiments. Blind recombination fails on rugged task landscapes; screening candidate offspring restores the gain. The optimal mating breadth narrows as skills entangle. Finally, I introduce model speciation: a merge barrier remaining after permutation-and-rescaling alignment tracks functional conflict, isolation did not emerge from compatible specialisation, and in a controlled test pre-merge functional disagreement predicted merge damage while weight-geometry baselines showed no detectable association. +AI development increasingly resembles a population process. Models are specialised, retrained on model output, and recombined by weight merging, and the practice is described in evolutionary vocabulary with little use of evolutionary theory. I treat multigenerational model populations as systems whose inheritance, diversity, and compatibility must be managed, and transfer the quantitative framework of the evolution of sex. Its starting point, that training on model output is genetic drift and model collapse its signature, is by now established from several independent directions; I develop the structure that follows from it. In a minimal biological model of inheritance that is literally Wright--Fisher, and measurably Wright--Fisher plus estimator bias in trained networks, I derive and test remedies. Grounding acts as immigration: a real-data fraction far below one retained most equilibrium diversity, with a per-capability observation floor that makes the rarest knowledge expensive under unstratified sampling. Refitting a child to the mean of its parents' output distributions cancels the multi-parent gain to first order in the rare-item regime; union-preserving operators realise it. Merged language-model specialists exceeded every parent in replicated experiments. Blind recombination fails on rugged task landscapes; screening candidate offspring restores the gain. The optimal mating breadth narrows as skills entangle. Finally, I introduce model speciation: a merge barrier remaining after permutation-and-rescaling alignment tracks functional conflict, isolation did not emerge from compatible specialisation, and in a controlled test pre-merge functional disagreement predicted merge damage while weight-geometry baselines showed no detectable association. \medskip\hrule\medskip @@ -16,53 +16,53 @@ The generations are coupled through data as well as through weights. Successive Training each generation of a model on the previous generation's output degrades it (\emph{model collapse}): rare capabilities vanish first, and the lineage drifts toward its own most common behaviour (21). That degradation is, mathematically, \emph{genetic drift}, the loss of rare variants that any finite population suffers when each generation is a finite sample of the last --- the same sampling accident by which rare surnames vanish from small villages and rare alleles (gene variants) drift out of island populations with no selection against them. The identification has been made repeatedly and independently: for sequential inference chains before deep learning (22), for language-model text ecosystems (23), as a closed-form first-extinction law placing collapse onset at the Wright--Fisher first-extinction time (24), and in quantitative-genetic form for self-consuming diffusion models (25). A diagnosis reached so often, from such different starting points, marks population genetics as the natural mathematics of the setting, though only as its entry point: population genetics is not, at heart, a theory of decay; it is a theory of the mechanisms that maintain and build populations despite decay (immigration, recombination, selection, population structure) and of where those mechanisms reach their limits. This paper develops that fuller structure for model populations: the arc from drift through its remedies to its limit, reproductive isolation --- the point at which diverged lineages can no longer produce working offspring, biology's boundary between species --- carried as one framework from closed forms to trained networks to language models. -An operator of a model population faces recurring decisions for which there is no principled guidance: how much verified real data does retraining need before a lineage decays; will combining two particular models compose their abilities or damage them; can incompatibility be detected before paying for a failed merge; and when should specialists be kept separate rather than consolidated? In practice these are settled by convention and by trial-and-error search. They are also, recognisably, machine learning's oldest problem at a new scale: \emph{continual learning}, the struggle to acquire new abilities without losing old ones (26, 27), transposed from a single network to a population whose members inherit from one another. Population genetics, I will argue, prices these decisions. Table 1 summarises the correspondences on which the argument runs. Fig. 1A maps the experimental programme built on them: the same abstractions tested at three tiers --- an exact population-genetic simulator, trained neural networks, and language models --- with the sections that follow climbing that ladder. Fig. 1B draws the shift of perspective the whole transfer rests on: the model ecosystem read not as a society in space, contemporaries exchanging messages, but as a society in time, generations coupled by inheritance, recombination, and immigration. +An operator of a model population faces recurring decisions for which there is no principled guidance: how much verified real data does retraining need before a lineage decays; will combining two particular models compose their abilities or damage them; can incompatibility be detected before paying for a failed merge; and when should specialists be kept separate rather than consolidated? In practice these are settled by convention and by trial-and-error search. They are also, recognisably, machine learning's oldest problem at a new scale: \emph{continual learning}, the struggle to acquire new abilities without losing old ones (26, 27), transposed from a single network to a population whose members inherit from one another. Population genetics, I will argue, prices these decisions. Table 1 summarises the correspondences on which the argument runs. Fig. 1A maps the experimental programme built on them: the same abstractions tested at three tiers --- a biological model in simulation, trained neural networks, and language models --- with the sections that follow climbing that ladder. Fig. 1B draws the shift of perspective the whole transfer rests on: the model ecosystem read not as a society in space, contemporaries exchanging messages, but as a society in time, generations coupled by inheritance, recombination, and immigration. \begin{figure*}[p]\centering % fig1 \includegraphics[width=\textwidth]{figs/fig1a.pdf}\\[6pt] \includegraphics[width=\textwidth]{figs/fig1b.pdf} -\caption{(A) The experimental programme. Each population-genetic abstraction (Table 1) is tested at up to three tiers, ordered left to right by increasing realism: an exact Wright--Fisher simulator over knowledge distributions (closed forms; bitwise-reproducible), trained neural networks measured against exact oracles (recurrent, feedforward, and variational-autoencoder generators on a synthetic mode universe, and a convolutional VAE on MNIST), and language models (LoRA specialists on Qwen bases at 0.5B and 7B, scored by an exact-match verifier). Colour separates the two categories: the population-genetic theory tier in blue, the two AI-model tiers in oranges. The same abstractions are carried across all three. Rows are the framework's mechanisms, each defined at the left margin; filled cells name the experiments run at each tier, and each carries, in its corner, the figure or table where that result is reported, so this figure doubles as a map of the paper. Each claim is tested at the cheapest tier that can falsify it, and a costlier tier is entered only where it adds a discriminating test rather than a replication: grounding at language-model scale is established in prior work (21, 30) and is not re-run; epistasis and the society skip the middle tier, whose distinctive value (exact oracles) does not bear on those operator-level questions; and the society at language-model scale is the integrative experiment this paper specifies but does not run --- its stated gap. (B) The conceptual basis of the transfer. A population of models is usually pictured as a society in space: contemporaries exchanging messages. The couplings this paper studies run instead between generations --- training on model output (inheritance), weight-space merging (recombination), verified real data entering each generation (immigration from reality) --- a society in time, the object population genetics was built to describe. Dots mark capabilities: the rare one (gold) is lost under single-parent inheritance, reassembled by merging complementary parents, and re-supplied by grounding.}\label{fig1} +\caption{(A) The experimental programme. Each population-genetic abstraction (Table 1) is tested at up to three tiers, ordered left to right by increasing realism: a biological model (a Wright--Fisher simulator over knowledge distributions; closed forms, bitwise-reproducible), trained neural networks measured against exact oracles (recurrent, feedforward, and variational-autoencoder generators on a synthetic mode universe, and a convolutional VAE on MNIST), and language models (LoRA specialists on Qwen bases at 0.5B and 7B, scored by an exact-match verifier). Colour separates the two categories: the biological model in blue, the two AI-model tiers in oranges. The same abstractions are carried across all three. Rows are the framework's mechanisms, each defined at the left margin; filled cells name the experiments run at each tier, and each carries, in its corner, the figure or table where that result is reported, so this figure doubles as a map of the paper. Each claim is tested at the cheapest tier that can falsify it, and a costlier tier is entered only where it adds a discriminating test rather than a replication: grounding at language-model scale is established in prior work (21, 30) and is not re-run; epistasis and the society skip the middle tier, whose distinctive value (exact oracles) does not bear on those operator-level questions; and the society at language-model scale is the integrative experiment this paper specifies but does not run --- its stated gap. (B) The conceptual basis of the transfer. A population of models is usually pictured as a society in space: contemporaries exchanging messages. The couplings this paper studies run instead between generations --- training on model output (inheritance), weight-space merging (recombination), verified real data entering each generation (immigration from reality) --- a society in time, the object population genetics was built to describe. Dots mark capabilities: the rare one (gold) is lost under single-parent inheritance, reassembled by merging complementary parents, and re-supplied by grounding.}\label{fig1} \end{figure*} -\section*{The minimal model, and where its exactness ends} +\section*{The biological model, and where trained learners depart from it} -Knowledge is modelled as a distribution \texttt{p\_t} over \texttt{K} discrete items (capabilities, facts, modes of behaviour), with a fixed true distribution \texttt{p*} whose rare tail carries the knowledge most at risk. One generation is: *draw \texttt{n} samples from the parent's distribution, optionally mix in \texttt{m} verified real samples (``grounding'', \texttt{g = m/(n+m)}), and refit the child\emph{. In this minimal inheritance model the resampling step }is* the Wright--Fisher process, population genetics' canonical model of neutral evolution, in which each new generation is a random sample of size \texttt{n} from the previous one and every statistical property of drift follows from that one step. Diversity throughout this paper is \emph{heterozygosity}, \texttt{H = 1 − Σ p\_i²}: the probability that two random draws differ (one minus a collision probability), high when many items share the mass, zero at total collapse. The identity is exploited as an engineering gate: the simulator reproduces the classical closed forms (heterozygosity decay \texttt{E[H\_t] = H\_0(1 − 1/n)\textasciicircum{}t}; the exact immigration--drift equilibrium; the closed-form multi-teacher union) to within 0.5\%, and these are standing tests in the codebase, not one-off checks. +Knowledge is modelled as a distribution \texttt{p\_t} over \texttt{K} discrete items (capabilities, facts, modes of behaviour), with a fixed true distribution \texttt{p*} whose rare tail carries the knowledge most at risk. One generation is: *draw \texttt{n} samples from the parent's distribution, optionally mix in \texttt{m} verified real samples (``grounding'', \texttt{g = m/(n+m)}), and refit the child\emph{. This construction is the paper's }biological model\emph{, deliberately minimal, and its resampling step }is* the Wright--Fisher process, population genetics' canonical model of neutral evolution, in which each new generation is a random sample of size \texttt{n} from the previous one and every statistical property of drift follows from that one step. Diversity throughout this paper is \emph{heterozygosity}, \texttt{H = 1 − Σ p\_i²}: the probability that two random draws differ (one minus a collision probability), high when many items share the mass, zero at total collapse. The identity is exploited as an engineering gate: the simulator reproduces the classical closed forms (heterozygosity decay \texttt{E[H\_t] = H\_0(1 − 1/n)\textasciicircum{}t}; the exact immigration--drift equilibrium; the closed-form multi-teacher union) to within 0.5\%, and these are standing tests in the codebase, not one-off checks. -The boundary of the exactness matters, and I measured it rather than assumed it. Real training adds approximation, optimisation noise, and inductive bias, and when trained networks are fit against the exact drift null they deviate in \emph{opposite, architecture-specific} directions: a smoothing recurrent network resists collapse (keeping spurious variants alive), while a sharpening image generator accelerates it. A one-parameter \emph{learning kernel} (a smoothing knob and a sharpening knob on the refit) reproduces both. Throughout, a real learner is therefore treated as Wright--Fisher \emph{plus a signed, measurable estimator bias}, and the drift signs (rare-first loss; the grounding response) survived that bias in every architecture I tested, including a convolutional VAE retrained on its own generated digits, where the dry lineage collapses to a single blurred digit class while 10\% grounding holds all thirty modes (Fig. 2). Retraining on a single parent is \emph{asexual reproduction}, and sustained loss under it carries the defining consequence of \emph{Muller's ratchet} (28), the mechanism by which lineages that never recombine decay irreversibly --- the reason non-recombining genomes such as the Y chromosome have shed most of their ancestral genes. Once every copy of a rare capability is gone from all parents and sources, no recombination can rebuild it: each such loss is a click of the ratchet, and remedies must act while copies still survive somewhere (a consequence-level correspondence; the minimal model lacks the ratchet's recurrent-mutation driver). +Where a real learner departs from this model matters, and I measured that boundary rather than assumed it. Real training adds approximation, optimisation noise, and inductive bias, and when trained networks are fit against the exact drift null they deviate in \emph{opposite, architecture-specific} directions: a smoothing recurrent network resists collapse (keeping spurious variants alive), while a sharpening image generator accelerates it. A one-parameter \emph{learning kernel} (a smoothing knob and a sharpening knob on the refit) reproduces both. Throughout, a real learner is therefore treated as Wright--Fisher \emph{plus a signed, measurable estimator bias}, and the drift signs (rare-first loss; the grounding response) survived that bias in every architecture I tested, including a convolutional VAE retrained on its own generated digits, where the dry lineage collapses to a single blurred digit class while 10\% grounding holds all thirty modes (Fig. 2). Retraining on a single parent is \emph{asexual reproduction}, and sustained loss under it carries the defining consequence of \emph{Muller's ratchet} (28), the mechanism by which lineages that never recombine decay irreversibly --- the reason non-recombining genomes such as the Y chromosome have shed most of their ancestral genes. Once every copy of a rare capability is gone from all parents and sources, no recombination can rebuild it: each such loss is a click of the ratchet, and remedies must act while copies still survive somewhere (a consequence-level correspondence; the biological model lacks the ratchet's recurrent-mutation driver). -\textbf{Table 1.} The dictionary. Each biological term is introduced in the section that develops it; each correspondence is stated with the level of support it currently has (exact = closed form in the minimal model; empirical = measured in trained systems; hypothesis = stated with a falsifier, untested or unconfirmed). The full claim-by-claim ledger with assumptions and known limits is SI Appendix, Table S1. +\textbf{Table 1.} The dictionary. Each biological term is introduced in the section that develops it; each correspondence is stated with the level of support it currently has (closed form = derived in the biological model and verified against simulation; empirical = measured in trained systems; hypothesis = stated with a falsifier, untested or unconfirmed). The full claim-by-claim ledger with assumptions and known limits is SI Appendix, Table S1. \medskip\noindent\begin{center}\footnotesize \begin{tabular}{p{0.307\textwidth} p{0.307\textwidth} p{0.307\textwidth}} \hline Population genetics & Model populations & Support \\ \hline -Genetic drift in a finite population & Training on finite samples of model output & Exact (minimal model); signs in trained nets; diagnosis conceded to prior work \\[3pt] -Immigration from a fixed source & Grounding with verified real data & Exact equilibrium; signs in RNN/MLP/VAE/MNIST \\[3pt] +Genetic drift in a finite population & Training on finite samples of model output & Closed form (biological model); signs in trained nets; diagnosis conceded to prior work \\[3pt] +Immigration from a fixed source & Grounding with verified real data & Closed-form equilibrium; signs in RNN/MLP/VAE/MNIST \\[3pt] Muller's ratchet (asexual decay) & Irreversible arm of model collapse & Correspondence, scoped: applies to unrecoverable loss \\[3pt] Recombination / sexual reproduction & Model merging & Empirical at 0.5B--7B \\[3pt] -Fisher--Muller effect & Merged specialists exceed every parent & Analytic model; replicated in LLMs \\[3pt] -Outbreeding depression under epistasis & Merging entangled skills harms offspring & Analytic model (NK landscapes); hypothesis at LLM scale \\[3pt] -Mating systems / population structure & Who merges with whom (breadth of the parent pool) & Analytic model; hypothesis for real populations \\[3pt] +Fisher--Muller effect & Merged specialists exceed every parent & Biological model; replicated in LLMs \\[3pt] +Outbreeding depression under epistasis & Merging entangled skills harms offspring & Biological model (NK landscapes); hypothesis at LLM scale \\[3pt] +Mating systems / population structure & Who merges with whom (breadth of the parent pool) & Biological model; hypothesis for real populations \\[3pt] Reproductive isolation (BDM incompatibilities) & Merge failure from functional conflict & Empirical (MLP + LLM tiers, conflict-associated); emergent form not observed \\[3pt] -Selection on a fitness function & Verifier-anchored selection (``reality that can say no'') & Analytic model (complementary with recombination and diversity in the tested society) \\[3pt] +Selection on a fitness function & Verifier-anchored selection (``reality that can say no'') & Biological model (complementary with recombination and diversity in the tested society) \\[3pt] \hline\end{tabular}\end{center}\medskip \section*{Results} \subsection*{Grounding is immigration: cheap, with a floor} -In the minimal model, grounding from a fixed real source is \emph{immigration} into a drifting population (29--31). Immigration is what conservation managers prescribe when a fragmented reserve loses diversity, and its striking property there is how little is needed --- the field's rule of thumb is that one migrant per generation holds an isolated population's diversity (32). The same economy appears here: the equilibrium diversity has a closed form the simulator matches exactly. That equilibrium is \emph{smooth} in the grounding fraction (there is no phase transition in aggregate diversity), so the practical number is an operational threshold, and I define it as such: under the tested population size and Zipf source distribution, \texttt{g \(\approx\) 0.05} retained most (\(\geq\)95\%) of equilibrium diversity indefinitely, with the required fraction depending on sample size, source distribution, and the chosen retention target (dependencies in SI). Verified real data remains, on any of these definitions, cheap insurance at fractions far below one. But the same analysis yields a floor the field's average-loss framing misses: under unstratified sampling from the source, a capability of rarity \texttt{p} appears in a real-data batch of size \texttt{m} with probability \texttt{1 − e\textasciicircum{}{−m\(\cdot\)p}}, so \texttt{m\(\cdot\)p \(\approx\) 1} marks roughly a 63\% chance of one example per batch: a soft observation floor, with higher confidence priced accordingly, and with distinct consequences for continuous retention, stationary occupancy, and reintroduction after loss (immigration can restore an absent item; SI separates these). Protecting the rarest knowledge under unstratified grounding is therefore priced per item at cost \texttt{\(\propto\) 1/p}; targeted or stratified sampling changes that cost, and recombination can recover rare capabilities \emph{that are still retained across complementary parents} (next section). In trained networks the \emph{sign} of the grounding response transfers everywhere I looked, with two deviations, both traced to the estimator bias above: sharp thresholds soften, and support-counting metrics decouple from truth (forward-KL is the operative collapse metric for a smoothing learner). On real images (Fig. 2B), dry self-training collapses a convolutional VAE to one mode while \textasciitilde{}10\% grounding holds all thirty (the trained model needs roughly twice the exact-operator fraction, the measured price of the estimator bias). +In the biological model, grounding from a fixed real source is \emph{immigration} into a drifting population (29--31). Immigration is what conservation managers prescribe when a fragmented reserve loses diversity, and its striking property there is how little is needed --- the field's rule of thumb is that one migrant per generation holds an isolated population's diversity (32). The same economy appears here: the equilibrium diversity has a closed form the simulator matches exactly. That equilibrium is \emph{smooth} in the grounding fraction (there is no phase transition in aggregate diversity), so the practical number is an operational threshold, and I define it as such: under the tested population size and Zipf source distribution, \texttt{g \(\approx\) 0.05} retained most (\(\geq\)95\%) of equilibrium diversity indefinitely, with the required fraction depending on sample size, source distribution, and the chosen retention target (dependencies in SI). Verified real data remains, on any of these definitions, cheap insurance at fractions far below one. But the same analysis yields a floor the field's average-loss framing misses: under unstratified sampling from the source, a capability of rarity \texttt{p} appears in a real-data batch of size \texttt{m} with probability \texttt{1 − e\textasciicircum{}{−m\(\cdot\)p}}, so \texttt{m\(\cdot\)p \(\approx\) 1} marks roughly a 63\% chance of one example per batch: a soft observation floor, with higher confidence priced accordingly, and with distinct consequences for continuous retention, stationary occupancy, and reintroduction after loss (immigration can restore an absent item; SI separates these). Protecting the rarest knowledge under unstratified grounding is therefore priced per item at cost \texttt{\(\propto\) 1/p}; targeted or stratified sampling changes that cost, and recombination can recover rare capabilities \emph{that are still retained across complementary parents} (next section). In trained networks the \emph{sign} of the grounding response transfers everywhere I looked, with two deviations, both traced to the estimator bias above: sharp thresholds soften, and support-counting metrics decouple from truth (forward-KL is the operative collapse metric for a smoothing learner). On real images (Fig. 2B), dry self-training collapses a convolutional VAE to one mode while \textasciitilde{}10\% grounding holds all thirty (the trained model needs roughly twice the biological model's fraction, the measured price of the estimator bias). \begin{figure*}[p]\centering % fig2 \includegraphics[width=\textwidth]{figs/fig2.pdf} -\caption{Grounding is immigration. (A) Stationary diversity against the grounding fraction in the minimal inheritance model: simulation (points, 95\% CI) matches the exact immigration--drift equilibrium (dashed). The equilibrium is smooth in $g$; $g \approx 0.05$ marks the operational threshold retaining 95\% of source diversity in this setting (red line, bootstrap CI shaded); the hollow point at $g=0$ is a finite-time value (the true equilibrium is zero). (B) The same signs on real images: samples from a convolutional VAE retrained each generation on its own output (rows: generations 0--15 of an ungrounded lineage) collapse toward a single blurred mode; 10\% grounding holds all thirty modes (quantified in SI).}\label{fig2} +\caption{Grounding is immigration. (A) Stationary diversity against the grounding fraction in the biological model: simulation (points, 95\% CI) matches the exact immigration--drift equilibrium (dashed). The equilibrium is smooth in $g$; $g \approx 0.05$ marks the operational threshold retaining 95\% of source diversity in this setting (red line, bootstrap CI shaded); the hollow point at $g=0$ is a finite-time value (the true equilibrium is zero). (B) The same signs on real images: samples from a convolutional VAE retrained each generation on its own output (rows: generations 0--15 of an ungrounded lineage) collapse toward a single blurred mode; 10\% grounding holds all thirty modes (quantified in SI).}\label{fig2} \end{figure*} \subsection*{Recombination: a conservation law, its operators, and offspring that exceed every parent} -The largest returns from the transfer concern merging. \emph{Blending inheritance} --- offspring as the average of their parents --- is the failure mode at the root of population genetics' founding controversy: the swamping argument pressed in Jenkin's 1867 review of \emph{The Origin of Species}, that under blending a rare advantageous variant is diluted toward the common type faster than selection can multiply it (33), an objection dissolved only by Mendel's particulate inheritance, in which discrete variants pass through generations undiluted. Refitting a child model to the mean of its parents' output distributions is blending inheritance, and the proposition below is Jenkin's dilution made exact. \textbf{Proposition (blending inheritance, rare-item regime).} Let K parents independently retain a rare item (mass \texttt{p} when retained), and let the child draw \texttt{n} samples either from one parent chosen at random or from the \emph{mean of the parents' output distributions}. Expected item mass is identical under the two schemes; and in the rare-item regime \texttt{n\(\cdot\)p/K \(\ll\) 1}, where per-item survival is first-order in sampled mass, expected \emph{survival} is also identical: the 1/K dilution of averaging cancels the K-parent union gain to first order, so in this regime adding parents through the output-mean does not increase expected tail retention. Two boundaries: outside that regime, survival is a convex function of mixed mass, so the variance reduction from averaging can \emph{reduce} extinction relative to a randomly chosen single parent; the cancellation is a first-order result about rare items, not a universal impossibility; and the contrasting union operator (keep each item's strongest source, then renormalise, which itself redistributes mass and presupposes a verifier or oracle to identify the strongest source) increases expected retention with K in all regimes in the minimal model. The practically important operators, \emph{weight averaging} (a nonlinear network's weight-mean does not compute its parents' output-mean) and \emph{routing among intact specialists} (34) (different storage and inference budgets from a single child), are its empirical cousins, and the measured bridge is a \emph{headroom rule}, stated qualitatively: in language models, union-preserving operators beat the weight-average where that average falls short of attainable performance, and add nothing where it does not (easy-versus-hard contrasts at two scales; a quantitative form of the relationship is untested). On easy tasks a capable base's average is already at ceiling and refinements add nothing; on hard tasks the average dilutes a fragile specialist below even the best single parent and routing wins by a wide margin (Fig. 3A--B). +The largest returns from the transfer concern merging. \emph{Blending inheritance} --- offspring as the average of their parents --- is the failure mode at the root of population genetics' founding controversy: the swamping argument pressed in Jenkin's 1867 review of \emph{The Origin of Species}, that under blending a rare advantageous variant is diluted toward the common type faster than selection can multiply it (33), an objection dissolved only by Mendel's particulate inheritance, in which discrete variants pass through generations undiluted. Refitting a child model to the mean of its parents' output distributions is blending inheritance, and the proposition below is Jenkin's dilution made exact. \textbf{Proposition (blending inheritance, rare-item regime).} Let K parents independently retain a rare item (mass \texttt{p} when retained), and let the child draw \texttt{n} samples either from one parent chosen at random or from the \emph{mean of the parents' output distributions}. Expected item mass is identical under the two schemes; and in the rare-item regime \texttt{n\(\cdot\)p/K \(\ll\) 1}, where per-item survival is first-order in sampled mass, expected \emph{survival} is also identical: the 1/K dilution of averaging cancels the K-parent union gain to first order, so in this regime adding parents through the output-mean does not increase expected tail retention. Two boundaries: outside that regime, survival is a convex function of mixed mass, so the variance reduction from averaging can \emph{reduce} extinction relative to a randomly chosen single parent; the cancellation is a first-order result about rare items, not a universal impossibility; and the contrasting union operator (keep each item's strongest source, then renormalise, which itself redistributes mass and presupposes a verifier or oracle to identify the strongest source) increases expected retention with K in all regimes in the biological model. The practically important operators, \emph{weight averaging} (a nonlinear network's weight-mean does not compute its parents' output-mean) and \emph{routing among intact specialists} (34) (different storage and inference budgets from a single child), are its empirical cousins, and the measured bridge is a \emph{headroom rule}, stated qualitatively: in language models, union-preserving operators beat the weight-average where that average falls short of attainable performance, and add nothing where it does not (easy-versus-hard contrasts at two scales; a quantitative form of the relationship is untested). On easy tasks a capable base's average is already at ceiling and refinements add nothing; on hard tasks the average dilutes a fragile specialist below even the best single parent and routing wins by a wide margin (Fig. 3A--B). -The generative payoff is the \emph{Fisher--Muller effect} (35, 36), the classical account of why sex speeds adaptation: in an asexual population, beneficial variants arising in different individuals can only compete until all but one lineage is lost, whereas recombination assembles them in one offspring, producing a \emph{genotype} (an individual's combination of variants, one at each \emph{locus}, or position) fitter than any parent. In the multi-locus model, sexual merging of decorrelated specialists climbs to the global optimum, a genotype no parent held, while the best single parent and the blended average both plateau below (Fig. 4). In real language models the signature replicates under seed replication: merges of three LoRA (37) specialists beat every parent overall (decisively at 7B: 0.87 vs 0.77), and on the sharper worst-family metric the merged models are the only ones competent everywhere, in every seed (Fig. 3A). +The generative payoff is the \emph{Fisher--Muller effect} (35, 36), the classical account of why sex speeds adaptation: in an asexual population, beneficial variants arising in different individuals can only compete until all but one lineage is lost, whereas recombination assembles them in one offspring, producing a \emph{genotype} (an individual's combination of variants, one at each \emph{locus}, or position) fitter than any parent. In the multi-locus biological model, sexual merging of decorrelated specialists climbs to the global optimum, a genotype no parent held, while the best single parent and the blended average both plateau below (Fig. 4). In real language models the signature replicates under seed replication: merges of three LoRA (37) specialists beat every parent overall (decisively at 7B: 0.87 vs 0.77), and on the sharper worst-family metric the merged models are the only ones competent everywhere, in every seed (Fig. 3A). Sex has risks and, for AI, an unfair advantage, both quantified on Kauffman's NK fitness landscapes (38), the standard model of \emph{epistasis}, biology's term for interaction between genes: the fitness contribution of a variant depends on which variants occupy the other loci, much as a component's value in an ML system depends on the components around it. Each of the landscape's \texttt{N} sites interacts with \texttt{K} others (the model's eponymous parameters), and raising that interaction count tunes the landscape from smooth and additive to rugged and many-peaked (Fig. 5). When skills are entangled, blind recombination produces offspring \emph{below} their parents, worsening with ruggedness, and the optimal recombination rate shrinks as entanglement grows. Biology knows this failure as \emph{outbreeding depression}, the reason conservation practice warns against crossing locally adapted populations: in the textbook case, an ibex herd in the Tatra Mountains restocked with animals from Turkey and Sinai produced fertile hybrids that bore their young in the coldest month of winter, and the herd died out (39). But an engineered population can do what biology cannot: recombine unbounded parents, choose complementary mates, and \emph{screen many candidate offspring against a verifier before keeping one}. This directed sex converts the outbreeding catastrophe into a reliable gain in the model (tracking or exceeding the best parent at every ruggedness) and replicates as a sign in language models: bred-and-screened merges beat the a-priori blend in every seed on headroom tasks, including one seed where the blend failed catastrophically and selection was immune (SI Appendix, Table S2). Finally, population \emph{structure} is itself a knob: sweeping the mate-pool breadth from monogamous (repeated local pairings) to promiscuous (\emph{panmixia}: any model may merge with any other) against ruggedness, wide mixing maximises the population mean while monotonically destroying diversity, and the best \emph{champion} shifts from wide breadth on smooth landscapes to intermediate breadth on rugged ones (Fig. 5C), the mating-system phenomenon known to structured-population search, mapped onto merging populations. @@ -73,7 +73,7 @@ Sex has risks and, for AI, an unfair advantage, both quantified on Kauffman's NK \begin{figure*}[p]\centering % fig4 \includegraphics[width=\textwidth]{figs/fig4.pdf} -\caption{Recombination in the minimal model: blending inheritance and the Fisher--Muller effect. (A) Expected rare-capability survival in a child refit from $K$ uncorrelated parents: the output-mean (blending) stays at the single-parent level --- the first-order cancellation --- while the union operator (strongest source per item, renormalised, oracle-identified) rises with parent count. (B) Multi-locus recombination of decorrelated specialists produces offspring fitter than any parent, approaching the optimum as parents are added; the best single parent and the blended average plateau below (mean $\pm$ 95\% CI).}\label{fig4} +\caption{Recombination in the biological model: blending inheritance and the Fisher--Muller effect. (A) Expected rare-capability survival in a child refit from $K$ uncorrelated parents: the output-mean (blending) stays at the single-parent level --- the first-order cancellation --- while the union operator (strongest source per item, renormalised, oracle-identified) rises with parent count. (B) Multi-locus recombination of decorrelated specialists produces offspring fitter than any parent, approaching the optimum as parents are added; the best single parent and the blended average plateau below (mean $\pm$ 95\% CI).}\label{fig4} \end{figure*} \begin{figure*}[p]\centering % fig5 @@ -92,7 +92,7 @@ Composing the operators (Fig. 6) requires one definitional distinction first. In \subsection*{The limit of sex: model speciation} -Recombination presupposes compatible parents. In biology, lineages pushed far enough apart become separate species (\emph{reproductive isolation}) through Bateson--Dobzhansky--Muller incompatibilities (41, 42): changes harmless on their own genetic background but deleterious in combination --- the mechanism behind the mule's sterility and the inviability of many between-species crosses, in which two genomes that each work perfectly cannot run in the same cell. A merged model is exactly the exposed hybrid. I built the analytic model (Fig. 7A): hybrid fitness tracks the parents while compatible, then peels off and crashes below the ancestor; the isolation cliff arrives earlier the denser the incompatibilities; and the incompatibility \emph{count} snowballs quadratically with divergence (42). Note that a super-linear count does not by itself entail a sharp performance cliff without the count-to-effect-size link, which the analytic model supplies under its assumptions and any neural test must establish separately. +Recombination presupposes compatible parents. In biology, lineages pushed far enough apart become separate species (\emph{reproductive isolation}) through Bateson--Dobzhansky--Muller incompatibilities (41, 42): changes harmless on their own genetic background but deleterious in combination --- the mechanism behind the mule's sterility and the inviability of many between-species crosses, in which two genomes that each work perfectly cannot run in the same cell. A merged model is exactly the exposed hybrid. I built the biological model of the process (Fig. 7A): hybrid fitness tracks the parents while compatible, then peels off and crashes below the ancestor; the isolation cliff arrives earlier the denser the incompatibilities; and the incompatibility \emph{count} snowballs quadratically with divergence (42). Note that a super-linear count does not by itself entail a sharp performance cliff without the count-to-effect-size link, which the biological model supplies under its assumptions and any neural test must establish separately. In trained networks, the claim must survive a known alternative: merge barriers between independently trained networks are famously \emph{coordinate artefacts}, removable by re-aligning hidden units (43); richer symmetry groups remove more (44), with known failures beyond the shared-data regime (45). I therefore aligned under the composition of permutation matching and exact per-unit rescaling (the unit symmetry group of plain ReLU MLPs, as the search space) and decomposed the barrier (Fig. 7 C and D): two networks trained from different initialisations on the \emph{same} task have a barrier that this alignment removes essentially entirely (residual \(\approx\) 0.001, the aligned merge performing at parent level): coordinate, not functional; two networks trained on \emph{conflicting} label maps have a barrier the same alignment leaves largely unchanged (0.502 \(\rightarrow\) 0.497), with the merged model functionally dead. The tested alignment removes the same-task barrier but leaves the conflict-associated barrier intact, supporting a functional-conflict interpretation without proving optimal alignment: exact recovery of a permuted-and-rescaled copy validates a special case, so the removable share is a lower bound and the residual an upper bound. Sweeping conflict traces the cliff as hybrid fitness, 0.97 \(\rightarrow\) 0.03. The conflict floor itself is information-theoretic (no single model can satisfy contradictory conventions; SI Appendix, Proposition S2), with the framework's role being the \emph{structure around it}: which divergences generate conflict, and what moves the cliff. @@ -100,7 +100,7 @@ The pre-registered \emph{emergent test} constrains the claim most: true BDM inco \begin{figure*}[p]\centering % fig7 \includegraphics[width=\textwidth]{figs/fig7.pdf} -\caption{Model speciation at three tiers. (A) Analytic model: hybrid fitness tracks the parents while lineages are compatible, then falls to inviability; the denser the incompatibilities, the earlier the fall. (B) The isolation cliff: probability of hybrid inviability against divergence, by incompatibility density. (C) Trained networks: the merge error barrier between two MLPs before and after permutation-and-rescaling alignment --- the same-task/different-start barrier is a coordinate artefact (removed by alignment); the conflicting-task barrier is left essentially unchanged. (D) Sweeping the fraction of conflicting classes: the residual barrier rises while merged-model accuracy falls from 0.97 to 0.03. (E) Language models (0.5B LoRA children of a shared base): on shared ambiguous prompts each parent performs under its own convention while the merged model falls below both --- function-specific hybrid breakdown. (F) Divergence without conflict: over-training disjoint specialists from 1 to 12 epochs produces no isolation; the merged model tracks or exceeds the parents throughout.}\label{fig7} +\caption{Model speciation at three tiers. (A) Biological model: hybrid fitness tracks the parents while lineages are compatible, then falls to inviability; the denser the incompatibilities, the earlier the fall. (B) The isolation cliff: probability of hybrid inviability against divergence, by incompatibility density. (C) Trained networks: the merge error barrier between two MLPs before and after permutation-and-rescaling alignment --- the same-task/different-start barrier is a coordinate artefact (removed by alignment); the conflicting-task barrier is left essentially unchanged. (D) Sweeping the fraction of conflicting classes: the residual barrier rises while merged-model accuracy falls from 0.97 to 0.03. (E) Language models (0.5B LoRA children of a shared base): on shared ambiguous prompts each parent performs under its own convention while the merged model falls below both --- function-specific hybrid breakdown. (F) Divergence without conflict: over-training disjoint specialists from 1 to 12 epochs produces no isolation; the merged model tracks or exceeds the parents throughout.}\label{fig7} \end{figure*} \subsection*{A controlled predictive test: functional conflict, measured pre-merge, predicts merge damage} @@ -115,11 +115,11 @@ Headline quantitative results, with sample sizes, uncertainty, and outcome defin \textbf{Design rules.} As engineering guidance, the results reduce to rules that an operator of a model population can apply, answering the four decisions posed in the Introduction. \emph{Ground every generation} in verified reality --- a few percent retained most diversity in the tested settings --- but price the rarest capabilities individually (observation probability \texttt{1 − e\textasciicircum{}{−m\(\cdot\)p}} per batch under unstratified sampling), consider targeted sampling for the deep tail, and use recombination to recover rare capabilities still retained across complementary parents. \emph{Merge, don't blend, when there is headroom}: keep specialists intact and route, or breed-and-screen candidate merges, whenever the naive average is far from ceiling; plain averaging is adequate only where a strong base has already composed the skills. \emph{Match the operator to entanglement}: merge freely when skills are additive; sparingly, with offspring selection, when they entangle; and expect the champion-optimal mating breadth to narrow as landscapes roughen. \emph{Preserve diversity as a first-class objective}, because selection can only preserve variety that exists, and in the tested society its removal produced a distinct failure mode. \emph{Before merging, measure functional conflict} --- cheap, pre-merge, and in the controlled setting predictive where the tested weight-distance baselines were not; and \emph{do not treat divergence or specialisation alone as evidence of incompatibility} --- in every regime tested here, what broke merging was conflicting conventions on shared circuitry, which is the thing to detect. -\textbf{Continual learning at the population scale.} Within a single network, the discipline's remedies for forgetting are this framework's operators writ small. Rehearsal and replay of stored data (26, 27) is grounded inheritance within one lineage, and the replay fractions the field settled on empirically, on the order of 1\% for instruction tuning (48) and 5\% to 25\% by distribution-shift strength in continual pretraining (49), sit where the minimal model's operational threshold lies. \emph{Pseudo-rehearsal}, the replay of a network's own generated samples, proposed as a cure in 1995 (50) and revived as generative replay (51), is precisely the ungrounded null studied here: immigration from a drifting source, benign for one hop, compounding over generations, with verifier-filtering (29, 52) converting it back into grounding. Parameter isolation (53), including frozen-base adapters, which forget far less (54), is engineered decorrelation; complementary-learning-systems consolidation (55--57) is the periodic adapter-into-base merge; the recent turn to merging as a continual-learning mechanism (58--61) applies recombination within one lineage over time, where this paper applies it across lineages; and the observation that rare examples and long-tail knowledge are forgotten first (62--64) is tail extinction seen one model at a time. The mechanisms differ (forgetting is largely deterministic interference, collapse is sampling drift) but the victims and the remedies coincide, and to my knowledge no prior work carries population-genetic formalism into continual learning. Read into that field, the results offer: (i) an equilibrium theory for the replay ratio, with the sharper prediction that the required fraction is set by the rarest capability one refuses to lose (the \texttt{1 − e\textasciicircum{}{−m\(\cdot\)p}} law) rather than by average loss, testable against published replay sweeps; (ii) a \emph{failure theory for generative replay}: self-generated rehearsal is safe for short horizons and compounds into collapse across generations unless verifier-filtered back into grounding (29, 50--52); (iii) \emph{pre-merge interference prediction with a mechanism}: where the current state of the art fits regressions over candidate metrics (47), the functional-conflict measure arrives at a convergent signal from principle and comes with an operator prescription --- when conflict is high, do not average; route or breed-and-screen; (iv) a candidate \emph{decision rule for the consolidate-versus-stay-modular question} that currently splits the field's practice (keep adapters separate vs merge them; 57--61): union-preserving operators where headroom exists, fusion where the base composes, consolidation as the slow-store step; and (v) \emph{tail monitoring as the leading indicator}: continual-learning evaluation that averages over capabilities hides exactly the losses that drift theory says come first and, past a threshold, become irreversible. On that last point I note the standing objection that apparent forgetting can be skewed task-inference over latent capability rather than erasure (65); the irreversibility results here concern oracle-measured behavioural distributions, and distinguishing latent from extinct capability at language-model scale is an open experiment whose outcome would be decisive for both readings. +\textbf{Continual learning at the population scale.} Within a single network, the discipline's remedies for forgetting are this framework's operators writ small. Rehearsal and replay of stored data (26, 27) is grounded inheritance within one lineage, and the replay fractions the field settled on empirically, on the order of 1\% for instruction tuning (48) and 5\% to 25\% by distribution-shift strength in continual pretraining (49), sit where the biological model's operational threshold lies. \emph{Pseudo-rehearsal}, the replay of a network's own generated samples, proposed as a cure in 1995 (50) and revived as generative replay (51), is precisely the ungrounded null studied here: immigration from a drifting source, benign for one hop, compounding over generations, with verifier-filtering (29, 52) converting it back into grounding. Parameter isolation (53), including frozen-base adapters, which forget far less (54), is engineered decorrelation; complementary-learning-systems consolidation (55--57) is the periodic adapter-into-base merge; the recent turn to merging as a continual-learning mechanism (58--61) applies recombination within one lineage over time, where this paper applies it across lineages; and the observation that rare examples and long-tail knowledge are forgotten first (62--64) is tail extinction seen one model at a time. The mechanisms differ (forgetting is largely deterministic interference, collapse is sampling drift) but the victims and the remedies coincide, and to my knowledge no prior work carries population-genetic formalism into continual learning. Read into that field, the results offer: (i) an equilibrium theory for the replay ratio, with the sharper prediction that the required fraction is set by the rarest capability one refuses to lose (the \texttt{1 − e\textasciicircum{}{−m\(\cdot\)p}} law) rather than by average loss, testable against published replay sweeps; (ii) a \emph{failure theory for generative replay}: self-generated rehearsal is safe for short horizons and compounds into collapse across generations unless verifier-filtered back into grounding (29, 50--52); (iii) \emph{pre-merge interference prediction with a mechanism}: where the current state of the art fits regressions over candidate metrics (47), the functional-conflict measure arrives at a convergent signal from principle and comes with an operator prescription --- when conflict is high, do not average; route or breed-and-screen; (iv) a candidate \emph{decision rule for the consolidate-versus-stay-modular question} that currently splits the field's practice (keep adapters separate vs merge them; 57--61): union-preserving operators where headroom exists, fusion where the base composes, consolidation as the slow-store step; and (v) \emph{tail monitoring as the leading indicator}: continual-learning evaluation that averages over capabilities hides exactly the losses that drift theory says come first and, past a threshold, become irreversible. On that last point I note the standing objection that apparent forgetting can be skewed task-inference over latent capability rather than erasure (65); the irreversibility results here concern oracle-measured behavioural distributions, and distinguishing latent from extinct capability at language-model scale is an open experiment whose outcome would be decisive for both readings. \textbf{What is borrowed and what is new.} The collapse-as-drift diagnosis is established prior work (21--25); so are the empirical facts that merges can beat parents, that decorrelated parents merge better, and that naive averaging loses to interference-aware or routed merges (4, 66, 67), that model populations can climb (5, 8--10), and that merge success admits ML-native predictors (47, 68), correlational where this framework supplies mechanism; the reading of sex as an algorithm for mixability in the theory of computation (69) anticipated the transfer before model merging existed. New here is the framework-level synthesis --- inheritance, diversity, and compatibility as managed quantities --- together with: the conservation law for blending inheritance and its operator boundaries; the per-item grounding floor; the society ablation with its complementary failure modes; model speciation as a named, tested question, with the coordinate-versus-functional decomposition under permutation-and-rescaling alignment and the emergent null that bounds it; and the controlled predictive test with its controls. I claim the framework generated these measurements and experiments; I do not claim that their outcomes validate a uniquely population-genetic mechanism, and one refinement it proposed was not supported. -\textbf{Limits and open problems.} The demonstrations are deliberately small: exact where small is a virtue, sign-level and seed-replicated at the language-model tier, on constructed task families with a trivially separable router and one model lineage (Qwen, 0.5B--7B). The composed society has not been built at language-model scale. The predictive test's next bars, in order of value: generalisation to \emph{unfamiliar} conflict structures and real task pairs; a demonstrably better \emph{budget-matched} merging decision; then scale replication. Beyond engineering, the framework's hardest open problem is the fitness function itself: selection optimises what is measured, and for knowledge systems the persuasive and the true compete --- grounding against a reality that can refuse is the only anchor I trust, and institutionalising that anchor (verification, replication, and challenge among models) is the society-level problem this paper poses but does not solve. What biology receives in return is a new model system: populations of learners where every genotype, environment, and mating decision is observable and manipulable --- where the evolution of sex can be studied with interventions (unbounded parents, offspring preview, directed mating) that no living system permits. +\textbf{Limits and open problems.} The demonstrations are deliberately small: closed-form where small is a virtue, sign-level and seed-replicated at the language-model tier, on constructed task families with a trivially separable router and one model lineage (Qwen, 0.5B--7B). The composed society has not been built at language-model scale. The predictive test's next bars, in order of value: generalisation to \emph{unfamiliar} conflict structures and real task pairs; a demonstrably better \emph{budget-matched} merging decision; then scale replication. Beyond engineering, the framework's hardest open problem is the fitness function itself: selection optimises what is measured, and for knowledge systems the persuasive and the true compete --- grounding against a reality that can refuse is the only anchor I trust, and institutionalising that anchor (verification, replication, and challenge among models) is the society-level problem this paper poses but does not solve. What biology receives in return is a new model system: populations of learners where every genotype, environment, and mating decision is observable and manipulable --- where the evolution of sex can be studied with interventions (unbounded parents, offspring preview, directed mating) that no living system permits. \textbf{Creative diversity.} Collapse is not confined to facts and skills. Homogenisation of \emph{style} is already measurable: models trained on model output lose lexical and syntactic diversity across generations (70), writing produced with model assistance is individually better but collectively less diverse than writing produced without it (71, 72), and the house styles of the large assistants are recognisable enough that their tics serve as signatures. In this framework these are the same phenomenon at a different locus. A voice is a distribution over rare stylistic variants, exactly the tail that drift erases first and that blending inheritance averages into a common register. The remedies transfer unchanged, though they are untested here: grounding on stylistically diverse human sources, decorrelated lineages maintained as distinct voices rather than merged into one, union-preserving recombination over blending, and selection that rewards being different as well as being good. Whether these preserve measured stylistic diversity at scale is an open experiment that the framework specifies. @@ -127,9 +127,9 @@ Headline quantitative results, with sample sizes, uncertainty, and outcome defin \section*{Materials and Methods} -\textbf{Analytic tier.} Pure NumPy/SciPy Wright--Fisher simulator over \texttt{K}-item distributions (knowledge as \texttt{p\_t}; Zipf-tailed truth \texttt{p*}; drift--grounding--refit generations), extended with a learning kernel (smoothing/sharpening refit), multi-locus genotypes on additive and Kauffman NK landscapes, n-parent crossover, and finite-population society loops. All parameters live in per-experiment YAML configs; every run derives all randomness from one master seed (\texttt{SeedSequence.spawn}) and is bitwise reproducible; scientific-validation tests assert the closed forms (heterozygosity decay, immigration equilibrium, closed-form union) to <0.5\% and run in CI with 151 further correctness tests. +\textbf{Biological-model tier.} Pure NumPy/SciPy Wright--Fisher simulator over \texttt{K}-item distributions (knowledge as \texttt{p\_t}; Zipf-tailed truth \texttt{p*}; drift--grounding--refit generations), extended with a learning kernel (smoothing/sharpening refit), multi-locus genotypes on additive and Kauffman NK landscapes, n-parent crossover, and finite-population society loops. All parameters live in per-experiment YAML configs; every run derives all randomness from one master seed (\texttt{SeedSequence.spawn}) and is bitwise reproducible; scientific-validation tests assert the closed forms (heterozygosity decay, immigration equilibrium, closed-form union) to <0.5\% and run in CI with 151 further correctness tests. -\textbf{Neural tier.} Trained-network experiments realise the same abstractions with an exact oracle: histogram/RNN/MLP/VAE generators on a synthetic mode universe (the histogram model reduces the harness exactly to the analytic tier --- the bridge gate), and a convolutional VAE on MNIST with a frozen CNN oracle (98.5\% mode accuracy; confusion matrix recorded as the measurement floor). Speciation experiments fork no-BatchNorm MLPs (784--512--512--10) from a shared base, weight-average, and measure linear-mode-connectivity error barriers before and after alignment; alignment composes deterministic Git Re-Basin permutation matching with exact per-unit scale canonicalisation (the unit symmetry group of this class, as the alignment search space; control recovery does not establish global optimality), gated by exact recovery of a permuted-and-rescaled copy. +\textbf{Neural tier.} Trained-network experiments realise the same abstractions with an exact oracle: histogram/RNN/MLP/VAE generators on a synthetic mode universe (the histogram model reduces the harness exactly to the biological model --- the bridge gate), and a convolutional VAE on MNIST with a frozen CNN oracle (98.5\% mode accuracy; confusion matrix recorded as the measurement floor). Speciation experiments fork no-BatchNorm MLPs (784--512--512--10) from a shared base, weight-average, and measure linear-mode-connectivity error barriers before and after alignment; alignment composes deterministic Git Re-Basin permutation matching with exact per-unit scale canonicalisation (the unit symmetry group of this class, as the alignment search space; control recovery does not establish global optimality), gated by exact recovery of a permuted-and-rescaled copy. \textbf{Language-model tier.} LoRA specialists (rank 16) on procedurally generated task families with an exact-match verifier, on frozen Qwen2.5-Instruct bases (0.5B on one 16 GB GPU; 7B on one L40S). Operators: weight-space merges (soup/TIES via adapter arithmetic), per-input routing, and Dirichlet-sampled offspring populations screened on held-out validation splits. Multi-seed protocols fix the test sets and vary the training seed. The predictive test computes all predictors pre-merge (generation confidence from token log-probabilities; base-model gradient cosines; exact r-space LoRA-delta geometry) and evaluates merges on held-out tests; robust statistics (condition-clustered bootstrap, paired predictor contrasts, leave-one-condition-out prediction, multi-reference outcomes) are produced by a committed script. Statistical, per-seed reproducibility is documented for GPU tiers. diff --git a/paper/pnas/build.py b/paper/pnas/build.py index b9301ec..68a0a14 100644 --- a/paper/pnas/build.py +++ b/paper/pnas/build.py @@ -23,13 +23,13 @@ OUT = HERE / "body.tex" FIGURES: dict[str, tuple[list[str], str]] = { "fig1": (["paper/pnas/figs/fig1a.pdf", "paper/pnas/figs/fig1b.pdf"], "(A) The experimental programme. Each population-genetic abstraction (Table 1) is tested at up " - "to three tiers, ordered left to right by increasing realism: an exact " - "Wright--Fisher simulator over knowledge distributions (closed forms; bitwise-reproducible), " + "to three tiers, ordered left to right by increasing realism: a biological model (a " + "Wright--Fisher simulator over knowledge distributions; closed forms, bitwise-reproducible), " "trained neural networks measured against exact oracles (recurrent, feedforward, and " "variational-autoencoder generators on a synthetic mode universe, and a convolutional VAE on " "MNIST), and language models (LoRA specialists on Qwen bases at 0.5B and 7B, scored by an " - "exact-match verifier). Colour separates the two categories: the population-genetic theory " - "tier in blue, the two AI-model tiers in oranges. The same abstractions are carried across " + "exact-match verifier). Colour separates the two categories: the biological model in blue, " + "the two AI-model tiers in oranges. The same abstractions are carried across " "all three. Rows are the framework's mechanisms, each defined at the left margin; filled " "cells name the experiments run at each tier, and each carries, in its corner, the figure " "or table where that result is reported, so this figure doubles as a map of the paper. Each claim is tested at the cheapest tier that " @@ -48,7 +48,7 @@ FIGURES: dict[str, tuple[list[str], str]] = { "grounding."), "fig2": (["paper/pnas/figs/fig2.pdf"], "Grounding is immigration. (A) Stationary diversity against the grounding fraction in the " - "minimal inheritance model: simulation (points, 95\\% CI) matches the exact immigration--drift " + "biological model: simulation (points, 95\\% CI) matches the exact immigration--drift " "equilibrium (dashed). The equilibrium is smooth in $g$; $g \\approx 0.05$ marks the " "operational threshold retaining 95\\% of source diversity in this setting (red line, " "bootstrap CI shaded); the hollow point at $g=0$ is a finite-time value (the true equilibrium " @@ -56,7 +56,7 @@ FIGURES: dict[str, tuple[list[str], str]] = { "generation on its own output (rows: generations 0--15 of an ungrounded lineage) collapse " "toward a single blurred mode; 10\\% grounding holds all thirty modes (quantified in SI)."), "fig4": (["paper/pnas/figs/fig4.pdf"], - "Recombination in the minimal model: blending inheritance and the Fisher--Muller effect. " + "Recombination in the biological model: blending inheritance and the Fisher--Muller effect. " "(A) Expected rare-capability survival in a child refit from $K$ uncorrelated parents: the " "output-mean (blending) stays at the single-parent level --- the first-order cancellation --- " "while the union operator (strongest source per item, renormalised, oracle-identified) rises " @@ -81,7 +81,7 @@ FIGURES: dict[str, tuple[list[str], str]] = { "preservation strands it lower. (B) Population diversity. (C) The self-consumption signature: " "conformity minus true fitness (mean $\\pm$ 95\\% CI, 12 replicates)."), "fig7": (["paper/pnas/figs/fig7.pdf"], - "Model speciation at three tiers. (A) Analytic model: hybrid fitness tracks the parents while " + "Model speciation at three tiers. (A) Biological model: hybrid fitness tracks the parents while " "lineages are compatible, then falls to inviability; the denser the incompatibilities, the " "earlier the fall. (B) The isolation cliff: probability of hybrid inviability against " "divergence, by incompatibility density. (C) Trained networks: the merge error barrier between " diff --git a/paper/pnas/figs/fig1a.pdf b/paper/pnas/figs/fig1a.pdf index 6d8f4f6bfdfc5d3c88c9b8268eae3c89ef5fcdc2..2ce40f0ce988d8c8c1c6b79c6eb39bb5ffd708dd 100644 GIT binary patch delta 13256 zcmZ{KWmuJ4*ELX3C z?-UW$`Qcoi5n3*Nn>3_AS98tHgTh7GFkc}x&ZrDD)!>YQo1CkYi=cf8XlB6O9ZGCV zIHRZ9!zFAU1E}Qoj>P*h5%d*omsLX>+ySfUP+UFqm$$0G=m&0`XRkImls_{Sru+4PoGS?#ekdn>yGGtNoe zX=swCsp^VgQaM6{y|+8fg*^wcYkll8V-mxU3)qtil{fTxv~}n80kHvY6bI7G!$tYo z+Nkia`j~)#3Cr9>q>#4@H38r7VM^{4kqLw8=27tA}&enEkM!;fsH3! ze)JRCpN1y^AsL_K?UNf{a|QDXUBn4;Fv_Q~@1AKlrK)mLtKIl8C}#IKu3Ec#=bd)t zj(#x%p7W<2-beJb?_iPEyRD|BGV#xlnw=VYg5CsO*NLatN8#Q zD9#)iTWUq$i?}jm+h|-9xcfD$(*B5jk?fkr9_cvCHgbBGp4R);BXF9g|8fy~*5Iwz ztjd85qujn>vGma?v?>u%;EDo`JcS~u`ilLz0Zk^36-tpdau^aO z$Vk+KV=8oOz@LgpTmrdIHcN=2uM0?BuqG~?3E@v3b?yZS9FXtDdGo*F^$m=rkH|k# zjED*}rr}#P_?)CyWgReVq@o*`{YL)i=zkVeXCCHOYx)nxHbBnDN~6+ z_=4j)7;NzhWLY3G;_nb%H2+p=v;oQW`6raAeAr*l>Ijs`zvd>=W{rCAX9v`uEo5R9 z+DspRCP8`F;8Bp(nL0nAocM0YPASbu$S|GtVocPfPHWQn6cV zq~9_ArK5eCLO)C>lCM%P$YYu&?HYLsfDGS=q-fbsV8Zd!kICJ$DUoc}JR$c`x@SDg z+5>!maoW3Cr4iY#Et+JG%czR`l^ghKbKSp(tb19=32O|ZO2_Qt=YDGU>|nIJvHa=W z(XYC86ffj$Yy8xvnaXjH|4#aFT8d%sHr2IcWm$tcI-r)TnzAne!%99(IXphVSeSOs zU^mII$_=()RGW5NO~s>_Bfalh+QYB+c_{cAd6$)?2(+#E!ThDhd#IlogZBqnl-msZ zxQh`H@$YnJc)H%O7_-v$Jz37TG4Aze&|f zR_I3*zX5^N4Y!t1tw`x8=D0-uHu*Pu)cg)1z4Q^ZHtPC=S6KbZ=*DkIiN|yDw8@u> zt`PIY-n*ST?OH(Dd_Z787{yHH_mO{G^si062EV?d3}dk2cEBFZr_&)`o=k#IXg!+p z z4}OQIt=&1736%;r=32_fkMrQUpH)LDX-PwWv4~#jk?ININ-(I{8_lfm_%ixy_r!-<)E-f?Ou$YgYcv&fl434o_3LvgPrxV6#0ZnszQ%re+6PE_61L}Fik))5areu z%#J^GrCi}l{N7fs72u>aVq7{Y#D9c~rGb=ll~A5=vv$|gG^Wjr;M?NsRAy1C8gmQV zY{$fEfQUJ19)aclmPQl;PD;9}FV&H9wywy@-9?1YXe3{a7%7aUs$_g@ckRUPjZl7; z^e#1KB}TN~( zv&ET}ErU>^xFWmz6R))-rVrbX>*t))yQ1t#jPK;_Jw?qbhDF;k5)KaCPI){$(- zv|Y#3vIputlkI>?E<0Qgzmm_;(LBXWx7zd$10hb9e%SKV{XRz8!H$R{suHzgmuO6U zog`H&FrkO~6cf4;J>2!-$tDrSYPA`q&a@B9|m+GL!8mQVbnD#QSXI*v3cz=?tI zWNF(NWYeuN&yTb;nz_q%I@G&A}4?)Ba{Ash|nC~l{UMfR3)b-b&5WlVFS{34c zT8bBsNBO(lDh9Teyf%(%B@JleUNLx}PWXwsJG`>NM=VZdSKK$L{ROvW*IvDD zcu|B4zAIh08LgukQ3UkMdW}Mb@(sKsCN?owjOIFKg8hhpz2W%m3Ld2lf79MI3YW_8 z{;q4g{h51ojOliv?oD~~p#__dLEb-zYX}{YJhZMZH6{dl!S%FxPW>*ZQ+K4e)mvtd z%4B9wK9<0E34ElyygPzp?Zffo1o@^1>W_Da`|5v`F}GZOiHIq?XJ)f-+ygu3z7gHF zpIDpG-qY`oVvTA$5RomeA>kXY>LU)7vc(u6%BA~*ObV7woaB|SOxG2CTX;wZB^_RU zM5gPNs`ASjNaom{f%|lmq*2ziANy-kME|&24?oOTrBfOB+XT`C8x?iy&Q?kE zTj;Q~RhdYM@~IM2x6xDk87b@TM1I|@QAbe0*a1BY@3i59kln+K>gKoF*B7lyi31<^ z1(~TR8eQ9$$tRsbJ9^)K&3tw%>-UG---HSkn}I`BR3 z{CM)}7LYx6A+Zv0j+*J-RYI@iR^M6}riP&DJ8!Dopd>`Z1j!{l^oDzv2z`gTcI^=y zkFhpdN$IdTGkV`VN^u=8k5&!6Clmh4?U~3cYq(l7eT8E8Th;9(S1(yIuv%7SH;Mh( zt-6IMi#Gws%;p`8a z|8%eHczB1_GXQXW!t`8tMren_*+0=IB?FZgvd{E8(C)(LTMU=QbvE8tfZA=C5}rR# zIKK>G>tlWqLvk@D@uzrd&_H<4JM9@NXlBE^mt(&0SH*JQRm4*u3P=tO+fWffM z1<=CR7H_uMY>pR`rcFN}XnLZ8)I~DDX;_as5*WF+zsI2_r8go3*91~9EDkdSddL^Q z6G;nxG^*$PcrC3}$;3j|boV4L2BDWrswyd;Aj!Mj?Z<1x$k9#d%wf!a>_Gdi#r8RX z5N_O}x<0qdKLLHgGMe0nvFqhmHPY!F-D~!g`BZDHO-S%tVBNByXWl@BkrPCL*GQHv z5|aKq$mc!kxbXO(Q}zbeXqfCUN>w;Z;OBZ)T{X-0z(8U0ffEy$$*L{)cc{fq3rn?f zzmX>Wa;Y)#1EkCTLBIu?LF)lU3@=;bz-Xq}=R+sTaeVg&3e1gj%_TDCdv=4RCHR{k z$OS(**A*+_S4j)Cs-(M6tx}0a2GMkfBZ$Bx8AYZj_xTV8dAi}jslNLiF=j`VXasnE z!YRu*)d^6E&tKy*+?vYrv%@IAc=L7+QYw;xPw~#tS;DO}yil6G$?-=^A9wJsHBxJO z)etrkZ>tMI`U=w5Y}9gdYF-icgo@iPvav-a2#~Qnwp)LVDQa%^X&Qcyxktn3`$-t{ zd~@~+N+)oTQTp&S$niugS5(ALn0dt3VNG?cuUJNUOjAHk{8lGHL_$=N$HZAqvhm{^zAo~ z`%%v$Ii!M_)&yO*Qb`x(#Nc@N#EWb!R&M*$N)lHNMp(?PQ;)*~R%e#8Umbj~6Cm!1 z^b(9VS68|ndYHV&dPUYboZ(Yh#ynW+=6HFm3nZnRXoX<$?!&Ps?lh>byZBd9`UsCQ z<=8vylP&<9REMFzBGYw=F}He-VoCRC+$d1t9=o&ocelB0o_S@>;thq9ws~ID8{Ch< zC`QP+P5z91yGP?tiS)%+?)@2CKJPnH<1Dcrz@k6WSCmG{1}*Jx#ux1F_DSn%PAWMs zl+@=ORb3vNHgKU$vkVvHvdn7l^crIMVQ#{(_M3c z_@8#Yv1yz@9{J+cb#Bb_>2Ao3fFpz(buG~`YGR^%{3y1drpG>Nf|c@9&G!8a^B1o^ z%*@P6?ii!%eYJzVOC$Zq?8LHe%e&(ztGmhP4qUzZ?yZ&sAyH~7y_f#=D-R6toQnw= z=fvORA4If3fO$yq>w4ArolhpeVoc0;QuZ`&G)v&|-8Nv#7<`*ST%K4!IWt(lZ-BC3 zuXlRAM|%yN*HWaf81pc`k2mb=46j!5FCN+JnBn0_zX~VHl?prHljiOs(+U%+`uNkQ zT6g0gb#JT(=&{l`zWSn(eN|7|HWf6m@N)%&M`GkJ@3Aa7EXcv8r?J8Ha9ZihU9bCu zNw3IB5O>ht$w{T3PY1v0qu6!B5R=uY5|ex{pZk~&{M9gO#p>B6qW$k?5tKvz3K6^= z_Qy2aRe6O_^VWbx)5Cctvqksv#H^2O8?8Tg)Jd0S4`UnpMxW0LpnmVHbLMEqwn#Q= zQg7<3Wqe#HYP%@wdCKTAzo<85P)4L77dbRPIU5@%`6-`Q>q~ysAxWAZ$T~^`_46qu zbHeg?*Y9QJ!K~t8d^y|ohYim^C{BkCR&m*=n?7aJtUqn^{?($ka5Xpdx<_sDLo*SP z>K~4$nj%g-y%4wA5%RW<7eww42=vHQ{ZFV%RDk`J6Lmqt_KX!ja~Y6#*KK$E1ZJm+&imAF5RtjIndYit1wEc?!zK8 zSVrX2K0neoOnN?Ia7jF-nF#n~8Tu%tDEvIko`_~sxj3fg>1H;*ZAO>Rq(5UX@Jf4C zKDCHjR}KL5#!`Yj)ZI9oP8X>BcAZ!jb=`yXg1YBwXTAY?=9JsEu1NxHI?GZk+S8{c z$31DkPOW9)=)@9ZA2t#_?u!zM?{dy4({&;llNtev1h?Xyf0;6zQark6u1oCuaj0xm zFHShB86vbU$*XY&XsZ9%VPMp^(KUENEpBtIl9GKO9b2{pyg3YOUnC+J<5gZGw8g)^m;y z8qIW(3L^Mr@OXhe&~pDJkso8c^bJvhK>DV<))y=ST$xTLq9(jfzO6UE8C9~eE{a#H z=I~QD@g2R>b%5VJRfD5=Uzj5bJ(JQ?%3GqDKJB{ZsOT%{T_^4tl%n*>Q{*@_3vI6f zR7P^QS*|3Q`{bC2Fo~h2O$33Bx@6TZ6?9K}f1T&EsgS5XzI6n&f1m=JEjRpPI65?K z+6FPkOms)@cz|9`KYY*GA?=e%CelP9KF#G&g^S2;UBdrN4pJzP@$ z79ZpDwg%%bWr%1qZuLjzX8VhojjzzaKC*HGcj9)v&Ay+NJfvq9NIt&FGr%u^ulr_F z);2Wn(D>_P7o?MpKwQW!*yeg^d9gg1E2%Dvja${nZ*FgiFf%r*3e5^)s`tN+n$xMQ zehZ!VbjhkSS<2*-K(Qut_8=18I)|u-+V`3@BRmAgzpRi~?>%5ry=6Z*Q2jJ&xo8(u z8hB(OfA5(>FsIRNhL@sm=C@P~2Ds8hR~9C&-s8fAOETP>sZi;zpJEd$2)wl+Kwvi( zTfrGhhQBQ~`f04v;bh0rq%ga~+PO}it!l^#DJ0oYXE8U` ze3IHq`nAuODc*XM)a>Opghb=yP`*Rs89`py_w=|6>u2iE?wWwb6wKivREC4K!qJNL zp{CS4Qjka4#94Hj0j~<`)W-a7J`_)^a2ZB9xqopj>;3u~mEfDB{rW{UaSQX3f%v<- z93?EJ<o4N?MjS-J~epje_$qk{` zR7BdfV=}XkUzKyoA?H$DRt;`3wI1>K*f!_39ynjUyZogpcfjG|tl=t0*9hKvRdqHhBj4y`51wU-|s?KBzMO=+*Iw!PJlQg_j+72))m(n=5Hh(6x1NsUh=ZBGdS2 zr;-u49_K_%CG&y8V}aVuZ~0z%mslSQxrs&b)ilp(g zRm$?4t!@&rQy&JXmtGC2NPC>%o%Pg(hVM@43{E%9_3l?er)C92@kvytlQ>@vD{Kuo zUWM#blA^w-|MtF7<2QaM^|6Y!+q8dle$=t)s#*WkH>sdux7*R&!JaCNKJAw6N^dLt zGYDyP?fp5W$ycife3@w4AFFS-n7@ry)^liMRTGif*9zuupzsZQGFKFlBs+F)qeW`q zUX8edC=}H8ZM(D!5D@&I{Xt=PJSng=Lf@0gKIMSRxqS~;Jm<* z9g}8JHUX@EV2D`;heqQMsl90_qDVl*{ID_tGPRPF%-m$!s7$G^KKE{fcHfx zP1}o9`b8}fSrO)mTyfTu5&I+cZ0|s${a^x7pHMOJt;VpWoe7>qOOi<2CdSu!>ir)F z?&>7#Zd5*eGui&0YC6I6yGoS=D){!zP;s`Y=ZqEwEMl34*3@C44w-~nKYn=|34>@x zH36tSv~l@^ehy|z3$l!I{y@s3V$*IXvd)09hA3m6A5JIj8KG=LL-l))jWUI#zG+YK z?Dw4B7K**Usoa3)Wjf}WK?RzA_G^8+r4$LhRdDoigKKcRr-^Uad~}}AhxFQWgBhKT zq$thkQ0CH!;0ZG|Q_b_{*8Da%hI}ZAm#s2t{maDG;HWGs$BTF=Y63kP|F+5ZwuSZV z*D__gLwjlZF7FS<=BNRlMN7Ii1z5)j)edd$4aEjk*A}h&9T3>t zw0#iE-0|_T8B2ec!C5Ycc?cim`(!Bxyj{)vE1hO%Fvxb>byKw-cdr{O>sM$qFdl1@I(`)z*XT+a@ng)*tgksEFVEo=>bq-@IZL z95MD}B+<62=Nj9rOv$yEcYFgAHZl3zr;m)VA0LrA^J5)leqr$QgWR!#CAstSjbE|M zFFi2_UlZUB2FDrA4JY{0o4?NAJJ#-~SfE6|3#{y(75^@56j42WrDcgEeNO=xyzulj z_0s$?rmoL&a>x4#fw>O;3`FcJyYO{Q1$vq*-0x0UJ89ZfTI@7e00J#ilAN7BZqpQw z%dWIyS>?@IXc0s0w#l84o&Du6nQ>iLQ{4KzsFO})m<+Ypm*msy)awePjenG7`q-c_ zInb?f{?SZ{Qft>A1fOZOLUn~bLMcwh(-o$DLn8}kiI-ekgYxAJb?+7yCDz`r%q~ja z4JzDkdQ8@R181r)grYrUyHPKd7dOscjaaMy3MXi|DZ7Ypb^5x zn<93VvEZVHFLCLq)LMU10^v@d`dgQq$W#Cg`!8p|mw9we0l6I4zfeE){~FHkff_Ub zfOG0zu%(E9=d_>-u%60_OZ{~|8otxms1l~8)){;A@V7IFv%k2Ttw*!W8)Vt*srp-B zROFqQzGoke6=Xi1hQq!sDI?ga+hy$VJW0S=hD-DFsqZcRmoQkd`+Cw$hSb*gpjTCE zbuz`wiAyc|!M;y5eo}jtS;SetIzsJP=3Mo?sJzNq|NZft>!=sw1oLPe$97BPu6T*7 zMd(COM4NwQ2jLjI`&Y>lXUN)F7FW^DsuQ~H)4;C7T#GKM@_yFy{cz>&qHQRil!$%MJ1gDe+_jz04;92djI6|F)k-yNfuz4{Xkp6>l^EW)?r%>-!XH9ewdBWQ9C%-eg`-xO zc1gxz2XHIwqP%77x6s+?`Na$D-}b4BKbhx``tjG zDNhI{|I(tj^o`%3lECgBx76H=fWud^wY(pQ~Z&HwRR!?#&*z=nhpSUP@A01J` zKndaoaC)h$L}iXO?;^q>#3=I#~9u}Kvk^z*-U+d zx2B_(Q%O&~Yb#$%CQf5ytUlJ7?_C%ytU=GU4d(e?>o8dqms$$vsNO?}ZmlaTqriz% z%C(^GY=;qRhj)Q!E$eXx_M6EV>NO?hL)!Y)UnSfIrU6GGlTi=cBl=d%$hk^FirfjA z)P{a?2!5=b+#_kbrbo%^K+|gNho66q+)gi4dSQG7lPI5rFZXc7$Y(*TB-U!)B76fQ zCf4%wp7!RdWp^k2+D(>Mx7x>BV(4-ZL4c|F8m|QAxnIBWVC}cjA$)@9a|j<=n9mtK z%g^w$R?r44MTEEsBpu9+!AWs za2U6QHU$F2En$e>L9wC_l!b}7CGJw7xmCRXC+vBK3Iv}J4g;Y_RH5iFRWdY%S~r&> zHxP+!mf@eRK!CrV)VM*wzn-+8Z>gE%6Tpz@@prr^!y&gG0BzfusGl}Xcl|9bsw{a8 z3E3YC?+Y|(nj5KORIfJN?O(db{0#UTJiogVwP9Tjio&SpqF`D z;n17yzG>RvqX3P>dm-NxGj8qeNlJKo{sK7Zl5=%!AzQES-XlrByZ6$cg)4KLZRi5DSch*s(3T=PJJ8K8 zd8Ph9us?_DS-TsF^%-T`8M8#Q;S`NqU=v;)G0y zHBE%+8KP2op(_m|E_v#A-fr>iv6BFI=0C9i<0w92Yy@$fa009UP z0w)*=!&$?g@A6@QK-m9O^1p=ufk*@nsvr;&cMS#&4S@F0^0)^4!Bt9{3H0USU3O&Eif2}(^N1B3BHUB7=*&D z4h+T$<5IK!bP0zgND!{3usLwrgv5&K(taQS;4e*K>;EtOAOIM4xlbSf6sKYk0EUwU zgr)oCmVo{U=yGBp1o%$`|ARjW2!&mC2M`d6#5oKIjN=XoGg61Sgl5YFv^z1{wKEBsY4_RhSl955V?Lqah8Qk(wZ z4@LlRTm+23I_G6_A;37-78n5qT}B3sfWvTNkpNuIAi+q~WuF5h5x70Zdh_M#AlRwp zGSm>PRb1u*1OUTP69fqS+YV9xy-^@Q2oz^SAg*CU{ttdYuvgFJiXqs};p_*xtQ;)< z5Ug9^R0qb6OPAn>fMGZu00AR##zU;t;}95YThPnwhd{6{fV!LnyLwr65C|N8SrHHj z?t}or$^@rkC=~l4_aBb_FVaxNfA>TEJ?TSW034Tsz`+0X>i_HlfkAO}41vLMydMI` zVtqL=9DLc3Pyno$aE9MM1K(v+*&OTzNteFbGb?2m~%!v0otA|FF0O4ff)| z2}b@Ouwv22=_hvRzIL;d}sg*DvEJ%(aqFLMD3z+(7!@ZYg8 z+<5^C!#y;X3OHOqVW%ma8w=ZEoUbS-cAUaV48gTlCK*i*8l(j delta 13248 zcmZu#cRba7`?vSb&feSEBZ};ugi!VtvNy+!kmT4ady^S5vXz~zkZc**p_G27=YD$b z`*wckukY(~zMpHnulM!7u5%_o;n#k`ua3h-%_Xq+KHF|nI=*`x(*~{?oA`;hL~JR$ zWISgsTNBiA&ZQ(r7aPv38r(kgy!LgS;bh7{S;Y;#KxP)<*0_#b^n2r!p^DbKC;pt* z&n^xo!$x-l5Emy$OKr#7Ti-hRUo5?j5Qq8uX}NSRqDWSoo z$gHKVt%LQY4naco&47zKsqS{VoWY$s1y+5{w3Q;bFZMa; z&R9Df4;P|$HgQ%5z>$K3AxE#OhzN`uU9xZ+hx5|tjLK5m$SMw&_BUs2s74DJSc~XF z!qa{GgG-i(>$XRi)|Z+<%A-eM_X5GlpMv=AD!YQ^v689shDhH&gHE=AEmi*+Dio3E z1A)TCkyLzq_`0lY4;#`WmnmKNHe)3jn0OZ7hU)Fd9>}1fsv8%{6%orgwk8m9sK9zh zhf|$hQFyY54K*P0AlBdKt?(IlnwOS}cmP*!?cl8;w(Fy#L(j6=PM%W?N*}lBiVPNQq-PF#kZFp82}0e+TADYQYN9U8rAB z+0(Vqpg9y0vdLGFHkMXueiFY*(sg4@N7_w65Y2pvJSIRvkv3Q!U2e<#z53$Wv}8lFS=Fcty=L%#ojL$g4*v|p{n$7J|iug z^;J7G%K2Q1^yru$L^y#c?oTJ#C)GdJcu_!^-wYd?0QiT)f2fV`*OeAc1m-zlpgT|rED0PyG*)R(xrUc)2T)7`~DOHID4fa0sYQp{z>!x3t+ zZ0@o=1-WanO=j~+|Ji5*nT2|%P4$qnnb8F>6_qcUzBL(n?TatFIK6m z5#Ua9qRpWlxFfprwV2(4VuK{Px~$j1bII?lNRfI! z+Ora5EaUmCz(lHeEFtL6N;6%uJzONQECgZj^xU^}p6|KVt8Uh2+L|ty~kLR*1fkJFZ z$73K3O;!lx$~AFlYkTG4Le%1eh@;@%z)%bO=Q*nSS4sULVmA0RL*N(r_>q%%CAT-T z+?%Uyan*IKYdOqVKfj@5JRwb0Jtsho^aSEry0(o(n9tQ#H{J4Y8p$1Wey>3)3aw~| znLi~>DG(93pYWPAoxpQejcAS}(i4w|jp)Ujo@>_K*QR8mTnx&1y39#iLux8aHo5=` zO$2@6gE@apTv&B^D*lie%#%qiQsuHp7kHPJquN(wY*YQD$RLw9-2$~AZCgPF3pJYz zv5O|;$rEPGek>lJ`tgjolScCj^Lt2G6IIQBl3bh7KDcj}DAQLx(*?tvT7xHWmI(=ry&fMA{Ktq+3&#Oo!f+H*JTpcI<|Pf} zXnM#=qkXxr^poi zW1CC>(B#KDomg1QFJwS~_ZcKvu99EdPP*!NwIVW~up4iVl%@*r$)?~FBFg-WO}YgF zCY25;x?~W_B;YDH+1Y?=ZhwqzO2}S)DXVdxU-o zPnk};qy|VRMdSOPFi9>;&AN9a5r{oudfLrd1YUkh0doFQ~Em_W%qZE&Zrs}eBTBHkM;gsL-FjY04R4z9zCuY`e}(nvM9gN ztn)Iox!g_MqZ!pqFvYAt)v<%a7v!?jzHN zGr?UY?3=39f~y}}k<-j@pLH=(JV_&}CPcK2+k@{*ryCBalakXb_4D_CZ99E@-3oJcTex~44NxiNNiOrFohO)&zyDCc`_G4`%+&nVC;Mx&7!hxb5O1));RPx!{ zMo_CFcJC*6N+Ngr_(G% z){k@QZr-EarCTZQZhXa~pc`9T=`BZ)k=*ONnYy0<|8!1CQn9g;eT?S?%_$CLqi22d zrHv@;!I}Zh?jrXZn?IB{{T|UqTv#)KO6>7bhbXQUBS8p`OQ>+sdN(;6dF?n}hC5IU z5{U{V1@8U4puln0BP;wEEhoT7!{UR_@lE6g9l?zz8ns!WjzeE(9_z(B+U z<44tut%%St1QX4LWy@c zR$E21_RA)?ijXt9ZP6#Y*0JP9W{kznjSfj-25{O$;01sjK^S#MB5M3oe?CDTvjQ6ea+83nnWSO}1he_Jr(Rum(?ZnVgTjr#QqrY7 zX~gGh)6VirT0g_kP;6q-?YtkI(Tl?G0ITd}W_@&m3bu4Cw>ErnW1l?`cA%$GRM(rc z%%gc>V~^Wj{c=SA+1SJ*kI}V~X@$U1R49e`N!axp`ilk~LmmA@8f>^a-yxhS?r2Kw zogh=XVC`(@V56(oKkph)^a?;NMJQyv06)g8&epd`;Ani!f}TCbXXM>Dapnu7`8@L! z^FjmunyfCIosS!?%~WOyPy)$HB>YwCKfjmbZEs=g(1hMiGOo(ty?2l!7eNK3LFxLx zzZr#;)1U;AeBkt5gAMW2ws3Yc`*@!w4+oP{SW$jYO{cZ8x8mWPkVTM8>(T`o(j_t)_I;rq<_MWdns*YqJwO(V%iZ~!usAkfK&FWy$!GgQV zL_$$*@}W4>45w`(ZF*6+y6a!gQJ`=qT&a^G_-@3>B`W;Yy7bSb+4T0r!g3#1Uv-eM z&h>o0=Io`2>QliT=bIp{e7kkXQbzDbr(~V$DNXYne>pi@04vdfM(|mq*nXEo$hGSB zrKZPZbk}roGL&yGumXDUyEWE^(xng0@5e->;g{I>{t$6%naow`KiX(n zMqjSQn;yKk>DfCeEZB9~PAyP0TTP~D06tW=M5%CgUoq!_Gm!0RU%Xi_Qt6InZ>`Tp z6Gu}DNB|syLc58NE@cTZkWnK^g3V~@^iDv(j&_P3%Is~R0`K3*{27q?D^>@fdAtoY zy$hT|qh&ZZ21RT{UiwJ>`j$823eQpmh)&XY@Vdw0waaz1v{ zsE%Yo^L(>sIcFfrhev3$UVK*I zfBPm*Xi1pW;m#E@3PM7XNfMJc&)zVYe-sK4yotA!niy|Er^ZqWmdvdeC{5BWDjE0$ zZ7I{J@CcVxF^dqWR6$v-+1m{DSE>;E3?FlXHV)|zfBYcm+3=+@B(bqI>mB2}mwQef zO4&(fB~){x?j7{( zcs~{oWH(w#-Po`U?reyZ33taOr{5Ex{pchSy*^VD^s_tcj2P)F z3p?AHnq8xoOol_=cYqdgCFiIFt%Jg@N_28gx_#?jIU~MtSIdw(&*EG6d(Y-{)%9=# zYI*HCS0Uk6GS&o??OZV0YzaYTc2E8G#VvNB(@5I0L2F&2P#xF()j;yP)5hU6?P-#| zhOvX6wNl$n#4*vrAM$!7zMjxDNu5gQnSK%9!UW$!m`(^jS>l+G{Qrd2GL9<*_eN%i|&zLhi z;FP+hY1Wv*l;8Z#ZA4#XZvWwP@2A!YCDe=oBujWo2$cz!UecW%YXIR(QdY`0M7;JA zNiO#FUvdQpUAX1!JxNx@wgZz58e7a~XFlM3(Jilc(+F@0kMINo7(F0443WwNu$-?} zWaOXG$QoP}o+=sbZU`YGK|aoq0=W?6K@t8szH6x^!aZF+TM0MlW@(0bW%Uf#N<^xY zDGte4-uUR(3~_QT2dI5@k}qUQgSp1K>mAKiU}j0Kw%nuW$4b0GUuJk!pZ*lN(J7yn zhQFyj$lP_Cv^RLb=W4xNr9+g42HZGZ-HHMv+2B8b^5OcRd268FXW2$;fX{t<(ivAn zWfHe0WLxfr{j`r72z9r0SM7QE8YNnCzEym6;vqjG6}h_IDpMAtUd;SlsTJ8y#S@*b_SAcQ&t)yMR>$2Y1u)vX;eYLH zw@9Yo9qJLVP3z)fX>B)M4TF1wn%73We=_^Fw#lz?^E-##FHh0dy4DzvYyM2P zTUzUEh1>2lECe5_K5~~KolaZ|V3a01=8?^-f(fn_oWCc{<2*9 zj9cW>PyKqFYE+rzroNZ>6z0{c-Ab-1&=Jtl;6-hG%k1ELylHTFP*HDN%n~ z?=|XMw%yOSOS-%t(3UnylDV%#2sHPJj7B)uneW(qd;UaK_V`HaqE0BKck zfsCWd)N{3l=dRMid*R#LKuI<^%-wS4r-%lrt&kg=zZbLx4 zp+G=ss5yECzg;XOzhqk|nwQZ}PwI>AL@+cL74fve1n*2=otih(A}Vf>+~Y!ShU%C- zRNPN0U!dmbX1k$-sX3SOuLF{&)1=>^OtK^*eG`}>1AGfEA||p{@9J?p&u(TGD^F%o z`8ss8{KX%2jz28#t&A{)e6;-lE(KA#>V zOSx4G)cy?oQ@14My&v;2ag)066rJ>lS9b%+ogdc!H=EF?K$he^lgmNKXi{TOK znhkTtUs7d-UNS9lT^+T}Z}+mTh~l~J-R{(QWoWu=PdyyM5vA{D%3oSYZk+G9 zocJ{Em^ap2G*($Q$~t+Mv^ni0-_NXKt>jw_>xHJ!su0VCi{{(!{`#7@zss1uOEr1t zeD+RpbyK=+c(P5~M8%xJh zg1*NaZi;SM<>}^BT}JElD#9%V^klt{ zc3*|M+eP1WbdUFQsD(OL29;Njs?}BVhTgeW${qn&MeOuGk>5x&LmhvgPb{&`lopBm zoIFx$qA10oG^@&4s>OG!>4m2!v`n#P)N8&#ylzGAXx5)xP03rBpeo7mwm6- zi0=Jy>MJgCe4Ti$tSk+ckcD0vtmy@IpCiX#h7?Trjaz=<^Sm{wM1S0?=^9#H^NugL z;Iv)reTL**e!@*HQwzZg5@!Yy)k!nO1fRkvenQIYlVn!&R1O}C7xSnQn%?ABarZvh zEViypX)t+ntsACoTrZB%x^*6LeM8WVo~bhcDc|I2^+pv}{ubq(Ct=aNa^d$(opeKt zjg;g;->qy!j5xdtmnT=54LZ0v=-QW;rpF>S8c4N9yu@@ck- zIk|68@Zoz=%7;6T?hlTk8Z7qMtK`>2MqFl-%1_+vjI(z3+gk4Te0)S*{d&KWj%J}u z^_!7?i-Ac)-?cd_{tnsCv+v|n5GvQxhC&}*|H{R_$vF^_B=JfUln|x*<&^~diuXv= zzIa37xfVwqs}Xlp_B~p~`!DqO#()l4YaL-XiwRjY$4QFz3WA} zY8t!*t6rw>ioDty`-)Pisn5$czP_+s>dS;je`IYhA!lUlz<=vg4^r^T7{!E_b`8j` zP9gX9c3p~j{R5^$E%d8Vh+`@Zz8=R0^M`_FmuXjo^PBOQmsY4@|IeSs^MC+*>)Q=a z2)kZSsVT+mB}6c;EsXpmDKnfQt5LIeA)ihujo#IM)AB)qYAzgwMxOf`PhE(onrwW` zc1m&b=1Z26C~6#)E8NtGss3UT@+l>4J$uz-a7^aKi?fm8*)d|%X2YTwRr}F4nU7?- z3BD4|g&&E`O0*ZV=&}f2S28jafZ4wnAFquHFI_KQ&UfQ(`;ZW_z$WlCku(y{|Ds=> zKQ5+FF>7is!`O%(m6vQ1`NGJp>K5PX?s{{Pvfz=fqQ%yhNB7sfN&jbCgU>mPm`r0H zedSY379<)hq)!Rsd_J=#OBeXUSyXYo!|>o*P{NijP4h}9H)-d1J;9UC>`9R(ilymY z!{m6i@OFyV(+RgMeUY#mZd-KL{aJ5~6oL?OkKAUwwfM-phEWbSX?F3=USA|l`;o(? zq(}98u-76lsHYPEmdJ3@9jd@PBmVKIFDB$stSf8UU{9CJH zrMQFsgMRC%@mHwU$EG)i?m1D47iTb%ykD-gX|{SAbL3YQ?q$_oBBd^BE|osr@T1+m zYWwBA_s4ujWGPO!7+eMAgxe@<2DFyHJF8+^$-LX5_DQ!L^2ULMWsuL-|)rhvy5(7_N*$&m}y3UCz`M%*aXJ zcEO4J_pdYPs zX!r|ZzONUVQt>*3E%~f>+djpjvabLu0?MlbHgp%&JD$}#EnF-9#7^ljJsS+Y`{`ZF zvk}d+oT(3}3SMo6P-m-V+W^{HH?qh>`Z(Lnu3SV!v5Y9`O5>&$qIP8&rx`eYV0bsW zAf7lb$lc{t#d?ER20QmARkY`qU_Qrz+j2G#(}z&5dkfb=WIIh2Kw5$GgOT9X`g+A6 zW#!kA#$SK=jZ?kJ>tH2MQoFFKkDiac5E>Z>GvpFSk#oMZdwGTPT9*&I_}78Z(d_MS z?3zPuoE#OV2iTv9dry3N@1=Kc>Q74b*Z>)TPL2otF(SK=ipx%aha1xI+k+d6^7*}_w^o=U@Zebhji&=%cy zyi-R9!KiI`UGP<{XOl~V_&g=|7YUD!6uNdCM_#ksecugF3}q#ws>LHXDDv1`B28Bk zooO#JobB~wUwRn_ld}CRc&Ns2|6W(Lkz4$!;IdYKYHfRv4SCDJVgm%3yV7nV^?bH9 z_SBCx^FoXbWx@fOB0>0PlyxOR_Zmvf zOkz-qOzvzIa5k;c72>^Z%b<|nj+I(e2t>~rA}x3i~QmkG${``ip!fzYPQkqYs-TDxUg;W{6mx7 z*`R*)u(-Pp>?zO|dG@oj^~{1>kyU;XsCfA|U8Jo(VJue>9q$|4Lm8KO7fb!hZ!# zDCmVg6u5#1grR{Ze5g1nI(%*ld4H`XUYrZF5qSZ$h5Ij9g0~>0AG4T!sanc)m!9m} z`oRZ~@%clismjj|FYi869@r8FCM$>`pV$tS1PaPi#`$nlmZP>zq@vI5J<`-8s@~_m z$<2nf* zQ;`dZ#Gt10+X4vi*OLYp2>92Nb{(yX0WK~aeE{S|87F8y3u zm;Q=8GSJSk*q&HLBLX5v|2YIDGxzqIy~?nt%**p@(eK@lW@cO&Hp`!Tl9w|pJ&+bV zntVWW{k~F$RWl&p#ImP2`&vrD%)Ij;*&WMBWU0urc5^qwp8{UgHoHp_v~mTK3&EUU z6nAcnIH6>}Og2AE?WkIgpJ)I87C&A0+-=T~Etqe8NU`4UIcGU%+Izk(yUJ#3&UgK) z!TsV+?-M*#&yE+2lfkB0!O&lahP1Qaxu-UCHTwl$<^^%T67-0U!b`kj9Y=(JO(-n7 zRGDuKYA@N9T4XARHFyV(aWCe$R z;;pt|43~6N!A+~ytHeS}(Frev{3&!ZL-6U@dk8|f@MD&N49GG>1lftmOu5;M8Ix!E z(SJR@;o#gkwf_9Y?wxZ??u&&Lt$o-bXpA+M(V2YPM%noI9M`|Ghm_3P*~;C+&Dz|l zj!2z8i4X+&T{+H0Efg42x8ENwXD+h;JV0RxBnayaiDAd3_uv2wSFp~IaI7;R5OL`Y za|;Ij&l!4Wk&^%k0K(DKO9Ck5WrILC68_hVe+LMJ!(re{27z!8^s-|h0u05nhrr~0 z=`H{X0RMFt^|uoM5(K!MF#rjMVr7DaA+XO7AgrJ<8AC2P1_Hpae+7-g-~t4|5m-@U zqQN=?f`OPu`~CS}>3~2O=yExMKqUBbfItukf_(;sqb@@O1i^pz@V}jaf&Yc;ACAFb zB$i_^6oQp87>n^u9x2f zfw0g5gW*^xfFWS`WqUv*rpNwkPx+4qU^s(K5G3ppTmS$F!|KZ!g8*POaP0~Tg2es? zhGi85fFmz63>~ z2_|hY1lutf0siNEzk3;T6M!W_APh+^(-Q=NBCs-s0-%2i^QSH#C>Vl$2Ek$z2ns`i zFN-)vA(5Av1A-!d5ANTeztMt$v5|qnfR_~r(@|K10t5#_|4QT^S_*PpE%K4?;iy{|b=xKZh{LzYc$&Apz)`jVmY+5DN<=7>0!z z5`m2m5(&7B85jWk??C$-M=+*0F53eG5P#AApY;6QKU5*UcU zmID~*Up@67dV*otN{N}7@XL*c!39g+!7vQrFBJwO8dy^pLs%@?07HQ&FqRW2wm3mC znOtTL6bOS|W&{*~smY~aFp7`W0+{ysuWJ8;o|qwwVgGLu{iAj;1a>{)*ed}>WU(Y4 z42NRvF__f;Eq%Z72gCoLIw1i6l9At0fe}Eg?F$1w_%gM@2ng0L0>&&nmz4#K1VB-j zAp~RCf4Ql^NH}(lF-?EjF$B}MmzfU1sO05l#o42&ht z5EvAzjUh0M+Fy1I$4tzBpP>MlME<2nznKn!L$F2^1dhEmK;TI1G9j?r6M}#N|785X zgb{EIS^uMdfA>2^SpP8dZ_;4e1nV;L|A7_rv*5CRK#=hNN2D-6bN--!`txfGGdM2G zC=|2gU4{^|(_(2q6o6S8{}}vZ4~D%VKr#F4WiFtgQ277I@CO|AYVkpvG+CSW^rN#a2Bi6pZZzilOG^py9xOc8-UexxJ&c g8=15;m*gFLKWogN73iEjF;Xa!jGtfemJ-?j0ofv*BLDyZ diff --git a/paper/pnas/main.md b/paper/pnas/main.md index 5255b4d..6b37ef6 100644 --- a/paper/pnas/main.md +++ b/paper/pnas/main.md @@ -25,8 +25,8 @@ systems whose inheritance, diversity, and compatibility must be managed, and tra quantitative framework of the evolution of sex. Its starting point, that training on model output is genetic drift and model collapse its signature, is by now established from several independent directions; I develop the structure that follows from it. In a minimal -inheritance model that is exactly Wright–Fisher, and measurably Wright–Fisher plus estimator bias in -trained networks, I derive and test remedies. Grounding acts as immigration: a real-data fraction +biological model of inheritance that is literally Wright–Fisher, and measurably Wright–Fisher plus +estimator bias in trained networks, I derive and test remedies. Grounding acts as immigration: a real-data fraction far below one retained most equilibrium diversity, with a per-capability observation floor that makes the rarest knowledge expensive under unstratified sampling. Refitting a child to the mean of its parents' output distributions cancels the multi-parent gain to first order in the rare-item @@ -93,7 +93,7 @@ problem at a new scale: *continual learning*, the struggle to acquire new abilit ones (26, 27), transposed from a single network to a population whose members inherit from one another. Population genetics, I will argue, prices these decisions. Table 1 summarises the correspondences on which the argument runs. Fig. 1A maps the experimental programme built on -them: the same abstractions tested at three tiers — an exact population-genetic simulator, +them: the same abstractions tested at three tiers — a biological model in simulation, trained neural networks, and language models — with the sections that follow climbing that ladder. Fig. 1B draws the shift of perspective the whole transfer rests on: the model ecosystem read not as a society in space, contemporaries exchanging messages, but as a society in time, generations @@ -101,13 +101,13 @@ coupled by inheritance, recombination, and immigration. *(FIG:fig1)* -## The minimal model, and where its exactness ends +## The biological model, and where trained learners depart from it Knowledge is modelled as a distribution `p_t` over `K` discrete items (capabilities, facts, modes of behaviour), with a fixed true distribution `p*` whose rare tail carries the knowledge most at risk. One generation is: *draw `n` samples from the parent's distribution, optionally mix in `m` verified -real samples ("grounding", `g = m/(n+m)`), and refit the child*. In this minimal inheritance model the -resampling step *is* the Wright–Fisher process, population genetics' canonical model of neutral +real samples ("grounding", `g = m/(n+m)`), and refit the child*. This construction is the paper's +*biological model*, deliberately minimal, and its resampling step *is* the Wright–Fisher process, population genetics' canonical model of neutral evolution, in which each new generation is a random sample of size `n` from the previous one and every statistical property of drift follows from that one step. Diversity throughout this paper is *heterozygosity*, `H = 1 − Σ p_i²`: the probability that two random draws differ (one minus a @@ -117,7 +117,8 @@ decay `E[H_t] = H_0(1 − 1/n)^t`; the exact immigration–drift equilibrium; th multi-teacher union) to within 0.5%, and these are standing tests in the codebase, not one-off checks. -The boundary of the exactness matters, and I measured it rather than assumed it. Real training adds +Where a real learner departs from this model matters, and I measured that boundary rather than +assumed it. Real training adds approximation, optimisation noise, and inductive bias, and when trained networks are fit against the exact drift null they deviate in *opposite, architecture-specific* directions: a smoothing recurrent network resists collapse (keeping spurious variants alive), while a sharpening image generator @@ -131,32 +132,33 @@ of *Muller's ratchet* (28), the mechanism by which lineages that never recombine the reason non-recombining genomes such as the Y chromosome have shed most of their ancestral genes. Once every copy of a rare capability is gone from all parents and sources, no recombination can rebuild it: each such loss is a click of the ratchet, and remedies must act while copies still -survive somewhere (a consequence-level correspondence; the minimal model lacks the ratchet's +survive somewhere (a consequence-level correspondence; the biological model lacks the ratchet's recurrent-mutation driver). **Table 1.** The dictionary. Each biological term is introduced in the section that develops it; each correspondence is stated with the level of support it currently has -(exact = closed form in the minimal model; empirical = measured in trained systems; hypothesis = +(closed form = derived in the biological model and verified against simulation; empirical = +measured in trained systems; hypothesis = stated with a falsifier, untested or unconfirmed). The full claim-by-claim ledger with assumptions and known limits is SI Appendix, Table S1. | Population genetics | Model populations | Support | |---|---|---| -| Genetic drift in a finite population | Training on finite samples of model output | Exact (minimal model); signs in trained nets; diagnosis conceded to prior work | -| Immigration from a fixed source | Grounding with verified real data | Exact equilibrium; signs in RNN/MLP/VAE/MNIST | +| Genetic drift in a finite population | Training on finite samples of model output | Closed form (biological model); signs in trained nets; diagnosis conceded to prior work | +| Immigration from a fixed source | Grounding with verified real data | Closed-form equilibrium; signs in RNN/MLP/VAE/MNIST | | Muller's ratchet (asexual decay) | Irreversible arm of model collapse | Correspondence, scoped: applies to unrecoverable loss | | Recombination / sexual reproduction | Model merging | Empirical at 0.5B–7B | -| Fisher–Muller effect | Merged specialists exceed every parent | Analytic model; replicated in LLMs | -| Outbreeding depression under epistasis | Merging entangled skills harms offspring | Analytic model (NK landscapes); hypothesis at LLM scale | -| Mating systems / population structure | Who merges with whom (breadth of the parent pool) | Analytic model; hypothesis for real populations | +| Fisher–Muller effect | Merged specialists exceed every parent | Biological model; replicated in LLMs | +| Outbreeding depression under epistasis | Merging entangled skills harms offspring | Biological model (NK landscapes); hypothesis at LLM scale | +| Mating systems / population structure | Who merges with whom (breadth of the parent pool) | Biological model; hypothesis for real populations | | Reproductive isolation (BDM incompatibilities) | Merge failure from functional conflict | Empirical (MLP + LLM tiers, conflict-associated); emergent form not observed | -| Selection on a fitness function | Verifier-anchored selection ("reality that can say no") | Analytic model (complementary with recombination and diversity in the tested society) | +| Selection on a fitness function | Verifier-anchored selection ("reality that can say no") | Biological model (complementary with recombination and diversity in the tested society) | ## Results ### Grounding is immigration: cheap, with a floor -In the minimal model, grounding from a fixed real source is *immigration* into a drifting population +In the biological model, grounding from a fixed real source is *immigration* into a drifting population (29–31). Immigration is what conservation managers prescribe when a fragmented reserve loses diversity, and its striking property there is how little is needed — the field's rule of thumb is that one migrant per generation holds an isolated population's diversity (32). The same economy @@ -178,8 +180,8 @@ still retained across complementary parents* (next section). In trained networks looked, with two deviations, both traced to the estimator bias above: sharp thresholds soften, and support-counting metrics decouple from truth (forward-KL is the operative collapse metric for a smoothing learner). On real images (Fig. 2B), dry self-training collapses a convolutional VAE to one -mode while ~10% grounding holds all thirty (the trained model needs roughly twice the exact-operator -fraction, the measured price of the estimator bias). +mode while ~10% grounding holds all thirty (the trained model needs roughly twice the biological +model's fraction, the measured price of the estimator bias). *(FIG:fig2)* @@ -204,7 +206,7 @@ reduction from averaging can *reduce* extinction relative to a randomly chosen s cancellation is a first-order result about rare items, not a universal impossibility; and the contrasting union operator (keep each item's strongest source, then renormalise, which itself redistributes mass and presupposes a verifier or oracle to identify the strongest source) increases -expected retention with K in all regimes in the minimal model. The practically important +expected retention with K in all regimes in the biological model. The practically important operators, *weight averaging* (a nonlinear network's weight-mean does not compute its parents' output-mean) and *routing among intact specialists* (34) (different storage and inference budgets from a single child), are its empirical cousins, and the measured bridge is a *headroom rule*, stated qualitatively: in language models, @@ -219,7 +221,7 @@ adaptation: in an asexual population, beneficial variants arising in different i compete until all but one lineage is lost, whereas recombination assembles them in one offspring, producing a *genotype* (an individual's combination of variants, one at each *locus*, or position) fitter than any parent. -In the multi-locus model, sexual merging of decorrelated specialists climbs to the global optimum, a +In the multi-locus biological model, sexual merging of decorrelated specialists climbs to the global optimum, a genotype no parent held, while the best single parent and the blended average both plateau below (Fig. 4). In real language models the signature replicates under seed replication: merges of three LoRA (37) specialists beat every parent overall (decisively at 7B: 0.87 vs 0.77), and on the sharper @@ -282,11 +284,11 @@ separate species (*reproductive isolation*) through Bateson–Dobzhansky–Mulle changes harmless on their own genetic background but deleterious in combination — the mechanism behind the mule's sterility and the inviability of many between-species crosses, in which two genomes that each work perfectly cannot run in the same cell. A merged model is exactly the -exposed hybrid. I built the analytic model (Fig. 7A): hybrid fitness tracks the parents while +exposed hybrid. I built the biological model of the process (Fig. 7A): hybrid fitness tracks the parents while compatible, then peels off and crashes below the ancestor; the isolation cliff arrives earlier the denser the incompatibilities; and the incompatibility *count* snowballs quadratically with divergence (42). Note that a super-linear count does not by itself entail a sharp performance cliff without -the count-to-effect-size link, which the analytic model supplies under its assumptions and any neural +the count-to-effect-size link, which the biological model supplies under its assumptions and any neural test must establish separately. In trained networks, the claim must survive a known alternative: merge barriers between independently @@ -382,7 +384,7 @@ conventions on shared circuitry, which is the thing to detect. for forgetting are this framework's operators writ small. Rehearsal and replay of stored data (26, 27) is grounded inheritance within one lineage, and the replay fractions the field settled on empirically, on the order of 1% for instruction tuning (48) and 5% to 25% by distribution-shift strength in -continual pretraining (49), sit where the minimal model's operational threshold lies. +continual pretraining (49), sit where the biological model's operational threshold lies. *Pseudo-rehearsal*, the replay of a network's own generated samples, proposed as a cure in 1995 (50) and revived as generative replay (51), is precisely the ungrounded null studied here: immigration from a drifting source, benign for one hop, compounding over generations, with verifier-filtering (29, 52) @@ -426,7 +428,7 @@ null that bounds it; and the controlled predictive test with its controls. I cla these measurements and experiments; I do not claim that their outcomes validate a uniquely population-genetic mechanism, and one refinement it proposed was not supported. -**Limits and open problems.** The demonstrations are deliberately small: exact where small is a virtue, +**Limits and open problems.** The demonstrations are deliberately small: closed-form where small is a virtue, sign-level and seed-replicated at the language-model tier, on constructed task families with a trivially separable router and one model lineage (Qwen, 0.5B–7B). The composed society has not been built at language-model scale. The predictive test's next bars, in order of value: generalisation to @@ -470,7 +472,7 @@ paper tested. ## Materials and Methods -**Analytic tier.** Pure NumPy/SciPy Wright–Fisher simulator over `K`-item distributions (knowledge as +**Biological-model tier.** Pure NumPy/SciPy Wright–Fisher simulator over `K`-item distributions (knowledge as `p_t`; Zipf-tailed truth `p*`; drift–grounding–refit generations), extended with a learning kernel (smoothing/sharpening refit), multi-locus genotypes on additive and Kauffman NK landscapes, n-parent crossover, and finite-population society loops. All parameters live in per-experiment YAML configs; @@ -480,7 +482,7 @@ equilibrium, closed-form union) to <0.5% and run in CI with 151 further correctn **Neural tier.** Trained-network experiments realise the same abstractions with an exact oracle: histogram/RNN/MLP/VAE generators on a synthetic mode universe (the histogram model reduces the harness -exactly to the analytic tier — the bridge gate), and a convolutional VAE on MNIST with a frozen CNN +exactly to the biological model — the bridge gate), and a convolutional VAE on MNIST with a frozen CNN oracle (98.5% mode accuracy; confusion matrix recorded as the measurement floor). Speciation experiments fork no-BatchNorm MLPs (784–512–512–10) from a shared base, weight-average, and measure linear-mode-connectivity error barriers before and after alignment; alignment composes deterministic diff --git a/paper/pnas/main.pdf b/paper/pnas/main.pdf index efe11ed7a85b7daedaae5c98e8bca776fc03a5aa..24bfb3bb73ae89e4b1de39a8a98794bc0a24acda 100644 GIT binary patch delta 74836 zcmZ^qQ*fYN)TLwFwr$(CZ6_T&{l;d;w(WG1j-7OD+crAMd{r|wHUGt2?>f&q=W5qp z>+GTkoUtgJL~ekaMMwzF&E3_~)Dg~my;PsKp8PlJ^d~*WdcXaHP~ZqW!YWh_qd?}B z3CH@$5Wd492UICgVPQBZYimBgEj9K~!KJv8Ez{j1u&-2aw z%gRQ6&(_x~cEExQ`#OL5kDVT;a!`va}zXUsER$aX;(x zQ=-oos_xh9OAT;QiH?}lXUc^A7U${`8cBO+Az28l3{A&-^*L{qq4KjBeJyr)d}5E{=ryq!tvN6}&CU4_c;apUR*@ziS{&)% zkCl{jju@P_Q>2V)sx_v)?PRy@QAz8K`UWvu=5k!~)^wQYRLs6KPp|%*Mi;}z@~7TQ z8O`$?z4I6xQNy;$)jtkY`c$lDtgca8IFnY(Rwe}wHzJU-TWJd}xRDM<%7!v_Yn2y9 zdcOf{(+^~941DdWQD4FYjAB?kc#Uq^Jq&2IZMf<(G@PZf)_EKDo|>DVi;*gvah|%6 z@z+b6yf55Y^=8A8rFLhLKp(_F{85kU+Sy*=x8BX}leV9SLO$@r&NGsy6 zQS?9RAVTr-7Rr*D@>v9W6KT4SCFZu?OC?mTs&XYQ-;I_VWbO=;#dK+*HfJ_1IYN^# zS`gxk8o+i~2)c31(qmL`l*Mt<(V@Wqi4_;Nf703{5HJuWU03%C96M`m*n$DBap^yTf<|CmPAOx zoxoB``E1tM;AnFjGXS69`EtM8S-=+YkMkTbpD^?339x+dg7Ek?CGD1xqKYIdIUAnuXn;&T^%3dHquGKbywu-iR1PMiuTa6BkGwy2LSe9iOoH*rALT@0IiQ$ZG z_^ta#W;IjDIj%;g zpz;lvrK33!_voG?A=eg!{@1|fa0x!q>4Vc~tZJ`Mf0KBSpDq%2qb2US^N@i>ar%+WwpRP1y_Fw4M*sdcX1klz4#h*SGswo#nuNUX`v)BgwG1*D!`XmvKxX&Ay!kiDYRJp@b$C0rrBBYJ>yOZJ6L$j9Ji-E!LFLe5TVmiZ%JNW$Lew`a zKGcY+N=L3>m7vc|u}Ko?)a0<#)oNO$L}hlTFv|iNt6Q$f2TfV{!~vRu!TqSoX5xXG zm&0^^;2B@X>p))k&_!bV0UUt%<{5SjwU@Al8`6MEUtlOM#htU61u+Iwu!fp1MtA*U zpu_8sk&Yie7(VL5zaCIYGkr+3)d#{VnjIX9+f8((F-*ek zAF}ZneVHqnU?uFX&_$WIifWCV8O%Yh2N4Z%9zj=yYLcOqA=+}Toi35fb0d2}RlC0^^m}Y5~ zIcJ;_(aQESZR;w2eD2|e9^7pN6lSICb`x0Glm|YAu?C$cb+}bPU+RA&|lLX0W3aQ$9^@VyJ`zW3{5;` z$5=l4HYXLvHiJl;x)l|rhuO>({%Gyqi9?Lb%pqt17jshSqfyli9?w6qnQ#!$9})4) zs;zM1`xv$kLzf&I$a-Au<6C*T%Rq!qrqi?(8p!(sQ$Ba1P(rmbYV8S_a zEESG0+WP#ximzU)tO|q5ydG81h{oZ5*F8P)n%$neJ-BCohIYpuA@_& zpKXJTF!9gc|1zw7Xr&?IGEkLUhAfp?$va?Vkt^bJY_^PM;owRyKOC9*3L*Cp-skeX z1yxiLELpRcETwP1z*;3a##0&8iI8(N^h=_TeGS+#_1Hyf2Jcg^L7qj>!6p1@(}lu3 zh3PRM4H6igq1=TO^D*{uXjH@t&)>Z|5ikSk7%wi*_$RgA`HR%m;;fTJ$}D4ULc((H zSP~H?WYb$DN3OjEExyfYlsl*LZ{ZA`GHFFMIiBIEu!!Wp-U^aRKGa89q^3D(=*)8i zM;t)#v`WVQS0O8k+}-smMK|AwZ*S!Z>V*d!)rCpBL3_q>oDy{v26X|LS5v=Fcl75n z3?D8L(!R}g>Fy%TEG8jGaDk(#Z33ylknF)iO&`HlxvwE1TkjNb}^@ zqQYWCCRCpMH=`h>pDzySFY*2LWXuKGjCa8Fbf7l0zu@-tjACsEMo*C|W&nB_yb$rP zaue^cDKhI7W*R4K_Jr>V5kmd4--KnmiS-tiXN3o_sAYa?jcPKd`PUgIydO2s6_Ayg z*mY%#$=thDXfqtLGlha{(j6i6#8Dh`RobkeI-ATb7|mFR^g~W+dzcH^H*t7aBLYUo z2BTPV`b;N)8e|Oxt;;0)B^TsO#gHqajg&2IsLBmm%LaS1&P#I&kjpl{bGd=5f$lH3 zdE@j|Zr-T0P@C~@qKp03kG)SFJu~dfoE48KBSObvRIWf@m9B({+-~?uZXq5#=aha6 za)k>p&w1Xon)YBL??`p>U7No@@PMe7>%%*1$=pw;g^;9-Zt&cbIfCr+SXrDqY z_=9M5G4FkM7Lt1<;=*3d1m@9B!-yT84_uf}oCE8yq-er_q&Sy%ttRl=5A& zykwl-QYpPKFra!6g;yxM;Dx7|2vQ4p-ut*f#(C9M6@F!AIUOPf`SdvOOZj&{fEbov z7-D-uO{G*Zb z*|jxi;vYuHXA00QwuDMdMJV#Nd`u@u=?r}ytI`@FiMu2mgxji}Na#N$S6NFUQr~Q) zF`kMg=Fs{Chv&|l09)6BZ9Nxdh)-|-7-$~VeCJEQl(>j_A9?S2 z@~_XI%+nAoQ%9ejYFj3w{0L{xBW<8)0Gp{(85W z59qm7H!da+CjgEk5M98xU{CkjckcO@b!^o0MYFK;%xAy(_cx&t)QZtF()ZV0Lcrte zoD};~3Y(PIkyVT$XvDW|>}PGwJ$&@O(?iEx#&jLbI*NVZbH$r-IK`O-=C)>jWY~Fq-EgCdpPaoFM1LGIU&Bmtg|(ERCQdmW&8{b zD<71+3c$a^qkh##SVB~5eGxk7*hBK4O+Rq1hb*tF72nMNLi)J(dR@@4ul0YYa}sAt zERQ^ihJWC2#fJrED5}+=-w(7dOx8VNR`aBJg&`aoACylfaw5_E@(uRBZFh_}BRmqQ zkWjTZD_(P~p^L`WlG2BLp{)DXKS=oreiZfQ8(qp@!*6vl*`5pq)vo?-zGs3Rr))jV z6*RnTA%-^XvH9XM@(iAf;h7iR-~GpP%48j&EZTh0W5EX;><+lI-BJiAf`=b2k>(-` ztx1KUB;tpN3(ax}ZTa8isA{6dsOI7?FN^sFR z4mDA8Ydv};qkb$^Jblo)4*9*;A?gr7+87ccYHA^xAC+TXKxV^1ft9xM{w#Rocr`-3 zV*}4P8|n=dR+wm5P-GV$A2b#%aER zJqc_~(@4I*F&7RpU3r%~hxPHDluF7e5Y8dh3v!X0Q*7#2=EDs09@JSCyZc3E4aCK+ zFP5vYh=^{>VjYhNH@dC#f$}wPk&Qz!i6EUG1Udo|nbwFmVF8a~uDN;@UNw0b><_4% z&FHy}pA|oYqqh+q$yZMpM-Cqp+8p{J_TC;#1wUf@mHcC`iaHy95B8~ilkfz z^H>2xnT9U_{n80maIl2KB{<_4PxhKwp%`Tt={5)2pnrkE)`+?F`c9?QfjFoih=NH# zCuQ?Ra3->hB|jH}nrWdOe4W#Jeu`QAv@N?s7y>>YjknyqrBpk$d$}(bM+V}`0A2V@ zb3U8TsXJPLizU~X$Gp1vD_zUYYG)4J+&z$SyM>PaYZC|BgEn^aj8wqSM2IIXF}zc> z7$3>NrOsxYwG;X1#e~V)+_HSOA2v)`u2JzX^!OK~6sTcyrRf$`fLe=bg|9zh6*bIF z!}K<3OM20{E~Fq8;ciRrJiRDf@sk+T!ZWfy|6WXy`s!KfKSaBDyBr`$*fcE4MhGy) zx?Hy09jeVVWEOa3A}_3(gR(aXJ^F`{l-cj`*YR(cvSohWiv(!`zn9UL`AxH%$(YaM zPsiGoKOL6X2P)J(($Ah=kCH<;H;{QzR3xGZ9(xsZh2>b*s2jA6G-XZ-yPqRuE7T;( zbK67#aibWkJCZQ@Q29P+IfV_TcsGD(Y*9#{ypBt@(^9$0%hPom*HNH}7~Q~qW;bIF zoSREk*;Ic7t=J5ARYA6R+uuXlp`oupBd8An=kI6y=p7^%n|DiGxCV z>kwjamGSl^z9jnp?iX^O=g-~B!OnMf?<~jP!1i>#z`zefDZ5z8NT@uTM&F~dhjE{% z3O&D|ffTNla+8}jW~vV1xa6EFyA%mIxIzQ+v4at7F&D}LZv>JDzfvAHenS^4dzxbT-wg~G zQY@hi>U7P28=gvxG}eS}S}g;1X8k^GIUIX>z0W!|0 zP5$gfwoFdH^L{;9bxp90ht2JEb>?_ezQ9md@Y;Hqag0KbWG|qGt*rvSxjr-%(442l z-COan@J?RG1^(C+;-$gGxdja5inm-9XSJc7-9CFbq93Vyq1@WKMpy(|;f3^MEwBi_ zicBQb9-VrDdC7Q!jQt|eF~3a5f{Zk16*y7BwNqJ=gn9mv`}f+*D3w2~IKn2@m$Ij1?C!;gNaec1uN4EM<{w~x3`4`{Ct_;wrf zdugO^vcT3{GbJ#YvutOwA3~HZ0X=I54uzT$6Ib4_lVlvk7EVAU+5v}}3E zuU7B^MAA54@|5>b81bF>0W`79#?YRsYG2Gc6*39#7z%HcMnte;r7}vB?e1}3j5-6f zSM(+Jq{%R@5T*F!l`6Z3h4Sp31P+5`iXb3z^=#Rw|3PER}3`l1IDGx4B`57O{u5dpb7is1V#0< z3OQaA`iZ+6=ujm6fiVjCe8}~wCNRg8YC)~AX$MpsS`;T#Fb+Xid7$=0Fc4Gd>NC8-Z03C!6nO`>SBs ze66LLta?Yre)ON5{Ru4WtsaUyQmxX4dR|gl15bk((`$uzTyikm8RuM2i=JD~xm(l+S}6}M8H@7!VW%QDIB0R>f9-9LS+nggAq5D;f1XimvHzhuH$ z3V=XeS8W=C^GrD?}x*D(Wm+<&S*{f!miXPT7<6hxu$h2h#XqIPX=q8(jQv zhDm;JB{CMW_C5x%cUAeS9|EToJa`Connz?=O=Du|$9jsoTJkWS zC6E{_krrQ*u4&m_8oU!B1^UiaFvIeD#Uj~U6~wB(I?FM@_*m+wVJ2cKygPeSkguzz zrR@p9rQu03>3ubUMXESq&QkHT!&$4ir{bTg`H1s?fZ_cwRniv?pZ52LpqdckkOjzO zdLKma?g!TChyaXlmK8m9rz9ZP)ZzA-0SFnV<*V88(RG2xRn7DXbtDf(Kj7M$?S-#P zF!gh3f(X1)QuTxeaZo)ebPM$+@9g>)9NWyzneo(eElRbs&$WffZ%{{Mwc&NBxszSR zXIq>|977X5FL^EWz+?8uoOZxC%?pGN)A84JZJZlpy5dbr-fBb2gL^az$ZKoK7T_P1 z6I0))69nXL;S**akG?xlYTpWy3BGfZ{pW0!zw8@RHeUrKu0YKHQGcSbgH$__+B)QZ zJp=LnS$FI=_~H*q&V0mAyB@&eLHt^n8CxT3fQ$Iuhv zB725YskJDm;?dkMrRcp>Iza$R zWhJ$9_XZ=Ul6@i9HpoSPnx<>cvgjFzl57)KQO4h&`{=%bDS zn%y9bvTaq}KSNvy<_g{=rAl)7{lg`2pU%sxeR?&@G+@gk=ziN1XT$51`^gmc91g2Q zh3~VY{rW1L%A(OY+__VtP{{3BfMvWzr;14E<=8Iepq12A(P1<2&x{V|mhC1jFBIAFz@p0>sF4nOD;M}+ zR}V*cvoQb>AK+~h_m4`HvhOqpPTt$_AT4Y&;K;ZI?(PW{yP?wvfU5gIy5J+r-)C%3 zn00PVsLArP;o*s@+HlcOkriv0-naLj#4ASgoTnzXNJ70VGYLvM~Vv9Fa`uHe7dR_B6M474O%EI%N)jLlu^M2Qr!DCa1ndcySUg6BgSYP9{Z4VJx@OHb z+=#hV3?UQxe)Oh0PWJ0a0+D0_fYyg>U%ki_j5S@r?y8vUMu?q=u+-ykjTM@REvlMmZSj?@nPP z(2#Pi>LpR&AjVo)g)}JR;q+`U(NvI^*ypN@t6@;BJIG))TE%{8?H9c}+=SY6Nkiq? z6m!{>zd9jDqFlLWT+v*f0jNR*5)>-D9|-Jkrtp!o8)(Zc@i0*O)xL$|BOzA*+-0(8 z6b~{{eMqP(PJTiT=t)8<$c^jvCX?5;PN1UD2%Txb3aw}Nj>x9RD~7QN#HEl6|GlSj zA3|p4usT??ymET-?NF)706U$kST{0}G#^mdFr3iJK&80$h!-Qh1-y^b=!!1I$f*R( zRs9vSDa@(DRo^)#%|XBuTG~dk+XO0%$-xq-iIDLUyU^r@GJ{O7UBwaI$;8%7Sad6O zYRP2W+lynP*IK`khZGIFZs zC5w?9Dir;0?F|*V1SW#HsY^j~q>S|a8}Fw7@L)M-Xc2kS$jln*v@P}7+3c(9Yr}3p zauso(LTrZaOCcCmB*^E=7H#SIp?7z5MHbOBc(>Sm{{zoOf3VtA<>(m?%0WMgU&!fC zhD^oahWbE!vJ26wUqIy`fXGpFEPLh*ZXbyTG7pn z2nnb*Q-WNysW!+rPz=|IudCLodh6~$PEgHS>_8m9|EW}+n+ejCg^g3)+MK*h9)g3w z#it9DA$30vzLYa#G1GC@fLq#?4J>y~pG`4udiG_kE)ARdWWs7(X})>0 zrMSQ{M$xlWs3#|906*QusW$o~D5R@-+h!2F9|0tf>!-2>ENm}UB8~$5f1&Gs$N)8b zJ3s&0cnKumugUG{vyy#NF@ynZoU+M@pJtE%@$BR0EgwdeEr)>THcMQJPnxrmpYcCP zWzwbh4VZouzyVeF(!dP@e6GIpMADCWn7U`3$b}W_%%~e%^)TTaYPyZAiP5PBA0;m? zpW%~Eo$mlsy)$BUbf~8(oC?{xhz*r4+VBN;sT1FGErRE zN-0{UAgE%;X3TEv98C8F)2@}LuTc^RF%It!TTcCDS?;e1_T;!sN`1ZYsI;FrC$)&r z<1n#j&tE8zGv|l86NB5gE$jv+c9cSs?`TrsB3@P*L6#zB!-%EzcRj`L;Unb!?hIl; zVt@ezdMr`!7$i3y{fMW4yIN)sK4K5A#y$B`MEE>tHP{N`(A~KPwb4oA6?u= zCSV;Vjqb}05@^oRSH^8E;Xh*l1A4NZ&O_}aW@xS2pg&$UG#eX+5?e1Ab18$nU)o|5 z!f}Oe|Aoc-%B8pK9d|Mi^t;*uU zSXoYL-v6ZZkQ{nk(j0~W;KQ9so#4v}3y&ovcD0OZ7oTE+vyOP{#WJ=>H;2MfaH>UpiIe96w1Kz<$v!`KuBjp+=;KPkQ%arLR#xpDDYAOghn zNIZdJh1Eu}E|3?Taf<>=3ZyOf3gbfi**R(G zgAr00L4Xlbdy+V1erNltWO3^T!QZy?QKVAFO|;|g`4b8iixJ*Zy0?PSPPW#*X%Dj8 zVdx=rgtNJ2vUiMx64oo_Od~RCti!|#xk71P#fnNu?MA$WEc1dPSq}UAa%b9f?+ir_ zc>Q{H;YIWO=fYqBt(i4dx`@hfex-|vkhk@GsyH@T4N*sqQ=wim8P{99NH(T_3J z2Vafseo0XWc^bmVs{XkCf{W-}Ua(6I+3jiEbULCoY7AmP)bk^YAW=QD_-Ch@IfFeJ z!G&EuQy&o;%(nG6cFg=x{jVyjKL$Y(4{^E~?&@<|;!B_w$aB8GsC5Z`CNp6W^{l;5 z5uU5YnS4;UalosKh+cB$xkO&(H-jlhurK4M{-Y4X$XEMFWUwJ*SAG%NtAjua zo+Zo6G4q!JFffK$&Z^vBb6gExF(D8HaVb%E$zV+vKrKbP%JgQe!}PaiS~R|b?v_E4 zXBQ!R#YvX3D*ed>5=T66tE#>jnkbBCZF8b48>#2YRk7N%N(Es)7v&3$7nO17#9tOzCs#Z+l+a8y6!4g!Yk`Mq;FW6rA({s$a{u2p_q$%BJ#2ACi0=|2X|( z#8)k`CV3>_b(La9{zBg~mK(SVCo8As5}LMB54 zCGXkd*^yjIor!y_1VUKoP)8&}Wm98Ao^!;k6h{aeco%zHUMEv~=B*iA3SW3Ys~pj| zP4RDFcc7-cBJ5{C3|1cFfT}~?xhXP+emDo=`Zdqa3xNT{b0faH2A!;qbB0lr?DHxR zAB)n3SM(t=h2Ff_Nh+~7<@g8G)bug=a#mfO1%h=`g=jJFr}Wn;S&rV6xbbDIz@n@3 zJw(Y+^;fjqa@*LrxK6pQ%dewuWd+FM34JC&5JEULcgB+lJ{<*H|D23F2|gVc+Db;r z%+q^{KHHGxNkO{taK+(#M9;<9iDlt5y*hd191Fwjjua_us@U-A*}C0f`Tb|^lJ#kx z;pc9F;cPZ3rSx%%mbhX^-|tDWmgTcR8~YTukO$f0k<1#4yPPp=B3uUAxRiasFya_U z%SJS0Dk(sG?_%i>R+>QzEhg`LO@}+mvn1t(ikOR?K{m%L*QjINtg+jztMP*f3Trf0 z+x`4gMx)ig}O6rkd5Q#g@ae=G>e!81_G?6oWK|gf4*sVL!SI%Q+k0Gwr&T4^u{e6bpR+gC=aU*7$0F;{nul1yMF=nsRKVIx~UZvgxq2vGRp zG)r?9Efr~3%2BHjyxc05389KqArXaXYi{p|4ji48OORM`JUS`x$6MQ#C&m|iFl6iO zAhKaqp!JRhYF8{BR8$AIG?4)%6E2OnvpVoL1EXn&AN9lZ!v*cwZ75BnO7`+|)pz43 z)f19J3tNJ9c$DzUYGR*^@^x-Mgq({e#wXM z-NS#+VoSqL1TosDpt)M0JbIMTzZRDpXIW6dF*z{P-+0{UWGP{Or=kPy+qz5+tA?c< z{E@XxR8wc@(sk-M40j(XNW3T6G%DA z9t3875KTSEW71qXbZh)zW^SaZ?Gfy!ep>p|aBeVBu}ifm%7v`JJ?sl}ICoWv!0h_m zmLZs3sFkSa-E)@JA|eYcTO%sHn=r{eygT6k9N2n!r;jHjGlG)tDfi^4$q~AGXhr6? zU|$SmB=#1}7PYdBk2J?F!TZ33yNduL$pTE$9gQ@AgIe>Y-KuZMj)B;)|HrC5mEblz< zX!G2-vD#t)(Ezh69UZf0?YsB*jXYc`Y_G`+M=e6m%%jHiZ1!PSx8x(wR7<2Ya`V{F z!Odf4n^{aJ05#X6atiZThP8Oyd}eAyyL>_gf&=_$<;I~EdEBi3&s2zMfBc2NiqH9j zl;S_}O86nw52%2Q=t>d75LI|Y{NA2f;e@2>XCIj!TPgKQ9TE*^7|^#9!Pnx~axR9E zF{ze86t_3x)i$NVoW3uPbXd3Pjq<;k-9lD0DRV+YWcJ-X9|>;1QXZYavYndfSl-U$ z%P5lD8OQ$fsc8(8H#PnD;OqsJFH1EDgU`Mm3D7YrwhjQuv(tZGI)iWJ^K*>pNgIL8 zb%vGMQt7;oU$Ya{J4ltR`t;*lT{fzprle~+r!r}bLiG4&#_gn1$Mx$5%OEWOF3K?v zi_x{uYp8HK&OKhU`Zdsp_i6Rkm0FZD;CGl_^vSa~RgI^InKNQXR}hwOyc}X!{hdWD+BuBwL6u~ z<_2&;M22 zSsI&Jt|dT|Z&}fkg%U`UV_2<{x!KX0!Pz)Kle^i`l4aRlo69)<;Dd6dC+&cfq@T+} zzyaL)$Zai1;Zj&lZeu%M#CC9)I)hW!b`&?q5zmcJ2SxWQ zJZt0(b~t}xSd`OkM4z1T0<=;l;yh(zz5K4l(#-Gf_?VrYK3Oaki|@G+rscb7gVa-U8e^Qary<`ht&G@6idnzU%>p})vECC@P`b=fE~DR9#mWVvM1O7N8J&*dnnVDy8c(X2Vv zPTiZZlo@^W?%gT->aq+xe38*1eA14s&xoRAzXM@|F`dCkSXgkr6vGJ=JHt@@$e*Py z5jXH%ImWjm|DgJbKd+YTIH~m7bT|!~FthFx+(O)gs9Z=Rx052O7tp=`1Zm0bw@WuI zg}`WrS1a}fIJo}Dn#Psj#xKJ5xa>b#Puh?2O~rcETiwZ?N2GENnx4sv>spRPmjX6c zhe?-n-Ze3VfUb}wU@iI_M$+%plpx)vp{qU48Mp8k9t4a8B>3k0dfaP3*FB6q zcb=_w_wuoPO}bB%5U_p%cO~C$z(T@p(}99*HKc$HH7>gz9u%@wuv_svwl*O?1w!fF zr!uiq@7fWgm%_a$f1&$KDPhTM@g9rn^`lAYGFnJIr~@?Ro;1SoG~xiOFdzz|=Ty@I zJBygbx@PPSbGq1R1=HQ<7P)H=k`Uq+PW{7=B-yE@ zqZxxI?xIM1))H<7);X8qoVKVmat(e454ltMozIPj3PF1o^MyjAxS1SfNjTQZ=hDdF z3Er)xTbNbazj@>50(*ps6TqVO--?2r@ zo9dU=8VJEjZ8e~|7Jx0yBxd~|!Yyd8!-rFaVc0-0gEbQ4=5B{&4uV>bGPqGohkvlo(jZDJDp}@HE{EmZepmsvS=xJ$k>Uwi#(e6+ zU_e@=FN_n3Ce>`vaGNDRw^gx@E@Ec94Q$bII6XYAUZF&xVD+;QR^>EdcGlC z9dGh!pfC6*qS@2eSAKGAh5I zO=XHE1Y2gqLJ7iDmj?{4YSSg&A5Fn|s}63QC!05d9& z<%=1k-GA9i`3D;!L7L4XgE0m}Ajv&|1w{rWYBEKF(>;&ifmIm;6^;v`2+l5(@?ROS zmln)-KB6D9|s$%K)@Gjj*mo@Mvc zW zQLa(yl7f;Y7NBbM`1}#N(i;!B0X`si!iFQAYXLoCD%<+ z@oghY@4$x5{H_2PaMUXlo_QqD0}WMWR9eaIo?=&Gb||#^4Aw-NcoT{i@-`F}HYhEq zNLCi_Qfw_&c+*C%=-41hdkX}HFzpeseLhzbFtQPm-7y@P$YqgB*xEESqa973`rkYD zZ%={40ikBL#D%I__MmQy=8;e&8Th$y_&haOB5`omwTeL4=V>E_1=$5?Szx?^xhZi_ zR7=9H#HKV0){49_P8$+@%m+`z1qR;?*E2?=+Q2zJ`g4Y}YL(f5*03MMJ8y z4mHVu>20t=jk;Ei`DX}BL@6B^L0TU@&g6M1GlF<%jEO=A#Q#MkP#GkTelHAtgxKN% zkYr`H7`do;mJT_VVEVSv!@NbbP|V&V=O&aeT7;2hIwWLSW_!46U{PzS01Cn!M=+}* z`sqNwy>jFNyiS01oGao6W}RtSlwyw310(wGm?b`Pco z9*K?Oo<})P|GyzuzC-XbT;xU~X8&EyMc5ox6opxW<~R(3Zvn=(G&K#70Z5HR)Nt>> ze(fZsYw;#{BMoX9iJXJqfdw-W5I?CJU_%;^8R2|X=l$2Xgh?#FXNA3XJV};5KFC1? zvz;%X9E0N5gF}10w*|cy|9rDvWS$ZS;NE8w8G7cwSoi*hPU|oJz7+CZ3_w#WOv9jA zB!ah}_#uPQ0`EU6+UW``R^VgUEpdPB8tt;0_A6<~JzCFziRfEV&ahS7h6&Gza1tvL*%Vs`C2@DnC5bUI!@;OcU@UA{mLD z4&+G*M>=F99z#b~i3c5qn5A=nNOuyYlcHFKTBcog8-#-I26q5DOi*>u8QL&nfSwh3 zL1!U5M;@5-W^7YuF?J`42ur-gT~Zd$RY^RRIrO}aCx7Oe`K`5uTR0T)zSoZM1f#zP ze&6Pej}t+#TQC)RJO@JfQX|r3ao7#k-z*QXcojGn|4iP4LDM!Z@^S+F>$&@eV?nX; z1|4i`n8AnSs7wJHFe4v4=eW^#3tvf5_<2^%0a4-#I57kY?B;dujncpKdDLv++97%7 zMkx7zsBAis{_fTjMR?^j_bp86OB1Eymi541g0h9}FBZzt0CFq!H4U(?L=+HHlI z!7NSA*l_^pG1mkUW&YAulb@y5;CFxAZJB||F3JNQYViOr7xIBV`ft=kaEO-BrMdh7 z;tm)M`bZ3<|Ni~O2>YIm2183XxC6(-9zqra$^$C1)a@26hxlT3la;8Ce=lyMXVDT_ zqS3Ht&vKHrXvzg-eNu03t5iseGHq8>60RW=vOoc=#%Nx0?C}B9x7Ihf-Y z4WfL<&hv#?wu(yBiZch%KvWRa5a59!9`rCySP4uCY`^5G=n>hJHi!la6&=&B;sECf|0xU~?ab*MLi2;u*5m+kL%6tmYu#oZr8w}siM~kk2V0ys zU=&zw$3)kq!42JJW)wC+w4v`M;kO_a$t$LSI1iIa{C3DRMaI_E`_6THNi>rTiVon6 zojZ{j_ERGu)L{A16roQ!qeg}1U=CpRjTBH?13&^}>fTU5Had%Ng6k@&ok0K2b| z4ucB(GB*Rak%(9b98S4b2|FHw%M6S7b~9PM0E{{7Y%_E!LxM4+6Zb^KN^NQw_tem) zY<0l;d75rv$MbcoXEPYe;8Z~e*xC_yr4fjgmiQJ@k=5p>?TUR(`zh!VB~UIhZkyDe z{PNf`jkp5|S8&JeZI`Dyn#=|+qjO`XL!eCC{vFJhcgP}}_1e$l$^?l6(Q1hTfjEs|Mu@?69pHHIQ*jevQKb+s)oxHbudZ^~c@tke2eTYUkrI6Rl7!H6aN z>{K32ZGG~8xzpvVsVR4ePFNV#UW;vRa}C`z`2|8f%z#4dCY6A^AT7s$M2C}BkxZv+ zEKt0X-Y*GV0}GX#Y?c|{Vlo^;i=GZuqdka#9rNd$8>3(%cQ;1c2S_2=?qkEz#zu$~ zXE8UyF>7n9)kMPRr*W^`!fy7nyyA;yZ$-%M5|z+}%B52oaD8OeJ`^{$rYnBg%(jwB zXn-}VTl_d`VT}gii)c8%9U^A-4D?jvBgI|@T@Usc3gsSgdho0dQH&ee3N+?OY%(PTO4;(>|KZ?%>77J_8QzO7B*=7` zS{MHj6o?NV4bU6-i3>&z8?u%)pN-Ny#EHuaYigAIBtGKGijT1TCIM!2xQjJ3%bOdF z%O^VzmkO)=tYfbs4kK0?t+Do)@1tN5&H8VGj}Nb-a=ZkxTtn0o`i zqM^pkM5Ppv6tA=w3t}n;Da7lC*F~!&W5B1prIHcB3`Pq^Hjrb~Xp$eRm9?WhBCY!O z+d|u;HBo8&bdrFc!$wkoo;!+_xQvWpmvasn%aIo!TlfG+$uIq10E0k$zcFOaM>7En zh9;Q`oxfK+V8v_LcbI8^n_BQ1)x1;RR%2YWW9l@bNIXAfMz4eaBzB$izz3V@1|Xnd zw+6fd8bsr>e+=(v*kd|aL6%Ft zytC5P@9E(k31m=?8RySAg?Iec9)9ra0tUd91?*2hY<{7(e(1M{z$_ z@Fe^o1_#K7J-abx{^8a9g&GyG4D5jCY5SEjeZe#nupd=o{^42pflWC_#(}7qf4EtO zt41+e&X>jD)of@=a3{TnWXUvOB^X$a9N5-Sl>bu$%H3fE{&H@01VJeo?BrbyvzP1S zw9g2H?`r7yZl0ZgvCM%*HA+#V#U}N(5V+a*CRKmf8q|m32eOWgPvG+Fh)J*l(sg7N zERtm*MQVJrOb@IfK)D+A+{O783b&?r98sG%-{j^^Ol{(JbAFrrzsXcz&Odm6EKE>| zWmMOPJKy|a6IYv~Vv8j7F$y~}qv^8z=gnmqCF+wIADml%j}ZnE)gKN5J9}eLMv=Kp+Xu{Rt}1f?yPYnP>oodWm#Ee;_VZaAyva%Tr6hFDP~0Eg2Ro zq(9@6?F#^Z4q8;WDf6*r!3iY?1-5d?d=%QZSTo_Mi;Wmg*HzP@E0}21<5%}RlJMBC z!W+z|nOTqyrHCOc6`u?!wKHBT@RpDvopf$4kv2%b)b7dsfn+u2Xv`=N}v%?xv}Fs7>D4vXtH{T5mx4D8rc>xUI@?HVGaH{12N*9zv8U;Fq&p5qi{;$VbVe^ia2pgZ(;yukr#whKA(Tt2LB+4{bzp;W%|;8lTj`>DuZ^UXM1@0I zO^4*B6PD#SyS);uyKN?!FTOH>08Q`~TFwtrj7dHJXF?DVJw8kyv!o~sYQ?AD02RD{ z^TYUH{a&d3gyl(DwI%JmlzX4`^fZ&pr^LwV@~k^5|1jqVffx|*ne>n@gZyz~1Pkkg zdkt9(&l3Z_t0o3CB*hU6y{MfwGHSq@P~3%FFAjXGgo76L7~=L6XKE&FJNga57CoI& zlGte@U(dNOqtCKYQWlyGhkih)!^n+)PgUWCW(bDqF&9G1AtwUqnddr{7(DM)+YnV7 z>5F?)S__)rW;zwfb<3wbKgI^<>eAq*X0p^5Qyts&6E(SSp6*#IrWVmhrQDyYqr%51 znFY5FSD<+ zS^o8|l34t~ST41K6}}r4rZHa{?W6(pr&(~bqhZPG9MF&?@+{8?ph489O<`4<43oA? zqN;_gk&a}I^vAPCI+8U~O>rm%vAcQ@yBf>C-pL_g8rG*7Lnh=Nw~&M!+D$bHs%uH4 zAXs^iAdXU~A^(bwmn0@N5QJubE1ZXn(ypl{!01&W-m)&qVkoBiH6*&20Z2Z2`;A$DJ%AB1GPdh$ zIWOcdSGO0raYauew|K+o#w@dS8$viVRk?OJo0xSZR?)B@d{#e9pL+}&JUn!(_x)Bz ziE)GV=UoEHxeo&D_|!h1_BivZT0PHvNQz9L?ZGc-hL9Yx1R+m5*o65*O{LRab)SGw zw!YsV&|zI0|R^CHoI?R?#CCzvl~-{&O1sUD>#oJD=JEhIiM<0~1QFDB8r!(&?d z16JM}3tbLYR3!v;h1U|Ws%yb-=PIPXJ%P|oF5X%#ibDJ#9|`4_y14t0IMj)_^MzUj zENdyj;t3~a?!p1~f@TMTW-w?qf1Y9Bczz)DbmdG_3bl7NHS;2W-5Vt{vjwZ2+z#a7Js_HgugGwoqBwPG`` zV>3Z*v#G6VY^Gd)>}QP@0`T7An$ie_1tG$uRzFn=ui8&X*i0kX&wXvlQ|J$86>m=B zcpB~I58gbAIeBkOU{M-h;t`fhvmObhlGFg}#aJe}oX!+`yU)Suw^p#QpGy_Vb*gH* z$q8#lD?v(`Ii!#&me2A=c!Q==yCTql08!3k- zYwecux4Qj*{H+ELPQFQPNQ5EFVaspqE6;fQqjk?siF(4yqvGi5PO{cMs8fibyShz1 z9-pN+Djrj-aSAchbG1|tSJyYYY3R7E=yNbZHs0*^ZV=vW6LTeKRj~2_FF*TER&1UM z@L_{_96o1siaW=GBR(AqOh+()dT&gxfrI%IK6pESFrI@9DN}t*3|}(8V2yA)G5n9R z@b>L)96k?*=aKiv`>*e|)}3XM_b>8T`^|p2iN|S1!G6t*>+~_}=`wAqk`nOsF3zsg zwyyV4l-AV!Q3(YcRtA$GTz-5)K{blm(Qwt>L^1z;E{*A6n*qvImg`02(K+GoT+Nml zp3jkgvJE7p))F482dP`KHiDqfPKwt8eg!&Y;l1-by!pD`Drsc@5IbQ-*j9m%loeLI zq^ILO^PW;gr=|SDt38fl z+?y{a2)Bj^IZ0q>$zdvSP_{&faC>@6$qAZ&Msr=QFhNDjv!+nI9P9`du=j`f`!B37 zXSQqf|7kiC2xjW&xWBRb?7xCx>C;vXd~6zm$fW~|G+JNXWwxXt-`1Vi#P{B?btoLg z>GrpB=M82***!Guy_>d(n*|?We*TsZ5Zy8+8WQsxYUq*;QcEN=qhU*b zK1f6OhjL+|if%)`e^+t;`I?Y=VAg8xZy}qj@1#Df?u6h~&97I=cW7v}%yO`U6Fxf+ z-URVgWx2u*y+61^r=9^SBw7OFRXXo?aYPu=u>JHq&r`32bhvuz!P)vuV8TVmz zp`=g;nV}vDs5)vJs90#Unxat2_1TIrnxzVPqeuaG6)Q`T3zvtleNAM zzO<7}=Yv#^Ij4?Bo^x)gmjUR13KV#+dQ(WP1O~xOgZ|0)@~5Qmo|5rVIFMc~$v{sI zlUMS9!`l+qV?GndhSZ`)>-X(WE%IJVr~Iv!jH#V~+@mzs^mI2))#pcUyZs}#Nn5vj z$7H+O!Pm>3Q?$#inejOv%EfNT>`8hgHxKmva91HR3!4?E!j)8ePUMDv1BS!}OT!Y> z2YXVf;M|2@I%nC}&spZ~_fqhwT2h@HqHH)bS@OttCsa(PXxI!E_> zf_*;v;>ZrKjE2+14GV#PhKCwHYN@YJSdO%NO9RU}Uasp=${!nh-U?=EUv(c?x%ao- z`+J(iuGQLb7$UbWUxb?v(^6YqZPQH?glY0tHC?i0i}PlVgf>LOt_pkN?s)OKH*s9I zUYe#Q(2{N0wp0`X7%Ge|Z1o9g@tjpUrs9i2AWJjL_$?h2=F{&0sy z+Yg^oI$jO`c=6fe^}~-B@83Mob%&L2U%r)vx|gt(_7~t8Ysa4_YNbz#T3ItN`@OOg)6~Jn8z@~00=ZFcB3zD!{0D;Pm{(j)a z!v$S3<9&khPFa7)sh%^Cfu}*R_&Bwmid3Ow;wfj}`YI0=-$Vqq02iabbFlaKsi0iq z@P$|T0@eDNNQFOo75W9MS$aJ>`T5d;uYI==GP?84q@ zI(ePGl6P}SP)5c+<;>t$Xg{}&9#LObnQ7_Rjz{#f{UZwdWe2l6D62PC?z%+p?8Zs~ zJ3f#W&h9%W8Zv8d8~hiU_vNyl|6TG`)|_+=&N1p#Ag$Cd5G1z{z9E;?2UF}o2umRVc+aGT}H7|xy=TbbOXhj zWVkj>8?X$f{*QkV#n{di?SvRdkYBEO-c-&IrozZ#SUvPiJTe5R7y z^;VIS(Z8Dx19-DQn9*NsQ7tFlHMsE_O6uhapDoAp8;Y2pi1`HaYv z8}jdIxvK5k9{n!Z-*F-2U{sa#4ZD?UB64@ycJX-8LdrkI#3OPasUF7kL?Qm*cJn3B4FdUd-BziX*-Lc@8zdOWp%dKmA`6gd5l(5ev*j7Q}ZZz>4X;JH1m zVpF|DN`rnXae7z1*xR}xwU5y-{;g8AmlW-7QB<$6R7&`dPuDVvV55UPKe=?^_wyxr zeq<~p0Gu3q+pI$PQxv8%cXZj`9p2f!c`@PfYacgpzIoH&@e7l$4i}N!jB6ebUv4&k zwHb4~-Va~CZS@A_;4h~NcL2X-7S(Ty78Uesu;>~s1M5OD@04sYsD9UAQ2FBV2)(v@oTA4Yw!JDcR)I0;H97{RDsx4$Nq~mC*qMkz< z&ePwXQZR#@fov$>+Enj)qe!QiVm za%Ev{3V7P3TFZ{?xDDOsE9wWjB2f>^fYFOV9zb>j^e)Ka`qordgJAxDGa@O9lx$ab zBe-bXvL#UzpNEuhf86}d)qnoBz-`Ivy_9uHx1Yb)cmJ(_|GfQiyXQ45fu{AJZ#ms# zS>1wDl}H@`AMOQ~vf{{JfB*5jyZ!Bdw_oc&Bvh8t=i@fQ`BDE|{SEJR-daaf9f6J(f%H{S8Wce}OC@OL6Dt_;`l} zm#>xJ&-z?&InGXzXD8nd;M$B`SEo9A$$+#ykHXIM!m}S!wnsul^nu&(bleu%!Lf+% z9dE|0HJ00)O1hDxjY1u1qrw{OGsVtCLJ7SWFsqRVmR4!nsM17jys6x)A*HxNlRDDP zxT1@U_yhe7PZ{OYfAzT!7DOaT;OyiFI#02V##0|5`)P{9odD02e3K`>g{RI?9&jH% zVf|YMUdG*>+Gtf;n<1kR_|1ZL2D!-qH3PfuNM{8d3%jK*Gmx9u)!srDgr_iF!p36$u%4$vpFfa*~MCDcI#$>f4_GUUYt1Vl1G_!m$qrfU4$1Q zNnZZo1?bDmHb&_n{?Y|4B4tL1C8EUVR+c0(NCZ6dF z|8N<~$}?Or-gc_1&N4A3L5%}EGbWa=DeEF6+q{Ve9h*6(C$inM1x#o~V#J(2C<-V+ zIVlV7e}`Il^N*IgFw%%8har`Z$$CMXLO-m30H;di1QBuEDmsHIbWF$R$&_iz_!uSL z-@HJ83xA5c`oMs@5^P7aBvUosgW5;@Rve~a_%SA3jxtCqY$~x7H44biU6MR=H88Zj znIo-m2>VDS7|C_m0}UvYcuM`tfPrRT0R|Zse{)gN%>%nXm+M33@sI>U?yy!&WcEbi-8RO*z+S zf53vOf|q)2E}@v7n8orK0Z;HYc=409O@1?10qw0(jW`Ucvm(S=@@{RnViBmUgSykSv}QW(MxQCS`*%>}2%DNj<}e-CA^6E6uv0cqEW;i@df?*t8#`uk`vkQtct z62No-NJy&!(<=Z%feTZ`U_pHYe|dZ_fmG6%EiiN~0|9-KU*Z>3t$X1*Eb=rc!Aqk= zRt&aGacM1A-I4A}9a(W1DU1y4UN#%_FzYpoHV}y&6T2NCz)V2hVkX-*JJniGf4OM1 zlyWfYNEnw|8Q&2IT9R=EH1y=>5`^cFe&|CJ(VGmu>VV8yV+D4;eyV z&mywjQ4s(c8I2s_0ghy;Bf1N+4?GSA` z)2^1RnY(2}A9I~6dEwW#KC~7PnyAo|*3PLKuNytw#fP-?yk+wU$dJ`Dq{8*2PGkTg zlcgl)i6UPs6`gm;TXSE9+^(@}t4ivsF_~SRENRA3NHF#$#hP8&?809`e%TEU7D2>k3qR zUNc^0m;(6OnZ}SVf31tT&nMlPs~Y`PM}O`w*=Vs`+yYsB@GNl+js%xMUnlp=iD=_o z?^(+R&oLhNt8*JdaziDpS};lTc+1E@?O@@S@m^=_^Fi~;6%if0nXNQP=So7vYc2ua z?odJwv!Y%#k{K4l;igyc6FhZS3d=4^ku zJ6=$BB20xwf4hlD%b~}@OuMPe?3`&Im^VppI&dF?Q28~&j3I90bv+3y#BD6FHRz{p z{v<;&Jcb*4lQV4!fAQOOlOce){BhAnZaM^8Fa7uO&&zLrm%o0w{PL@` z7rzB-m9v*GkBixYbH-n;eiJsh%a{AVb|H-V0H6NK_^XAmcOkSFLVk^(L;MVQPv1YT zN<$o}Zt)k=75q!1AYQld@t-gMxJV-sRo+EfaT)!Yrun!+h+p7QX7FiFl~&i~)^jpW z7O@f^OcQi7T9IbSPyOwGOV3FuH)E{rd08*Mz)cx84Nevk)aMa%;#MV;%_n-&Rpl~W z<&K8J=KM{5rBCjT^XW350=n}C+di%mf=x{FAKvUpLmH-`&7P@_l!CE)dk~B}paSn$ zC^OxVh(ti%JMfv5qw#s%LKgZ%y-KkK;{(ZQIoq@}?X(1w9b#4(C7t~{TFo`gK_)zdr&yNMU$%vNOz#kiY9P8NV$6=`A)aHgl- zjt52CDC3!+rCH;OSMavbqRY|HI>ZVMd`-AD<J`ve8xrsNjm+m-d zmL`v|d8KRQ_7*~aa4IgTwxAuECM}V>`qr35%$QA+DrcHT{}KhYlfWe;!qM>WZP6 z4@Fg5fO9#U6h6v~(zb|hb~@!$M;Rc-W>e#fN!Muh+tC$`D#jDP9T`ue?bhrTIfAT` zfQv{gWDy{LGtijTfIActbn<7_1Vnwa&KCKg73-rUj1c8D1hF?F$T?`J4&qdRddot( zNA%*th5eDgf$=x3IP;dPOj-L_uFQiklMf^>MczT**Oo}Dv}J44C%GTXQn_0;LuRu( zT3eWYWRr)b&wSw%{H1S=cjW4dZCgVvC7HixhK>7wN3y=8Vc8g_BTa`Ae*uKdcXj$a z9S6uQNariEvgEdy<_er)Z3C)TNZkKyr#Zm}4pKQ6$U-|uFsB#tTXV1^*0#Fh^cbmp zI(yQdHfg;(7zbz9$DMdh<^gpQpL+&_;6*rN2cfNwB|56#*fWZXi=uVjdcl`Gk{Xxj`j9=@S|;?F#~0jv17K39F$Z9a^8ju<<>5bV2eETwmj zE?xnG9_S=m2w1O%^hpm^6z_GMEm0GH+@GA-eIy^lVsgtUH6&~d!8PDgJw`N-IX^q65w@`z0}yY{wrGt(Ig<}Hnz1VEo6^C5h6#vb z9-x@hnv@9)HU%DIBjMcLw@g3{H}Rn5L;o<>V5~obOU4E=d4mBXw&~j%RI<3^x+x`{ znQX$~osHq4z`Ieh$w$>79L+~K{wQIR&M54a)-|a$k$&LgN!2-=8qc#?Cen!= zkS4B0iKiJ%XLg2i>$leuu$b+Ma+g{5ky9W9Kq&i@(niYZRi4xtGDi7RvfnxHBlixe z{PL^GdLx4Hv{%quoJw{%;{9U0ZjN#bBl(mDs3q7NFh@NNQ-DaH8p`|^R$NXW$zSQ02Nm#iF@0{)@jYh_? zJRUgF?RYMQjeM(k`=~@&jCvG#zgepQU1QuGtaD^nRk4Uo+x=EjP$@lDLPe93b6UPz zfItjG`U8rEF4cYHF&wFX3?9E4#d&rTM>3gI)oBKj<{y(CaAT-U6<(@3+_8LGV^CJc ze4oU&mXVkTarolBij%LkKhhGLh-vkd*`%=oTlR5MP<{Z zS#}Pu@X%ND^hr&_h@D3%OdKmA0SQZvA4%@QF9Pt5fcmWj$M|%A-7t?UvK;v183?XT zmAQ0R%DREh2j(k;#IiMeuyP5G7fsFDGc|T*a~2`EbJu@x#1!nC6|~Yuv9g1z<5s%Y zM2|)dheDSbP`}EdJTT?$4KM&&*ir~qdqaHMz)1i~2T-7E1nqyq+@#<{&|uIwe)D(_iJ+rV3m# zRlxFJqIG7i`nQms`o?+^LJ4NP-BKRFjKX5fdTWtht!&#Fn4gz4`b~Sy3*O(b5^o%K zuV^p~=Ha*Ze1drwy40|Vi}~DZ*!!`pp4u&IMrQI0hL^0_PI!{B?DeTnFkMxd*`Bqd z;cb+Eg^*>lWHIkaLP6S z?Nhy?N`qxJFznv3dzZ2xNZumm#tyQqIeUYN0H|HU6Qk%!u8l`A5YM=v5gIxIlqz^h zLl;w*yP5)v{E2d#tQD4Sb>!jTqad9oMMey{a-JGWMV6uMtq{!5M32oUd49TTc`m1a z#BCZGx`HRTmphY`FhjGh0aob)!#LA4BvIFT#`$RVp_!zVbnUjcdgpBx@$!Y$o**qi`ahdmFwt>)DbTP4w&eN`Wu`;7}r)G9Dd zcf0;nYi7`Ruw|z<7b8QKF223A<*9OiVD*rmBU(?6ZB3JxsT$MHAhkLtk1c&cWm@S? z1m#yb68SBSW7`-28bJOMDnb)S*i&snMP&L^7EDBl@u!ZfuvkkfO=%>_rLhAisAY$0ed0nvr?Dw3C9g_*Py zU$l9&46f=Ng&F%cmJR1Ll6N4&jk^!SQeHHl6(b&>LkM!CETBEv$B3pi@P>%heu(81 zbc~Jkw(gC!WEHB~JUwQvh5I{yb7^o6IM4Anxk1m$AxBx1(k2i`%bk}-3*+hKJFX-h z!>5#r>WZ9W(Xe)gxr5HI3YHz^S*VU!r7SNnL7+T7cO=0ZH9jZC_K%Ky;$G`u;gq%U z)jx^joKSyWR9e$x1E@-OTn%Z`H^XM&-Hg0M(W-@W@Gw?EjbipQ%-uJC6`)O_cUeL7 z9hH_|xyj>m!R^jbJ@0I0Epnia04J{d4&a1WEBA#)IMJ~%7dCi#0BN^p_6yebAKY}{ zbhQ4}mwdmk9iMw8 zUMReoT{rJM#!*wB7r2Rkuf>AFV8JkP<3s91^8UyI^Y$qBLVJ`FdbBjc?DM-%bNI(c zHm#$p&x4!k?W{!yiLjfAqSR8ZGyi;jmF>fb+Hb&5SeKr)TwBQPsaPDT53Dk$AGy-m zgrFuA1gj_2qea)$?i@G6O^t9(D_aI%;OhJddw@)@Al&9n>av-CQn)QwkzWZ*?DG6Y zu~mYEEVA*QH93{%mhT8-kr4CDvU^XsW>1E*Oqq}ZsrKy>~F^ioNJq{KP9+`0(1Fvf|p_`8k z^6N$yPM>Y_Q|U>6S-T8&k^1ofQ{dN?qeS1;yv+7l&vR?64jXTYfk~aWss;PJu${U~ ztlWDvq@gcuSA(BNk_D=Tcq=MGs>IEPF2_HCAS-{hDQ7FYu7a~dpRFYK(Gvsf_w%%1 z>n@_KXqS^||9+xmSZ-XJM{HmE-a_Cn%Y+^`V3&>@UPLy3?0;YMT|3iD1A?C~)dBL! z3+f}`2qp=dV{azYi{?66GFPeX`jYy7Bl$|PmUW0x=hu^WnM(`QjrI0)tU73E;h~oH zHG&R(d&2xOP5-{X-lWt9nW2Huwy952X83ILL_DZdDevAbPp>O*s5jcNar99;Hmi%M z7phH|`cX@Q$mrJBh37zKzU8Z z&C{y^zy0O^(W>H?apeOOmtc(nDStRIK0XR_baG{3Z3=kWrCd9YB)^T%{V682gDFeY z8)!^mz&3E%0h`@ju)U3OB?;ol9_}>rvVa|W{Bf&?C z<;c0D@bT;4=d*vE|Nr^%w;vyW{fGUq_9!9Pm_L4epU*$7Ie+K!@#FDdH-BTow=oWk zf&2NlHOAk+oH?I^JO4hP`-d_1moc&6@7Csg2EV>OL$p`M|9sbI)C-8b0&}5b$y%X$p7=>KicuSqxh8i@wT=GH^#>i&&4Ny z!^@O+?s115c<{ZcE*9CUIxCx5l@;%Qy;pvT!W z_Y<*39|@#*_>z1T20RP^+`a=M`b645eAS+Oi>37AV0O8G1ti>_9M<>)9N0{ATYD!0 zykxMp$*U-*4+4vB~F`thWt{f=^rZSu#K&xSe>q#QyZju&^+);SZja!-n= zZN1O?)2M9#$7@G^$Gu8!6X)86tF8Jz6I-$aD+n0cyeiO6J}{tDYtrP#srkrrpQk0+ zB-ByPmNA_FJ@ut|QkM{i`{{2xLHEM%d_0^_nFUa3LKl}XkO3BdXj$=65xB!!>(Wp{ z4-{(e%aKBXa3$2l>yXP$^Af2~eADdl6Rn1=sx9s#87a``(_%UY)Sf4wg}{apd5$z6 zNdWP7r61(Zok8+ZonMbYElEA}wSP~FpbT#6t1T!4O{oEnMziQvMBF|ydo83;khs*o zj~KfuKJ07)xpgUd%)w|JaK&i3Jg+DT!IFgxnX}N9s>bi;6Kvo8S&6BO3r~9V)`@X;FtvsQ& zI^jUImEjmKWu))CwdS3ZIn5wgkDRaAouILn#Y&mImoCPft<)lm&B^)8UOE!B`Uz^D7*4H>_<#c;v((cLDw@)1Jz0~FSO9c+zzVva>w4Z`< z6RNKax1o7|&VFVDGT@D6+~_qEL9D6$f;Uw9==0i9Wj6oxG0zc?x3k~Q=Rqu{Jf#R@ z3Kgs2_N!7Ux1srj;=UBJ#)!$h<0`aPJ2q8HE7{HmWYQe>rTDD9=?ZVe4}LP;As@M% zDvM2baQavrygHT^V^+QVlA;j*2u0aVQrs(gwck8{KDA(PhYWfY=P*Z>;~Zw{Q6yI_ z5XNbt{uSSe8X$=MlH@b5EtWxaXUnn#s~lE5Xc`Q-o<8LAwEm9`x+bs%Vf^a~ZHZ#v7Sei$k z3CS~223s)BW5P{i8t}ba!K+mYx?BV>CZ31tL;!=riXjM_Hx~gPjVZjOs zQzniB<&B9$B*La6K!%0rf57r>ndHGN`kaFIiVN|FWS$i z+al8RbosSaIv_mtOs#i8S%0H!YLX#G4`Qlc<9)o5+c-GT3C-NrdYaY1TsUlZ5^n0{ ztgmfDBkXzZ*41;L~lh=*|};e~pC zJ(m=lq1g@$(0&vi+4+l+Z9JljZmcz2aim=9jr6aZboekUaGO}PiGtbL8rz9a5E$Vc zDh994Fc|49wJFkL;6RN*&49QiJK`+ED%vNC7tH`HqC%VY!i9P!g%=vl`qKbFNXh`L zd*^*`$>B|m6dVd4U2a%Lfj#ZxvtfR&L&UnA&yRZO|Khzcq!d1P;nam(yTdlJc4l$Qd3zA z3fv@hV6f~ST`9wVX(HjTz1#XW34m+%ys@u76I|)yv-2Sl2HjyvyRmBTh8I#KKzz#<(a{bPz* z;?-tM+Iq+FH}xU*5Gd;2`q*5~DAu`QII1VC;Qa3|lFL`=J=e^PAI<~Gltjb!o0f`d z&NmH2m0Nbj%E!`=Z6AP)*q9FGSy)4^r%a}zCxBuLa8f^byI9kwv3-?)-L%-s08ebz zRoV@LlL>5CQ5XVZ+4UeN|1UR{Y^e8%ZcxqGI29te~l9ZVd<(V{$ZR4Jzt@xBVj$)kXnH^DJ-u?#rz$&&wl;IoB0-82wQG&8*K$pn_y zm<5K?C$)oFV5q77mc7rIUwHM^+En$3EfvS(NbfKRx7;yEh^zpAltVTx}%3@OSs7@m>JgUVjV@|CfeJm zM9~`?xzBAN6zydo#LM77D8iKZAMhpBeZZ9e!zXX#|C&o=oj+0udB*}PWV+i@*AT_5 z7J221(`9wpL0nwc^DtBhDZ^O{vf9VaZ@j>Lhmg=)tV!E{m$}R~tRLYX>kn|)5*3%# zvIJM;0kp(ew^Zq;1kz?uLMmrz45*}SIne%uPbSTykB|k5y$#FQ6wO@}K$17??GaLe zB@4DhFx{i9fB9<{yT9FMV@9UOXLr-iC-zf0>^7qVUEswgUn#eHO)dKGmk(yF%cC|DwPqOtiqR1w&v5*`#{%FM^s zR%k563;Ux4QLEOOU}w^M!(q~qW;s`RoK0@fs;A0~H~V;+6`c{(?Dd1L#5Y6T<}Q>Z ztB}ioRbsEdnH*1<(DTkveL`!i^u*-H@AO6PI1%>NWacrcgtaX7!6iMxq0~){Zz?{d zaTs*RQ6GA}4s_Wh8M6)s+XGT`ixCaI+}{;{8vy*EExjSElHPdoFiYUm3<7E*$4*L0 zZ{bupHhfhY>2(=>QRVI>$V8I}_*zwMlr4~dcW9lY+LXV2S0hr9N0H!adJHdot=Fbi z`ALv*jB7#e+zF9ODK5g)S&sU(m<#&YB{5>tyXS1ef=3G*&FtW7_u1q*$EcWQa8ONS37k2(h0aMb3qZM9ZLqnG;inUbvPs=;^S zw9(Ju{`Mg5qU_2X(Z=l8U@Z5x<{s1kD# zrAEQu%#To>;Q|)Yk%RLC(Ub~*pM0Et&*y1U*Zbqtu6~9_sc@>i&)vhB1br3hJ>of5 z7fmn1`|>TTgID5=%aa#Lw2OPRp&bl}k}LFnhUs}y&2RxQm#f>?tU{`PlJUL*30qpqZc})2kE@}7!nw14(r4J0 z^>(Htg=c6_EBUb6f@&Kz2H4-PW-{38q$5lp3W#POT8@i-_V-+HS=L7GhoAHrwy;Yv z+FAxjE;U|*6WR7`rIjvv4!RXP%GK~jrQK?r%iDzm;NS|vm(_SqQuJFcK_-qlCwj5% zUBGd9b#WVhl$YTLqa!1KVqKAkd~v~vOV7&NkzrZO#d>lRr=j9j$${*7eG8l3Ts6c`ULd$Q>k|a`b^V4Qh|WINm+ z(MZ^Bhi6}B_9u9=#`cy$Yh~%@ugM1L#TU*IYV#O%SBl`1)U02> zs9(>iK}o5D`?^i-ji=jv5(7(u&fi3ZP?2^vFFa|G>ps{H$ zbA6mLqs>jB{pEFkQ^}-vcV-=?xT}N zN(Z+Y#JOT^Y&)xkH-R4o_{5{Q=*tTD*il96TH=z~1?Y(ib)$SEYbe{^CN!Pf-BGu< zHliob@ECb!%F>A5zwvDaTt&RJYi+=1$%6C{2c4yi;V&P5$OdAlCbB{!Zo!tn`i;)^ zi#+SWb0aw}sH4Rr>m!y45zU zAx>E>Cu)a(vH1e5qOm}L&e`K!SGG^TSk!J4T<}$YdzF~GzFR2oT(tPvb8^7H`>B#~ z|Kjpi@>t@SHa+`NgipT%)AcP7*WO2vc&=ukqdsoWRHk+qIramL62p4ZIS6cvXs50| ztJRBdQknIN6=!O8CP-Q2o4J$qY)Xex9$~A;|ZV zu2#`?le_xeeTqG@C!;R6@IGLm48CEn_1f!P?Ch-K@QVfZPv?D7=Dwj;<~r!xX=WnO z#r4c|1QWwZE%;xTcfU=svXTvSu{PfMHgD=KOl_96sCB>05)T7G+|uXkqpy~^3X)uE zvsE&G>@rWjX)<^nn<)-dcZ1@n-6K>8H!DK~@@O6CV~QY($Lt)B_@RMhr>|BR zLh|@~EuzqA#C#m#7`UyvQGy}kDl)~5GGQ_BcN*)DCP*9#@UtWYvz>w0Uv0c7T8#Dx zC5c<+r;inkf}6xQHfHENAIU=++I9PpM)Sr8j2Luxxg0*ku%`Og_?bV&qpFL;<|cuE zv+^@|0_o}daJL9*(hKKNo8Hj!VxA{e!OC7S);>_rbSPB zT#X+1W9K!DfA#ET%BWkm_dL=8{2Md+qN6qw(Y!WY@3af#&*)! zX>8jzw~G(E`zPj^xpVJ1?%k3xTdm%YA@a7wc)fBH`+GYLPr^Dgmt`i*wbe|7oJ>da zL28$o=jBmIV#TvVWu@&Bxn+Vo+a|gK=|=WvYWP!S?YHl|L;W>*E_KzNt?4Z0X$EKL zc^em7`rPvN&=&C*?@sQG_xXq6n-g1r@B2~SC(>OPpVWVsoP6^^r~qz`{~ZJ7=(bJ$ zSIY?G+xk1>;8N* z>-q6v(v$cmQ)kPa)(Ct&=DnV8JzN>|e7^6Ur5QXi+B6$Sj=r1?7YcmY=RN6m`>%gF zF%JQs=FvU9J-rJ(9wx>goT+Cm{|ecT4{MguD;Lj0H&Q^q5^)R4l0(`stzREJpkmuU zt2xELy~W>Xc1l*w-yiYFL}7LIK6M zite0`o_rBrjLyEWYBs{w#bO<@W`41_%@6>2hlLIR3+6Au3U0nK={EP+p>Juqa99F9 zo^>pKZ9OWk8yY@yYr0&yw6mGd!*}^Z;W98Kd5jnjvUoOK<~Gkjhc3--bH}gdUb@1X z;Pg6cRyzpRod@G~hppL=wKyDP%BdR4WpB)Q&)lXO`aH{}p^>)S1)Q8UcUP@W2>>G^ z)xV_J`}pNFIalIH2ZvULOK|UQtq`X`@~$@+V9lh^@SZc>CZ3(zOXPRt3@}qIhWhm2 z{PfZDri=9>kvcc_oAW#&12cA><^Q0F81fiI*ylOg7Q8VG6-aT?Jc70GU}Q3Mw896i zLQN>potiiu3Ay05Ts^2@JYM^Jd;vUzxiwp(Pd=xIb)JGdNtAZvY-uLiLu0Z8e!pT* zot1tgcm~cci=lNpK>aR)9P(t=Ao$T51IIiG2fU$RCW0fIVId*j4+FTdV=qj&ypMtl=^mLD&8V7Aw9J+*8rRCds=DKgF2dLaN+Z zhkC`RZAGnl!e$je(8l2XA2D_o+LMUX~@Stj=;KWFQBJj|l>yzroW;RzKpv71D z!EbcK@9~!1_60vEX-hdGy;BoBVvn*w)55~dpqv{j;G{?$|B5xNk{1TLyZJ|z!AvgQ z{o|RGjfxm@U>#A6FM|Fyp~ng>(2g*P8gQ#qnOK48Ivkxi1deoQcIZV?SxD+Y50H?T z?FP=(gWOS_S{BA!1tl6qbl`XFJ%LlaVYA6uUhU?nK@x7lq`@dMdQpmZ69#0kWhJa8 zaz{cf#1*E)**poHQZxewh}q=wDL7|;lH>zsV9}g3kX2yk0<)=4?<4&vU8Xy($_dksYzSB*n&e0# zr-2k978(g-j<)&s*k&}n9D>&FgzaG=a)-o$R**lCe^ zIjAzyL%{fY1sawf!Ec%K%C??urS>#Mip>ccJ0c3E$L`eLk(y=7PIB4N-q9U7EL(gu zmpNkmO*mEK85u|(Wn7))MLIJTgdFj5?baGh!F5wBmj=Jp) zn+LUn*&^Bng#{1J7D3}R{~lfITOl;t%|617qR7h{nQ^|bg7Dvub%?<&N1^yp_4~SB zM{-II6h~gDGZ``3iFXf*ZH7%kjQE&2Y)0;;cj}EMso9DUhp+lLGnB;0PQK+8S~&?e zdaAjGK?7qTId7hz-U@8q6KlD_glo)?qGf5dwQzGOtCOj}LswJO{-|8hAn{Z5L_a>1 zLZRH%sM16eQ3o_Q&V?Lw1d5W~hj8JuYM0BpGU>T6*@Zh{Xj#I&Br_iM);P`)MBN`t zetQ|EFtP`YJ9+;+tpla+pUcS?@hJrf=l8fcBO=~fq2F!NHk+Y8@@Kyy%Z&g%0Km1)r~@zYm9kHMiP6< zwhH7EF(K+KjD?FLv1YD^Qz=p$SI;oQ$}L&wHP8npAYVOvUz+j>!db6GDV$qPA?ZkR z4@q|utCIQI7VfZpXb?tFgFcLDB!M}Ns*F13#sqY_h_?_XisEt2+UlNd40YBnB&1?K zQ%9uGIF1eVj#7`$`2zWB&L=8YU`c;Ope3Q6lZ?32Clv&!JHtZR zH&&?rmWn#E%PvZmvM-E11#!;igcM#SJ|fSz?itPe6yVaA=?%o1u`GfSu~rD8*PFHT zKQo^(l@FinJ@#p6JM}hYS=JY;j?4E8zjLNJY;`zu3zUrJ`{-Lw4k6LuA9+ek(U*?D}z+gdJwsqgpS8poO53 zal>dFTto9?EJ~L6C|Wbg#1p4P7x*$=*0TKeB#|%kbhlIXCNK3-VCo_-1<#Z}nAg}= z3ydI)IX-sg#h>`{NIHfuWHiUbZ#bG0$)I6ojozwHUqb^+Ryr5t-PNAW3#aDdSGk?u+_ z#@l0?YR%|ip~n`TMbn9y;UE5LX?LsfmBO<}kV!(v9N)h($3M4}3>iC3u)B@9V+xcF zCrG9@vpa)P6mAi@irQO6*q+FpbQ1>>)hy7Q!w#swEtbxx%}ism#^$QARk}{yK6n3y ztYga2f%-@YX>&VHHQI*;y92Jp4nGd#38Uk0U{k11x9YK`c09|hRvEublLX*2yoUVc z??(M8)4V8Fd&~X=N7-|otZGdIaq#b_#swL`BF5_<#lVT2HnZA6m;g~d*)z^3e6_eM zK8-iyfv=FS%sKV?Mbsd>>NpX79 z?4GtlU~poUhvyDoVaIKU^Ap+l@4E8+r1nFq{4;tE#;O(kIvc|^-dNizr)6X?J(S`h z3e`LIEXPDu)k{Fa-42N`6gm`!Kiq@gQD7*KWONr;bX8X}#kA)Ahhmu;*R8u;qw)%t z3GcQ}EbIrgb9iWTdp!X1_jbw}cc*Nr+OkFj>d61zWq`v-zyr;a_g|G*s5nj%><^?e6yK&HwoGk0~f_yp2*C!nC zZ3~{ScOpW4D0BvhlYgukeQ`5j?qRwkZcfSI!BQf*J1EE+BFwOr{S$~|XWY`oxh;Rt zQ5)%P+Ky=rNb!}lyvwD%CVY-0e}gj#0lRJPYdMjI9|xFCR{!!RHeZTnAKQ7Lu$va2j0)xBe;LzLP$>LHJdAl{R*8+-tOMM{Sp5 z;svCfrNWK;yX;Tm>IoYOE2>IVA(ljjOoMYc@s((h$JpJ-HDy7Q zAms1W?k)BZU-{m0VfK{0dSUjB;>Rz00Y3nHXT6RdZccnd;%iF4XN;#P7UC8|`?0R- zAf;chuR5IxV@f6FRBDeqt-0%yvgWomOqN*N_L{q{8q3SX(`0y&lPZW-cWR>{YBYBr zHuKiq0ec!!>r;rW1J zO4e9#l%=1x!B`xr*~3It@o^p^x|dQy)5H$Zdy!_kGOWI7CvmNFb8t9AFLNz<6Y9xm zkWm6!>(W&I$3mt(jk?%U2Gcd0k9tE=ThU&|cnOxIuow?yF5?@A!HGEG`sfwnNT@tv z;Q7|c)yPPWbC6`CHrg5jkp6+CH2{pvPgHI0@wYg(1?h3!e`r7unK2!?)Z*r-a)eeL zW5??Ch@}#hZuUX!p%RzL=NjI38;ma zcjgK9z4mZHzsl}E&(^=cd8$mgAr(Drn-k4HVTU<^S)YM8bjXF?=*~`k^8>Uj_sbou zey;_Usn7;;fioIo`^ohy$K+Peu@+YUPO5AjKah5ydNMev&b~!;p5nJ*Vfy<|k_f`U zA$PE1t=yB|o?Xvn6u+X=cr#bWjNpd(VLcM`0Nq*Gc;pDGL{0y8wJ#!CSe;Q0E>aSf z#cF90hl{v_K&L60N^LNw@Cz_aSSx{a!HF$aGqq!%`k8KL4k^%U4-$Pst-- zpBKpMWgOWbr8W+;>DyIs3e98ubw#g1@J1RJ$L>52%Ch68p*@%xa5jT~cEv4DvD+0- zgErZ(X_MdF_!>i)g&Cu`c}R)kW940bJ|N>-rA_Kn(0wpfjybr~4h5ju-7FywJ?%3p zKysxwNy2&!@eQ|_-p@+h+!56Gns%1aHOsYkTd@V%F^6FzK0Xx5qW*?bXTAx5G^-YO zT&+0SR{VHb6<~)%g3-Oog797tq;jOBctqMz8;lwksks0Aa&+ezglUU!!XbHJk^mlX5m|HG)G<6>*y@-I z3U4Y`Fx165h+_ng46UxN+9H>c<#c(eM5sTSM30WPB1jJf>f7VZs+thS<>5yK+; zb@j>Os7H%54iiz#!DG`c>ffBGoapOWg#~(5|kt^Q4(xgTwwC?$HY62HZ5Vl$4a1j!u7hB%3lJ5@;=4 z8BM4E{y`wjGQ}A9a03uN_Gf>cOF9!$r;;8e_6^s3@#lT~eR;Yf{KqG_X#ql!Y-0a2 zG1?ox^ZnH73DXtu=leVZ%%c^CzW<@GRkRdDy<;sC%6!BDa&FClMRS=U_>B=czpb6RcBP z>xN=r#nk?ec3P`}KfL8pW!11YQ9>}j1bjWmx%)bT8|VslKTeHeid33B8+CRgNF0*T zSn0Xp)>7irDM_M-y+7di+JW>q!RuG`GOu=H^BH<6>_Nt%AFrGUeHYDGW!c)dd!=?2 zwP0U5iHp^q`y2sa#T~8;7hHxVz2Pxm~Xj(;c z)s6lYt71Hcb4V7GTkHRq1p|%Rq@}LaHmDIlrGVEW$ zLuXQuXCp||8ovLUd5y$16GNDhrsS)}&i>x<`Bk>p;ja&XablyKC)_D@ZBG^9M)+f0 z4w1$AdrD0+U3`AqVK|pcvZ;&Jal3gqS^LS%y2aksdb0^J>w&~aQ}x!!m2~{7>yx8n zf06TsHCx9eU--n`12K#qW23J6ZH^+Wbg}X#tA2 zLPyIXCjuS7iC8Q4c4gsZZhd;F6fG-Q)cW~QxWdxA^=UO{DWSqMwb-n^?NZS5erEfj zVavU;b|-s1sY!fF>1PqU9#a$}qJJFwFISvBmYXbH-Nw5iPos*1krWOv%9PP^M< zr~2sPE*365v*f+%+6O~4FETOs@sXjq*?xZ%trZ_|T1QNK2Z@Ir8Cr>o2?BGb`^m|W zVy1L~%Vo+@RrOu=wSI<-^2M!PL2AcwlxKPLiPuEna*Efao3f~GRg}zj9PMI0o${1Z zCpz1$tm6tZS94YP3l0Uri$?DDPQ!f+4$YrLi)>IPaTPvm2Ls2qZ)lsCz{K=7rkiZ4 zaJV@j9)N$z1z$0I*YWnSn%TU$NUKwWfmwlwYcFHRt)$pZ=y_|L(FHMxc+80X7Jbig zJQyKdq_O4X8E?wcWSM?)avJ2Z_1*_|)es67a!@a*T(5my&7K;w4F^ZySzt9SnGeMJ z**6MT>9uHxCE?S9(ysZB^q%=oxWt~rT;Kw*)wG{fj;Hgq`b173=?DZus>JXw+N+vT z*E{AwnCrJ1Ni}nx%uhNr_B*9L6=inU{YeMo(+W+J1Gzabv*vIlDMpiq-j6&4jpUIVgzi;-0b5hEb0ZBH^*Cq4lK#@>td9w$w1xvR_5KUuO{ zi1AHj)$mSYw#gnk^l2M>ap63?A`i-O3!?}=*l>yxu7Za%U$3L!4*Vo*P2aq?Np)HU z$q=F6Ie(IdQdPQly>iZSO5cZSU=w`*5&NUD@K%QqKd!Xzp-dlt^Nm!g>UOECb9mXqh7+#5xs)-c$XpUa3>E^t!509NBkx@5kT?1BNtzj<<|2 zOir>cc)9CRW!OCJm)$TFmxIx$=~cgl>(0G%8>_u@-H=57r4tN z3xwj>_J~&b#ZLK?3onyL=f1C}XayM>T=4x>Tv#qJ!)u(&E}R{jbzg_{XN8D;P9{RZ zJJW%L<6U%!H^qJpmM1=-#-S6bK8i6Au?xZLe3bot$ksxD4B5Nv~zCVF=w8EkzURG~{eL_${BdY)TxatD%hd;)#eN zN=B|wz~lqX#SrsB29F#+3pdLw<$G75^^ZKkEymu~=hH;uImZkWm$u`uS)lJZOOVWy zbKbX^t(aaRvQ(9VN^v!pBlhmacBPJejU8}toR*&3QGmZn7K`bv93dLV)W)akrP-wr zGKaFL5gnal(8g=yWYjGlG&(GM$QzC!uFI-_jrsPPo#L}^vvi`f*!sxj>AP@zeSD1@ z>yq(LOtxkgh_o!C4({`1+{tcRmqs6>#CVknu63!-M)3qg+C}FR&gCR=6F2|SB_Gdj zPbgmQ4*)8N#1RX*ir~!j#aJ#HuoJ(3t;K-TI`(o}5W7?$BKr|8L3efrS#=f9z5n>cWcK z8KXX*?5aa);W4r5?Wlw0qBnu!f|%I!Bq$jL>yd*16vi26Smx z6FX^(_RRQCh4#O?1Pi}KILr`}6JkwpNn4ITNEF$rbE!iuET_>xp)`6cvE9yVc zT1@zUweG89JC5bW7cw7MU^7w&a>e~=nNo$HDwRt*G2&fU0M~(Du4(dQQ(*`JaVp2O zzPPhYE|^2yNTxbN%WRWU4L)0Cy**@N=D~;!coaE2jX4OphRRT%p+uHQ>F&Nf|Zx^Rnqnt-ysEBC-ikyIoC+t(!R%Ziou z2=u1;Kez%~6PN-U=#cl*TG4>C3W@3#?({Wevo6y_L5ThRrX`xW@SuS0<~W{HIVJ8d zF?YDLyLbrCFHXyDeYxG=z?Cx^BpSi=i#-M{#MFn`X+>fNXp;m2x1iN@>Wu57yPicK z>{x~zMhw|z*&ya#SMyr;;$nM2>zp*cUP}imed9#ArdWhcqg+C=rNRVs?mu{|<(ApR zmAbQw2uJ#wS-rD?!oDw0e`4*ii`ejO_Az-CE*0)v=Xu+u?NOQua8_JR?~{W)<}!$= zblr{@s3wGqqQ4v1B>pTRp%2F&+jeJNL)DEc;rrHHFj<)`NMX5Z)#tRj1KuI^e6fJ; z;>A7~tpSYqq$j@k^ww1^@=-;{oXUPX*Zl#W8kZO%&E47*vhuwHh3k);QURF6aK$!R zG=X_rWKqjE>U294K(I>Wx8lY4EHeVLW=i5|2Wykc&#~=i+U&DAZGbDh*Zw9R#?^S9 zW5NRlF-y+e5;SBBkz1#-v$d5t0P$AWehkO*O__{t-Mp8?f{V7!q}uXg6v5U9J*7;0 zIhZo&*+F*D31X><;Gc7#sYvwVm2S(O+dVVNYy1>0=eyY%VElNps}zHxyJ9VKB$l#t zMM+8w&()N+8Ka?bUG+_0_MMjx$31WheCohJ4-3PG*KCT>MtEIelX4L=hqM>r3Bd+- zesci5*5WZ8tytawoOvAMxBj*+e%O_{2SPcv-Vh{X{maxE>DWe&zRKZ7=iov;)@>Nq z7Q5u{ivT7{;3*xC(~TJnj+mD<_#hxCqm$manjB%fl>|D2b)Phw(%j?8ihcf9%{QZj z{5)qEhTxsi^zw8w}(= z(-~SEr)KujGPi%w8#dXQn3hfql-Xw^^|jAcX&f&J3ys+At4vK}9zSuf^@DZjzc0n2 zZ7D{N&He#L-Yq?M3;Z1AKv#!+73Q}_t;(gnizhiXZtE;^CCy`BX=+@lo#@ZbAu2v zT>W}If5R0=tdO1p**07eB2j5rQ89DSdtM22^!42}_D+o%g}ZkZW6>B3C~EYQhNC+X zzJ<%EomB(`4v9DQe74BQ26bP@nbqHr`1f&|06U7Ro3!GdF;SPSEw}AY7`pAQ4|Y|{ zkD>QM8@g;`A%9=AKcCere9~wF`gKajDO^8>M9st}!%>y>XE&RhO$_8TV|85-FaptR zPIx+%+_rOKja9jMeE;I_pcN#Pk+%t6ig!IBi@Rb_4|2GaWZke6mKdjw-awRqqWfRJblDPdp>d}qWE>U zW8=4S-h>uxYe$iI%38&*E-}kzEGhMuRQj3ZhPH9M&OfMaAc2Q}1 zXj+T%Dx`iVduf=g{Vcu&(U?j3)Bd48&=p1B%DMe5uzqf8D^Hc5TQY8}|8`%|;$CDFhGySD@ z3>NNXs~mG>+PRt73ruSKJ!sQ^MFOf{o*~OoksWDL(eqHnT*iT)@>E)@s+=tQf(x(+ zbe2h*M{`KFKlM|IFmy&3C5u-cx1l?5m4otDUc|mAW+C#t8qikxmjYun2Ic-VQNcB> z_A#UZULmGj)%JFy8`1BOl7pDOF}zsastI3;*E3It2=+^|_1$h-NtT9)-+=u9SnXdm z1?5XvOpv}bHSDEX-g8my%IAJCiu-(E=-fTvRMmZNT}Fcju5;v=(dDJSDPxglPZicm zOMpF1$z+z@0LXIn;BA3wQ;Xlv+37)x5m^f6Dv!DG3x)Z<;m)_o$X42?L2l}d9weUC zUhbnJKJoAWz@k?E^{D>uoU%%nJC3deW$J^L zqcMnc*Y8^p0=y<QHW%xu%sJd1FJ<(8g+zXeoRm(8BA!&R-@8Q_Yn z;2WCpIyd~>(bvNv;b}bn9<2UPu6<#}l{@75Zv90t^f(jO^D)Cop>1TNa97n~S=gB2 z9MJM1S2z>|fYDpRL-F6ghU+6rA^?7ZS6jMmNXAsN4+rX>Asdz`NXIDSZ#~^#7Z+=j z8XnnqoM#Tqs+@?7?8d;K#{5IxCl@5H#7E3H>~Ih}<*S6D)zz$bxceWF!`Aybab+!K znVO$7-sr-Yb*UtXA3xE#%w?8APXeJfwVM}qMp})tC(APsZ=d+bHto9ZTL7*J`J}zp zf~z~kfA;fUwV{A8*|XA(WT)$#L?p|#Hx`Ih>IJrZdGx_l!?=YqCywm{F`8zpuUNL=|NTRx<_klw(I z$JhGqZ?}R@jU4WR1pTNh%K*B>B%*{U2f~jvC|07JZ|BYUX&DMMp~N+IWdb`Yj&;gV zzc3>z@+O8yN;ku#2Se)c_VMBe7g_AN+8s80yvTOxn=hf>2ArJK$!P2e%wW`B^VPaT zrafMJ>EDF?>oHAfXAQdi{Oes?f^2_)(kj_5uaV%ow-MZ5d>=H*YX&qHC59~pSFM#u zA6TW{z`#@a5Sb71m-kF2cGEBQ_91!iJ>nKcqX_b}O$U8d?l0A(Y!S+j5cX8H z)c+;-r0@9EJoAbD&TiMb-K`01MQ*IJ)91CJwzQ-89ZLpR!HOZtpOZsm}-n405)D5(QQX1Fg1&Yl{B0yU3 zF<%8Ps}6Q|M>WO$?CCxjF!mUmWCiSiF-tGuk`z z^DfJIM1Tqg#uNoW`s5s*cr zcsuk=Ksf-6c<~Gw)vr1uK3q-$SEMQ|4TL&Vck$Ov3}7$I=EiH`1;N=W325{zD-Vt-p`si`>&H|n!v*j zB#RHW9MeckxD(!;77F3NdYjgcz2Q#(fc-OXpX~*3Bl4`&yji*Nb=fXMkFr;L!TMw$wr%*^C*wHa$Of6K9=2&}I;_f6HhEJH zA8R%p%Yp8GV7~TC<2C(>BPs`Phb*93maOkwM1tXsnngPH0)oLuc?KiD>~dM8@3TYH>$IB9~MksBzOl+l&n z`b`UF$czwO&(T>XH3TWi;W=Dt6CVw&$ypw-#f{se>?n>W(#D1SZ}PF9=40}x{zF*Qu#z1Zz7loYThSM;DN*~an#8+O#h4- zK(WKE?#1NHiw*vZB-PxaQvJcmLxLy*4Vv9iMd4cagd(CIY6jtMp172%YZB%)>j4Xx z%8q$_#-4o@W4`seFxt=I1TDD42n`#P@O;@&oTqZeN7XEqT5j(cVoRQW{CpEHf^dlK z3p-pU`Ue-2#|G1mRXx^l?qz{;Q zQKAP#2Yx%(CAk;`Hiv0`RLf6WYgc*!Yy8#LT~Bob8>*UfL8}BTRA(g$xal=y54% ztt;C0?E6(ga)k~bP8X7S=rW@TD!esv8BUNhQHuJ@?x;vDy>F_HO%~wi9!He*fD5%R6>@t_d+=P5zPcAkXY znXdBanuQTAo;(XhCI;BUP;N245uPFjAVp=dgpI`W;XPnYMu`-S4G*qEN-g*AbkUK_ z`H$roZeyTy%1}Q(&L82L(cN_~qg*9yVy9%{_j5OS0NeSq72~_#RBcJcFI;G-RRFSE<0hq52>HhNFlP zJ!TcxU6Y!G_9a-xwBJu zgk&HQNZ(lM4^**=h>fL_SRsSD)Afvgm+WNbK(KxA9~s9wxkCsMFIOlpBCl&y9KL4u z)1fmV^Jp5P|J;pG~?idg^zQGxhgBPT(|OLIm0n^5%qd2#k3Lux`{IE;y#r zf5R@CuNR+CLBUq`)~UrQ_{B98G=cyw9ekhi-J%Dl_wzsi<6>_XQ30ceV#%!T8sy2ip zYW!GOK`F?gkp{M2OLWVD`ajJFW^J>-${S-k8}!YAR0`au-ql!*L`hJ0MR4lt{C*q= z+=#I^#h6YzN7UCS#h!c3T`#qhD0X@G{?bH2Mi5jQtw39lRq$-#p;$g0^n`{B=f89P3zwRO9wk^ zoJM*9e6t}sk$fSqOIzCIaJyo_dPiChH0l5fPfBiIK#01p7Uv0p}%W5p%xOs%?b4t93YT>huhm zn0{&e)%&Ibx&!C`?_vuDY(fqRMc7|F*9d->Li!X z=s))3{o{{ducA{1BAW%yLl}AQ#n@N~Vn8L(RPjWPRv+HC#vnRJcDRQvAR8-ni^M$zn{w@sBkjn$rrSJyWhc8~gi zt}T!NXHQZYCK@=~uyL0SboIaaPMG=3l_EG#=p}1k4K~CewC!GWATj7yxAq6x$Ao~x zYa$eayvcT2}i z&oF@94T)(U10HOgv$ixqnZl#+(`Geehpd`E-c3a>g&~;fAilf0zYy>MM-ayT*2SF4 zsz|4KgU|-K@TEVxEq5{?FpUz4LXMyRk+!~kx~LfsR~Esb3qlM-bcO8RI1yf>o4wie zhbpdnZ72JWL(tumAF=5c5{6MOcPY!3j~bd?D$DNysYvoj9KUM9P_6)W? z`M`)reoeJc^a2bRjT?~Zn%jZm=C9L8dk2os>gHc~4Y>jK)`qU)Gcn_~mFRo8G7UAB zTB^fyY}WFrq}Qa)xjJKTq@y5Dxu$b7c#B|OG8p#GL*O2M1Lb#{bNl~t4mq==UX7DSli96r@0wnW_>5K#C`iU@ixd5A+q(pW#$KKV2%?!j<^k%5uNlGb z8-sTdL`Q-WdmvGy;NW!8`-^;cSUgD}>x{NwjlmyZ`&0|FgPvl9(C_=CgfR@VYc+b8 z(2T;Ih=qaWH92va&TPRdXGA`#2a~w83w7OhIRFiOOIf*-ZlR7Grb?vgm0Y?2jxJ!9 z#qn-sj5*Z%c*5GJ9Nl z{)^sR=p$&LbT>^)4-QJ`4Qd!U6)=OrI;G3MA9m-Mjh5!>x%^S7HfXHHr1_dXNcs8f z*1UxE6ydz2au>LN0889URs2EsH`W>GhwBL;6ZD59L=fRiXae_hlY+!}M&iFmG!nXYI+>(5-O=^xn;6k!GZ#ye8+c%T0=ludDtEVy?;p%?6EeM;* z6=kkkdQ)*He>8!%oFJHYPqoYsOtK7^}aJ(Gm(B<^8s?!fX~OXo2x9Bow>H%?5JMu zueWDE4+i`mjVF|ks-7>;xbpi{gFCEqEzoYpH%a*6H*JB(O@WrUug{0SLCJhdyR+)X zQ5Qt+)A1k7GW~h-lJZ^>#0h+W z*B{rkSNy?ye3&!GGEACh7MqdFn|6ESTG{N*fIkUV+oGag4!{-)M*F07N0cjPO|Rfx zdLO|e07GDdj0D@Gh=*B;)lBlpE}Wxmd?A)lcs$nv%sp0=;bN&bpFy3F2v`y6QYs)_ zf1n9kqn?n-JzbavwkkoRdSL@L?K7R*ey*o+yzm#j#!QP@s@YHIm?J&;HOCM)X|y=O zb2KaEG208XOo%4A9t0#}$h7ZdITPN7Iy{W_0h@fX=o0DNL7&RU=igGqSrc{QLk}KH z>sBrU7CY|J9;@(TRRA{{vv~nyh@73!@Y<^Mi-GCFE{)EZq)EJQG(Dz2nm%g5rY$Oy zBs|*r&ss@bh%^%Vi0#**xQR)p(|_>Wd~b%N8!;pm-P#j_U7sDI#FD&F*P8eyZHOQR z15|dxxnyWey2v&1ObRv})vq%D^?rK2K}Hk^jp6+3C+F2eeL08rMIrg@a*ZWv__C8~ znej}u8CDC1%Mb5!G}I|BB!*3s5NbT6kK7+P@oIF;aaLeh6GX+Qe`#_oHEb|A&n4G; zs(G5R!WFvixSzeSSj>c|l_V-5BO3d!3BNeBohXAexFTm@UxR1L!Ou;Pr~=+D#Ehy( zRB}6ry~d=9lw?&4>!Bmm-!7q~|8XmSo&!yRyluded0O0#i&WIw}_?v)ovCnUALbPy{Iv2DYZJRtku`+t~hPgwAto zJR-KJJE)cKi7G=jnP$(U#Mlb17H^3`qHpIWX9s8d z`!ldtGGdpd4|#EytFvBBV+?d18c<@xNTUcXPMZ3U-zh+B?RRd-c@0gxmm*LKGs|)Lc!v7(TjsClJcOn}rJIH3ZUI742jg<}u zRl3GnyUa}Al`aqeGBO5LoNh(ZA3Y>1_4eFF!#(_zJ)bIeh6pnIyM$N3+qA! zkPa`rIjX*6G}-z__Qc#pDCi|R&)NdI$I`bJKTJ3d`p@tL3fZMaY_dP}SjZV)7ul@S zp;5oV7Z<4c&T8-tK@QP{Z`>9G{!svH>|De%Xf>>U&99*DF8!5!qZ zksr=2qmXEy#W`v7e|1>W?VMj#$pjyrm;0M*2?7Wr(RYPO3g;KMG0Wc2my+%TyeU9C zHONY1MUfYnv1$N42^56iH%NFzq-2Xmd$F=H(0+{^`yns{IgLGR!pb(#C(QHEnr{Rg zpb+H(NkHSN0#RGp=7cBaRg}6?jG8J_V8GrwJ9c?3>uvKxH{!HX*u&z{fm7aK5craH zG4`$uw++&d&WT?)`V!o(+xg*Kyys_3=0EXlD#kvIoKpd6qWWJTXM}jdvN<4&N;QXh z!C|mau|_*Ot8-qrzo&1xqI&Y)YRx}~W~IPU=eY;Sl4P!3A#YNl@2gVdp3f^7ES;C6 zemcPszt+M0d}TEo46-DW&;hx)fp_iRnEH%j9!Jzdy*dgHcf8cf+_o#J40}*5*@i(9 zA+KwX6uN+S-Qr!jyQR~Bb8*W{+qwh61zP)L6cYpoP70d!Ug@mGXgYXqafZ$z4Q3>1 z`2U9W9uXddDyJJ_V&KeaEOPIphM5N>=6a^x-~ z$wj?k2tDqUiH}lY#r5GGvp&LC@S98)@S{F-59uMj;`?5%g+;2Pb`RhmLo5k>@dLVR z$~SmCUGU?W2}#7RJI&^?w7i(On~eb#vtR{6(E&~gR0aJ7!zLOHqom*lYI?Z&}II++0SJ9O%MTWG5`q{ulQHR-_;zxI9 z4Po2&Ef9_#UUAvY6t8#4M|pr4;tNH{YBADYeJ;Y`VtnwK3EDIsfN?i?c9ru3E>6^J zl69z==OvsZ%pO0P6R%A~3Q*dfWKC$kev!&YnQbs7_I-t6(s)e@U68r}m!{qr)WNJ?=`0FcFv7Fu?LZwYS%=`P?|)BywKH^VSt zvaKoX-v{Y!jgr&|`gQLaOp12LN{M4^RH@{m-RYIT^FT}TM3FiD-!GzQEq44iAzRO0 zrkSoSYdXkPoXZkD^9#?gU3h8%>v8g|Va(9+Np8i}oKl zm#A99FWmGm?MnC{^$5afO?%$5(~5LTzapiDa}PYYUjp{LON{~RV!cO$XU`pnGar<1 z5H2_3xC`|O3-rZpCtQwX?v)WLxrxRUHmG+N)xWoC+r&IAWzA@`$C=QD&FnEq-{=na zw*v;8DiI~On;r0);s=K72v$JGMEn1mKPefiEG4xun)K8_1(GHhdMgXmtp|=O8Yu2 zVeZ{`I2XarGFWF+?bgm$RaNC+5!+X|cWKJONQDk?~qHMzKYH05{ zO7ixp2+?`YWu5@puq)VLEECh*E4*e7w|=dnW~IsLF2Ls@RsN0Y9t5!1T11 zU|mvH9Ml&hypDU1LPQx6PBt&EW8(#WXisuiptSWT#4-DrR+3O3`nEMOcvG2t7+Rgw zvI+NteJ}&2QE&d)#f`q5DUoE3Dx9pgZwt|d^ri?lcpDndp)1jjnYqK7-#L+^8TTI0 zo8zhx6}l|M(^hE>R{p-|&k!^mr*cxAKnu$=yhRE+F$&zGrtx`01x4-%F25t`M>BPt7L%0tAml8hcJitmW<<;4c2Zpyn5#al9#uEfhN&_L3VEuBEDCKx?z_!ZXIO>G zon&cSC87=4(t+t0IwSc$J7yyH;&iPtuppxIgM#s&9J@{g?Znrnj5^9c9#X!++%7~D z*){FFB8ECs(aPp_Iu9=yGpQ=;mX4?ke{lfd)g%6?WS@#mIeZaMUvco~;}vVP^Gxd# zY}L0TcOHCa{l#p26zgXEx3SJj(TqE;Un zgZh*E4JSUmy+rTRnH34s1EM$_n4M?bJfup0Yjb1_FXvlxXSuy-V6X^G|Ib5fLTLNZqM%KhY%P_`k4KFVT7A!SilIl;6&6WhEN`&4z(8;70^!)znUR-cZzvlZ#tv2nf@ zkv_Al$K8e2?wowgc(B%WnRoA05>dwFY!Z}UW7AG2DNIbNT(x4BN`v1+vC{&I=?QsU zt=JYO+t0c%*^n8Y9*4s1VYPra8i8WWnQ{(z^+8;=lDP~ARgc&;kEdfvON|#b)^jqt zowrT8DK`6@ni9LS@W@3Koiy0309h40t20LY1mSs%dn3L?(jN{s43XD5YcZN%nTwHz z-I~^78+{{W>-7K@+W`B_Pv5#KD+^Sswde6R7*67%Dx;4@eAU;Ax!q+OVlFpUg&|{ElmmFhR(6Y%iIq zq}Q`8YuU}P88OTkNmbrS;D?^p|QU-oWp%;qkE%@}zP1%dLPS8;; z%whREth>oMG^uH>8|#2KyW|ZU9d;l@7bLJY{%|MLaV_O$zgs&2$#3pH8yO@d?34|) zD2!W}VS?wgb?P34#F7m&Ec7-SV^R`;lPC~^iDX)_2LOGpr{vYi@fiPMScg&vxa~tH zBe$CEM&fwPEtN#Of&r{(Qps}Wo%cXg?QYb^KRf{gl;e%##00QXQ(3&$FF8$dkv{S1 z@hO7E1xHAuw&_wvv=VbeG8;LJVT(yLqO9aOYs`;>O=PZUsySp-*J<}Fq+};il+q@M zGDTaxRC4!FM?YYP|LvFiKj7|cJ9f!CG>&WMLD1-h-4PqA<4*>Me7D3x0hU7D!mK?mN``Gu>$9@pt{69d9P1&Ehu{)Tm)qx9acc@Xc+U ze)#C}*2aBP$_bBLR)*Rl^3>OG70u23cj7oobT{=Lt#9!9czzcO;P393neLeBJxZ^3 z@o25}zU%c+iX0z*oHoBUzU2q@(TDn!iY|8J%r?4j(6&I+n+9}$)P6iULZznBAc5i% zev?>mK{@6@+U7-W?+-TINcq32x1AU#A<4NHUXxozjAFKC$&;aSC$dP!n9xU{tUM9k zi66gV8yYcHnqj>GSO~R$5i1H@TA7zjd*(Hn_ZqCftb0=v|3Qyga!=p?)8+k}97(-; zx9&N|{GefoUnE%>G_oTH33)>kxTk9fx_r!S!6MK@MK0>{I*H=cpee>Vk}j2Z$SS@m-+0rX|*N!Uf9y~!-=ehM zxM(}E=#r|s-RPj-*Jx{K%yR zS^EZZ7@Sok;MdNKA;d;tVkP3OwG0sTLvpm>q0lA_ggq1$RtA7uElqcfs@&z;*B1yq zGrC}`=B_rMh=;PUVfT#ESWb+|$lA?$j`b-&Jfar=7SFBWNWuT_`7$;$zhC+dY{Qs+g=(}%VvV)inTC_TSDXHwUne+7sphL=snF~kz!d(N)S4OSR~KqXE5Bi)+kCpV&WLm| z*KBBbQ8Mp(l%gi<`92=z1V6YIPtJiUlzD(S)?Prv*@HZ;dv6wg3S;3RiM3_jqDS4i zUORdb*BV?_i7h$Y3+3Y+Zrsw&pJhrnavQ^XoOV{99JUyml&H`}L#hv4(}ZgyD0C%u#<0Ypr{)2zgyQox)G zd+4ujSnJseHFQ(l?yhtRJpt=*$=>srUHcyD{fM!CR13jDu>;)j_kj@oTNOwMWf7QK z(xfNUS}Syg!~$6U-v`QxYCLx9jEGzBG^|a5s(0?* zzybeRgIr;DDlZYmU5ax?zV$ub4q_87NGaijkv*=j+ewO#N9BTjUBeg$V~!IXALjQ_ zDnE0-mrp&?@@r#6Dmw8i23Nh(H5+hp;gdHaW+}#b2(Gj2-#*;hinohV06uS{Xd~p-=QGoZbW7P=bUKEB!^Rt1uKk-9$1hFpfQ0-a zO=;SGfn!F=?!$@^ay{&6b9hbUg zDUMZ%VW{8gsq516Bu47=MZ2-vLSI&7>q0G92A98ny@wDt@AIMYhM?9;(4plw&S&B^ z5*Nxns>wl1N-vQv$DvC*#-?Pa^e3?ebO!SBg#0{Dcj!`^sqJP_a;LY2p`wCfO+>7o z?~W5K5YBJU8?$OS2i)cTVPB0sll5v?_fA_5%Ux1UYf+jB8AID@e5!D)Fih@ku7JUj zS)HYwac;?J%o1BCm$W+H`p038VU@@!^1Cl77mc@m*jkyw7?RtjVjKZrgWcF@vZSJc zN?ns?LE=l-0lH~LZ%7!8b)-?-=u#@h{ZYVN0EBF=C5Gx11gMogGTad?^iYE9G4w6ks#tZL@J2S=jLQ z=syObn@39SopM)ap`VpYmt+mNVKE+`@#GaFesQzg8&@wADWbG;b#Eis!^u?(s~5Iz z>T}2yiAkvwmMYAZ))0irg`~`HZN+PEsW&!Cb~YI^NpD5Ilm|BGAoX2^#TZ1TL{@Wz zW*-~}je|`r4b5C3zob{xU$6A#MG2V6#Ps&5`CU=5gQXzs+(lq1FW>;la~C&1y}Nw_ z!KjrS#rKM%gb;OrHWPJ#4T}R9|7&VzYK%thvm#ExjqFkP+z_RZ3fO|apdT1C zt~+6a-G3(=+USt3ZbHNLqia??E(*#1Cg(}Lt5pu58Z@2qkTmps*IVjJyw0AF{u4Dz z8#{8g9q;16zUn2gHdJ`wZfu@xGZ3l&{gPG0@GYl54mfVM@_>tHyEj6QzjyXQBN2p@8%>@H`z)$1zy2pJzO)^N?|fY_q{*7;8J?Z&TzV|+K5&#LXlLb*5u zYvt_$ZA}`~;Yu#vF~w;%FcrPfUe$R4)EquniMvB)&!zs^5JtmKNSbd6YbPy(ni@^s zM1(0O2GEEULY@?ue!1kqTLw95IwYJg3%8HR7azc--M0Y66%pg%(g;rAQ!{UhPPUgw zZf^nc1_l_K>hV*Ho2;9vQ+I(q;xRk<;Y;>RbYSjTKo0Jp zREDLE!1C=2OfR>jeBoB?O*YO8<^>=aQXPKd0ptsM38Ogg9V9!ax5-gXnPBqF&9#}~ z#Yth)@WA;{+2LfIy;jFMgCD1I4M|HIV!KAa|854-hs}+TN(rF?3^*s*qXGvErs=T3 zDmV7=sCHf^hSF0|d^IU4G$1FyCSZ%AJ;r-sRoIlH91{zb*~9fthemhn(PZ+LB%=9$ z_Q@1-ga;v1qFkv7@8w{|Z{JZ+jfzZiS!axXi6K-1j$~Lz)h=Qz!m&GgBB(lchHu-L zYmTeLZo-M1sr&dn5_Eslhj^7I7*c19#eTJ%>CUN+Cpkhq5ueG{{tGaVc-wQcm{1ZI zpWdCCdsSBKODS?Z$39~_f`5d4Z){Ac1pF{YgP?FeHbba9hG}~X%Gs;9W`Wm2_g^Tz zk0ybG$)c)0Z0@1zOY-Km%}>oBjPjO#Dz**sZ;mZL2)$^I``}`0U`(KiQmT)Jvm0F8Ma z?l)bdX6QLvKp8Q&$mUgBf0;r!ABUA9Geey2A}c!aV@YqUee8S_u?p4u&V?>oLApo# zx!gdpT7n~gXLGAq?m!GDjJnHgc?SvBWb1hAquWqxRg>T`;RfSV(O%HLocl60RpayQ zPW2;(aR?hdcrn|R?h&8(9W8}7;7kfcZH=F=WBn~|I!?4G`pHbmrS@FS`^f2Qxqgj{ z4aVgw#uK}hX6oCnfr&mph!J*%Yfbo4=o5EMaq7HQ;nu7#270h0cB7&?yQ5TkpZsT_CjcVyecaFzoh93GVPoFz8^3 z;Puw5LQz@u%lezqX75bsg@vE?S8Y6cIIXxE3Yi1Q^l^_l2b)bGMw2-(f+-#rNf(If znB=$v^XqFh_nY$+nRTH>9P%ARMuMObT8al}dRQ1t(|YI`Vol=>072r2U)kfuz|2t! zLeQOG>J~$frqR;`B}~k^;6vBZrl(fsScS2QxVG3yUc4BJFjCv@&88G)i}rfUmIO(y zvG=7r?qJC;&2g+(gk#t3xxJ5@(Qbt}yMKC+p1Ed_>dTSN^PH+Kh``RUh?;9M<)<-* z`T*{iYUAq1kQI=QT<@lK%xb+7-o^RxHZEU2=hzs3>xr5i1`RF@VEZp#m`h!&AC}p; zPhY^GcZAz`RxeJA3r(ZvTAl20lo3xT4S_{!okzYpDYe7X^7Wq_Ca@3|Rx*|*CAabc zWV2vUR=v&QU)VVoWktv^fFk6oxFt0uwJ~BWYMV5v(lJbBDqSw`ifR#K(Lxilh5VuAr(C^MFQ!U*0X!h_UhrpCN{qRLP~WmGD%P>`A`fe6fw^c(XGRwqUzo=h7yQ!-m{|*)d-#OqBlEoy!dXUE73il( z%!T)-6tOUsgg7kVP4`FguyZ?u$VbuM^erNfRHBp=`LR9H#Da%75yh+;2)2-i@?}}M z^)y16StA9@72D9Wl(M#Fn1VQF5c&YzA%-Dx7)$dp1x%Km{Wz?oHaQ#x(Nx0w2}(b? zQoX+@7RHf=G`QI1bJ+5keJisL{EA;%%40*Qox8KXDorn@#*WrYO=?JW# zq?VOiyZ+|V)+nO%5i5J!MmI&sjf>5N-K9}dyfV~P@M>}(r4iKb;5A4DZpV~`5E zCyY?;CfzB>34RGyLlG;hI^7>Q;rwlb0t4*aB=mZrJeZMaEztONB2hcAhQEcb{`CFN z6;V`eA#QX-u4tr z!Z4|C%i!X68d4tApg%`RX#^h%zCSSHozbYMWc(INlF!yMvz-D|4N>tA!Au24r|{TT zQ3=)J4lUwbUY^No)G=5+JK|DC5!{x!J8*ntlN5}`cyCH9hXx zW|lTjZoc=%E>Aj#h7Bi1yb)UloD{o|c}0@4~G>t84Llc(;U%sk)6@29V1 z{NMM!p3JRjd@a1)&zH@Q&++-II5Itu`16BO_c=uEBm^SwW$BP9*WKRlm%rJGhC)^^ z_Rg|he3*UT4Kx7|&cmQ}SDFwM_y&%2VKo3>_I!XsE#>9ACxH_&*-R{zAQVYq4#9r2 zLeKs9eSZJwh3u!dTlbgYor4-dM%1>v>s44>>M7C>j?EH&STU z&7zzYah0{JgR>JyPbQx3h~(*^;&{*Z$9@99+S~Q!Xg6TEy6V@kx%zYSwEWiMJH`I9 zI%er@`1<-W!S)6Qyx7yH1$7`v$1WWA{h5U$d;bg^+o9a)`%L=d)XUzi$=g?dh2*t;I56ugSt>L3*}zQo%#u={!Wq&T77APPm4K-)Hzh|t%%YL{8ackg zOpfNFxvH|DYO1WLjCGA#U^5et`2YZqBcTxFHW}~9TYRvJecc8#W|Hx+ z+z{~zXRB_vo`y7-cwCM&34A@_H@JP^J0H!xm+7M)kJQby-yJb?q0pgeI=snyvK~ol zWII>L-Ni%}vcmVi^u#gcVwB|GFjue)h00IcGZ+NNl>(W(n;OO~uJ4yiO2Nrgo*ph* z9|QVd;)<5%nTmr@Gr6SXRB;z8)Jwrst1}wPkE!|aBh8cNqwVln;o)nF#zOTi;{q1O z&^#O*hBEOoyVJ8erxVAQAFm!>+yiR8SX(^+k=kDj1f z6hDo4c4GeY*lck|^pyP6G?3-X-UI5kO9gx952>y)^spZt71O5=$WGkc@p7^Sv|6EH zEj4^;#xH=LcS?q8Z;cUr5}K8Lg+`$HAT|h%+AM2@rfB>a=kNSjV$DgT3FimeBS-bw zJ{HMl^5pq@w*|XFvPaAzwoN%Bz=x|3f4YvWF$S=1T@<>k0T~@w`T6K6Ya=F?=jqfn zN}E)luI%r)dp!&1SOqARHblw1)sS)MO?i;XFV14g@io1FDo@v}2b#CAt2IX**nTF90I05hbXNvPbd zVh5;?KFPRZcB%mWzYsw!twaLv6p(DkhIz&W^9NBihJ@Gwa|fuiJ~S^pCXfnML=;3L zFA*6c(d>EO532=4pY7g^V!;>NFp+C^Af0Dv0u;eAJ7E**R;MnNIq}Uk0n`an`P;;n znXJ@Sy%pgN(f3IW*s|XR0K`1}$gso&sB*OZcwhPlIp(TIxlZltSR}tcWRulNGFCnx zfBLn)vnDJrVni1;G_-o@$2Jz?mOHa5eQx)A_lMWrhL-voSS8O!Pup+NY)1;rD9lfl zOhLsb%#_B==Ra+X%o8WHmin>@!$_3oM;t7)DoVV0=+a&*G}2H+07<`O!q7DB%#;v+ zcRy8F%ow!hwotYDj~fn9He-rORdYqSy6J!3*jJa>cCW)k`snv(K;?xQpf48Xrv1Vy zgMN~2j)jSs1@kKQl7JTb6Yx0?N?dJx1|7Q$w+Yr>LQ;T??=JS&RfbQTI`KB3v$(zSQfq7CTp+vo??y3gksUWa;NYqdu?M`GxF@Dhx6*A|f)aiFusFCFCPOq~}hhaO)fB}~`3lMRLfEt>^;%Cmg>JuMY z#-LbMFAzpaDj=(gAjC4AS6&y5iKxdCWM1jw4)y$p#9=HEyvd@4NXbOne`TcwGt$Cj z(kJkD0`p%C6=-h_&+ZUdXNbKqVZc)TC4;?jfjI02rh`G~FZB<7SPo{SnIB~n`?Mr- zOpdms7lE`B7r>ibC=Op`Ul2nePU0(IV)P8zBLFGuFP^@RP8L9hHp4{D#M{H;OZ-e{Gy^J7_s8{AncyF!Y#D2PI5HOn?@#>T z_;S+*r=;~nSlHtOK`}&3vRZs(B#OR248v^y61=eMei5}_(I&{iJWljw(Q>fBlFiFo z5gP{s`hiA7uW1`>Ef@3hwJ3eOmEcrHag~`Ag3^91d}0Qn2vX499W~Mrd4$q}Pv6v~ zT8q4slW5H@?eAag?!wJ2@@Ovx5}xNXeFhS0*FEzDbP;uZx@N6&lU~e=GvMQGG%l9d zz=#=a(3%!W6&`p9`$I1Ft|YZI>p1>uj?#W) zOQgD}J9jOZ8;TxF128{k!{plZtm9opLwnTq_I2Ooe$PmQn=Xx z*dSp1GGCfeVG~q=d2XR+MmiR7H%(TnIL2LjHGldSp`vG-bm>`dVxGlkJc|FB&=BP;gaEuvQ+REBSN^ zi1yDNolLlx0*H*6i!N0ScD|y%5oiHoTWXVYH#Q#wzTVJ6iTIj~Tq`TbQnt?&O=%pr z;b5s)@|$AO&JYN;=gymtCZ*9vV#*c#RDAM9F6?`I-2SRchu3Ro)$DZ zkLny>qZ*}6A`N4TEGY+ruNUCdHrZ(DKyqv(&Ukps|h6ROf?0%QP_%E_ea2d&QU4w!bQjQ0;jOXFgnSy}-kp zXqIsm4X(&8^o%8`>s$+1OmIt0X zI{m!svK}o2-Cpn4qlys%zsc&7r1=XEdFb14@_f&8Jfz>LxdpGrfgfP~uenUV8K4#z zgiblMOzb52m_7lbT=#eg_sD;WL)T>lZPZ}N9#!j$O@DX495$^_O#O4MnJCn&M)}<> zCCa1fxz8iXzQXSIdwvmMc0agJAns-Y2u~k&rYSmUinZjhzA+0G6y{b)TuReo;V53B zUsX1PdA~Rnn8I0;?-@AI5@V}`NH$w^7Jf_AgTITD z;Brm{V+gK#Jg);p|LT6XeRwp+Iff+&bD(asapdZJLrYe|24@slk<8O-l5Dh&AG1bL zN;OlGVHrdeM7xVZMmyq>LPe2wdXfZRG|fh>m*HYm2nzBc3p*{s1TIN&c#A}6Fqo1` z?!^Yj2X;!{!3M|B;`=p^8TNe6tY5TpvW#=h(I8$Bf_83;`O#i6W^!a@s2c2Iwp}J; z)~Qx8_8{}x*bH;WEleQUm2n>IZ5lPuX|u742s=-m!s^qYiDl2;pq|IfPL;vByl#`M zhy#uSP#TG(7q(UFpQQ-NUuL*Avra*$L)Ps@d0H@_;}*XcYvzAx%{Vj*E-_m&l;tkA zhzPoRlHSw>n(;x4E-{1xwcTZD7Fd+1=-iLBhk1|KVucELah*n>$oX z(?u;!_R^N=z`H@~JZ-0AmU;cy9kXvQa_xCh|Y{M?*L2d@e{_$x#81b}QMWV7S!ymQIDzi;5 z-nDHHM>n8*A``b)M3|_%TxH5RD39=}dE8@RaixF6MiNdb(>w(#{CNSH&!jU;42d8F z_ob@S0Q1W+8@4>I+7VSoP~1ko_u}Zt4UNJ^fSWvE)MXDi`pasF+GRO#R}zjHcHGGui0GWGvIKHh zT)qk?fMD&*#X*W=i({fKkR5!Rx*;@~z0@YRQ>yZ^9aev>8t1U?-Sc_Wt=$ds7lAOw z5eaP$NGhxZ4le6GP%=RdK>&Rc|N106=6T@Orlaig+Ew}I&Yq`y5)O6?Ah&gweS5z{ zA2?DO0scrdEEnkl#C6Hz-Tz-PE_1|E*Ul&u8H;@Y3jR!Yr=>35_%&}4-M zmU6|q09H>ZGK`UzrEo{Uuqs0T;+B|{PpT>cNm_ONnBm`-h@Y^nmhgberIJuH#RRmRq#fUQ_*XnxP9#GgGqgk$aa zQDr%8wXe(WESEk3*?5b9^Olw6)GO9_+*JFH%E3~G%WtQ`;k{K-c;k_e(G-HSMFmv4 zS}pkdUJiH*c@&%BP8Ci88Hy1Ok5XA3NV`7^3|PCrYry5NDf9{eJN_3aa}g)p2&RiR zr*>S2RsU)`{@V)|=`{?e_FoPJabY^yB^OCOk_~%-1*NKVK@hPQSs{&kj~%wjMR~;* z-IKNkrqL8KQ>!J_CNHnwcAfk$Z-(x#6R{oZOAX$VL@g|(s$hg}ionk+A@TmrI#D9N|@Khsj=SgSo!a8A`eUM>NhgUIw7 z<)21>#jhQ5Cw?9YXmX>-Wc{RnY*RjX#7mstsUj7(o64yX`p+swm;IUM)xIq(f*eaL zq!0094SHmjHjvRacu4^KE+yvkG-l<5wSt)GAO{Mk&O{iMg;rTXjcshq@ zJnl6Rl`}V@6z^x~HcvE~gbY#Zi{uaF2O#(Jcg(xDE&;5VE?i%79{;l!^s8)}70Vx} zJO=d(WAR6KCDdGO+*ad&wQk>CocF;4oRh!vckI0wg7Oix*H$>K}Vn7J+4bbaekay#jl+wb?Q<@Tl~Kq}tvGg|7n#MLT~6Bt`Ak zx!U*lDomx{0xH-zo&jSXCdo&!+8B`6e1O-x!Cu|I-X&99Yu%y$DV*$Gc_sct1=`@G z*o^mBYeSf=voo&Ob#q3VTQM(~f;(YPUjs}^#6>oyUD4XKXYKW^6{_+Tc?K zn{!QOY*JWG*c1M{Y67vzmMZ8bYc`{+%y?3znJ)N!5St8ed-_m@^YozXIm)%1hSSC< z)8c?q^>W;dDRgTND0^zU#nW}|;H$CnIXO+E|9zF_@_M`3e+|g)dOMW^cWwzXrp~k& z8;Mjs?x9yN{3TBx$-E4A=gX&QYnIS@R0LT+yhO>9c^aHIAEW~mo#b)jg9K2ELCUwf zFk>&Xqt?S33);timOmBG+EC9QCFt^Q?0kNknmv#^+E#OtnYozxwR4n>Iwt<19<{u9 zDzRne$1VT+VrSINW0b&>{u*ntIgTF?a_BnOF^>?B1uSvuGK2A13Q z>F6BdNBW7N>Pne0_FHU+linyoCF<3ld)S*(cZiKAb3ZL}%3)G2`Nxi1%!z2sz$lO9zQrtgT1KP{XZhknNvFOZ)?sq$sKw0O z=_ht-IkSdV&z!VhTt%+p<_RgEYO-H!v`=KaC6E|})4_5i-8p-}2sNRfe$aoUn!4|P z72iBn#`ry2OFwzy;$N=pnPRLv<{}&;l*)tDxp8$cPBA#|f0K>Wt8-?)3@;CGnX2;P z-M^Z@e!7+;gaY2YC>0)k9o(jJydS*AtnPT*dCY2iB+_qXl}Qyk;VXBb8i$8hCa1Rj ziWh-BE@>}Bvs15f2mRt+&8pc~Y2TergyfDY$z>tI0(ky~Rvxw<{^uJ*YSOpBw9dpV z5hjzuUBoShAes2{No1Ik@DD(%ZvlKAxhTp5fmlWthcOaaxP>Jh8qchHZQ@i(zPNaP zburz}bvpV=ICy-u(++`=2OrAkI+^nbJ?r(L=tkZCG;ADVxvQnrknT2BWMx%r>T^i@FWi zs8q;0=&bz5S^g*bj}yIK$FFSzrGGW<_0MeT0^MlupT|O?ixVLzhL5|1{*$4`qna zoC#S+s<8&GU>~OaM3$*S)~F{atk;R-wK`(1heD6CvtHPX9D_OYC0$epMMki-@MWej z|8YP4&yu(AlBdnD)5s-3&vMs(2mlB^+RVa2IIaj;2cE6}ILlQfYn2~ZJ6~%QR)ec0 zPv_4+a`Cde(eqO8J1OdG2)4jOnTk^tgu1r8bK12TG?zvTVfBT@Ja}D2_3$PKQC+&V zAv70Jk1hB!@bp2)JR0c=2*v?e>?J?cP$ZNG$I&g@_qkKNs;U44SG#M8)KIPj6$4(af(8J@^2{G(1YH9 zY1FSVuQA#?0O1t^n`t-18+0;G0PNU*TtSEZSxAdz*UH~YQAb*598K7oUIV*0O4S@h zHc_6i7)CbRQG%g7Lz)QydmkTzY1OsV(lprE+elMIS2f&EA40E4+oppLcI)2ZLYQ^4)r+Bxz;AZjDKbHUgY2U?4lTd!>9A!sj-oFiGFun zvriLbr#Deq<91H01Ee48wkwV3sJRW#uYN$4l;zha-^)_Xb!D2Od4YfSER_0QZW!^J zi}%h@0#uRpWKG0X zxSM$@DBn9Y6U6x(^#S)|oPVmE7@$UwB_&X;2ZT>vpnqm?!F~r-pnb5S@YKMRsY)Z3 zEy|Rr^$`WNYxx^pmNEM5@!J}O1*1>QStpYdVY=jxQSzxQA_f6Z=%aruyK=e0by;6+ZIeB;N zkMNOKb#QS+_+?A~*4f8hNW#xwZ`^br`~Cu|!M|qzpP!1&1ODU#;QX&EM?MVS1 z_+tTH3rJ2nnHwZ-km*IdLC}37CZjO4ME1ynf5I6_8X6HDmbqK8fryP>E)x*}tsa)m zl(vcOQbDePMSxThsyGy6nUIC!!$GMoA~rzP+2?9OY{Vu-K~c!8maiB*G$=~v><|hF z|K$)B0qw9#m4$L%_dY0DOJ8)Jp;qy@wR6yD*&5PZ9pa>;l;LP@62Yv2QjD;UcmCle zqyH)XuNr!Q)QFiWK1FnbJOj$SFDONNf&?8AK)gkajwB7u(U++q=#ESi>IThV1ivd= zL9&Fx|CgcGe?{U2=}UwVNdyc9n>YagMdFIIJAh9d_K)1?9~iMFH3)r|4zxPT!_i;C z2(vts5$h1m#8ur>-aqABHhMPvq%^U7aR8Ai^5j?#^mw`e+8^MR|Lp9fG#_VJBYHOC zPklp8MUA)Zp$(d$%Ov;2TeWfj&>cH-y6bvslb6l)%Af+Ss+b|C5nV_2&uRk*X$sp8G*d2ez+uhJSN=8E1;R6`ot-XA{9u&=c=_RU8HU}y){S|X^WV!`RCF)F z-!G$99;ogKJb7YWSPhSeiStw|Zhdcp?Z<@T2EW)1?xMX=P`DZ0HLitAn681T2jE!wmlWHJ_YlYlwIg5Te%OQ0BY= z(*z?2gxMDsT#MWzer$be_!*BglW=-3smVH^6oNd2c+>$RqL;qHV_mcfdFD)=1lk@v z%+lzrzBQrBx)R-=WS?CCq<`fEwFEvdH5as?18&kxs?%gx@NE!9t1-(Hp|_=iWTTqg zn_t$hi`=WHY&7M3SQ#OiH*gh0mWaQJek}#S4S|em!VYrr^{wEy)7OX_XhSzS^(4KC z{qt808|$QG(X3ui9`_Wrlf5%%3^7<6eMX7^6%|iZ)9zKoiBom~Kx;lQ@Vfia!7SIS z;63RCOIrKuf4j1C_NRyR=KZS=PKVMlq|p19Oy;*e5+q$2xwt zH?ar$n-3_e@k=ub;J$kc6;BRT-OGA9Bl@BQtmsZod4uk{@`>h?G;YW;fvWhEkNX1w zo>cH2WM|ITJSF{TR+E2&iF-)k6MMka`(W!r8gfFpJ@odfL(IXRv67thYE{&esuL58I@z;eGSCILrcqBQBiM-V58 zc?H{lF;e-&+lj-iBay+znEv6f_(J%u<-{p)HAzj4@*Kibyig%wn%ZlKy>g-(u1wH*Rtu;qm8~#a`tA}#E-ksO`VgqfNZ$^|YzgC~=@|INr z8v7>R>UH}Lz|}udc_(=f8qoZ^8>?K1+b9l}x1K`noOZHeeD4<^9ejx(2#sInl1@lx2B5IpYwuQ)|)q&gCNB!*;xGMS6h6`fMvz;zJbCz6sBS1rn#jXc|}Cu z1EMlz;xgo0%ps{D)xMZLcZu&ftP}-ftn-tiwn%_ZE^+S^<#}6hLniKT0$;GWVYO)q z#`lJT(j1`Eu$6|0ivm*a8j{jidoNuEp1RIo1|Tx0AIR2gD3|m=cyJHKDI%H3)Q0IG zBZJ#JfKw#jHG4GJ{m~6!)3$s#PG9)CB4S<5Hnr%j@tr-N4@Z#HLSHtr*$gIWyBTIL zMKS7-_feCE>Mml2?v{Uit3F)~$sQ3$bdKWlHKrBnz2I!(x)&2xHW!7{;Q0S!*Aosu zk2X0(H&eeqlMQSLSAUJj9jFa0Z%%GJL|KG@?? zkeSd1!5^;nbG$2#+?+RitzI6e_k0nPDblpUA%I#A6ybpkR=SZj%181pf_-v2guS# zj`X0-OD&0Ev3(iut`dmOXffV?p7#Rbv!}Y==u||U4=!T=ecyjUWmZTB*b{^#3qVD9 zk<*mZQngLporB`ZM8Yy83=?3Wz$t6EuxG}G^Hih`cwMK>J6Bt71+c~GJ|gN~Qh9nOSFza(dvipGBP(Ni!W64_+ZFt}rIAw=$^qP} zfmDfve^Tr@oOOhkKUTloX=ER7-w{S%NCR?Tw`d)(B$_kQ@GYb$UGC6`RMK@V#wd(5 ze@G$FFVP)-3J{futN6;jlOa0v;T}Lp3BE6BHEcre(wfjHEZ;FSJ79`L{O^* ztpo$-@K~OxH*iaUTGyP2M=H>5%@XmbM@b($qP;(%bMjoS2_Y87PV)`541_nu?9BW3 z3}*Y!4e_vLnc9oGa^z_&Ph{yI6EL67|F@Lr#d%}&2)TJSnf2?eV1)K`r&DwCOAS4X zjWVbB)c0V2sxpE<1XQ@fjr^TxXX)B`8?)uZS>|OQ@P#~(3^lCuauXs!-V8F7pUH2s zI8-F(4Y_&`krFn0l|&Eg8$eb1OafGU-_QZssY^#f>Y3wT{Hmz6eQwIcsX&G(yiz^A zHt+9yIt3LbAs(~*`mRrE269tZIHTilIL=7T<%7)UsCsJa2#|DP#Cz-_8CVl|Ke?Z5 zVU&E2J!~5EU}sOZ8K0KZsG88tuW!58MBX9egY|I(Q+t|yS9&VNVo(s;`t<2E8W_m?o(K z0qUw#F%vcfj+gTZB17az*`nLRvHN2y2}Jg(F-&PibRsS!QDfT0MTeTQwEE{3a1G9K zAe)@@9YjSe4919eWY>@>to7eqP~0>{e)|g5L+QSrP^SJRQ38U=XTTVP6hw%+S0Q8Q z9O7mwT`~9OXFSi*wy@zM^<3BsPH2B*i>5{=ZandzS0ijhiNOO=0&(u-9mg~Q^Y0?< zQu?p+i3?%c*+tUZZRm>>-Q+Oza4>umGWQWWU&Yxh7(eo58efe}NL6Jx8*`@%r$gKi;jWo=!0h()j(JxRX806KQoNf&;gSB?f@#D=tVMU07B*;&HYqftXs z#ft`E4+G%*9cQ(!fOp)(Kl@4yi4{Z;Zs)(l&(hFo5i9Y=fee4X1c5oC^;zdBiRoNB zfeamp%Z;Yv&eBS4gtLRlIr@0*!^Hf+l6F;musGxwY20tAFGUSsFv#)^8LS=tFh6c- zC=XV7bG(y`NoAJB)NHHK5%Z*;llE`VI7cP@64^P4dU0!ya{qv>Q{XR69ujKQ)O0l& z67t9&+EL-24}^HrUTocE`)$EOHRe1ZCTETdo1V(B-)wh50#a%0X^b25(T7g)yl(A? zsu64~qON3DXfJMfR``x-?~-?H8kI6Zuuy>rMO@bEcd-$*Wq8n%uv>BHwDu>Ii-wj= z1m;y(%Oyepuac)S$!wed7MAbS@Nb7yoJIexy02cEq&UKyEAg8b3~JS?mC{PW561mb<>p4C)cqoFO{I(@L>=t zOM}k`_`GblgU3%Ke;2_Bhn3_J`N|YhlHQS?lBi4$jt|x0`(*IR| zz`DPN*hDc=^Z*y;FaBC=O?G@dEXM(f%V2#2wc2IVHpp9oH#k@_+3H}dia#OkLC7!* z68Nv>5;TlGK4@uB_)r4vDJ!Z~=)r-lg^)Aw%W&Qu>E=go@CNq|+Gbb$ zRJrt`+PhxP#XF}^tdpFY7u(ukrqTUYYs={M z>!*w)A9J~E3V(=H{~_f2_{{apN8HyBU{Nd#m=C8ozvS6}@6Z`!IWFmcP{WbL)kjDb7oV&2CF@V8O=ruT=o~BHPL;$H zj)Q9CLqKi#CC^RxCTZ_^^2Gu3<^@UxWf>Z^GDy0BV&L1avEF&Z%T`b6mHchdiaxqp z_sm{e9FB>M=Pi-WSLxtnuN#*Q5jaEISl>6&wX+Wx?rovrZrtn{CTh^(N+Vqn8CaIi z+Q-K;lm&lZiaAX7ag;1kqlWg%T_gsg$ZkTVFCIn@`Ncr!6Q@JKH(XZ}W z*8X;LSQT^5-gc!5-mlna$ZNdgc5AX9!DtrM!>_tL+qA1*> zQDmJKhw}k{eEbf7>xu!YECx4&Z_cRmscL+*PnkGU?2ssvx&l~XpjC>m!xfKZk0XMd zKj~aLGJ{Z%T(U8WHP~z8*tZ zS>r$Z2En*^WUo+EKLL5^7iJFp-Q7P#EG5sM#$)r(|A(i=^9~_`ao0d);HEm*N^1a@ z9y#RA8xZ-c-_!pf)Py0#FmiyOz>*FbXCTHqlUU39-FUFu@Z13U>&KN3NySnvsaPgF z&lSo!FMDjILY-sAVyTf-BD#hO-3Tbb4Pwej6q_j)CriZ8v6k-XvJSsn8YY(SBYuo4 zZq}%b%aGtCu{<=BF7`))Nf(E&^b^!nX%p7428Zn$_A2W-d0OQc)BqVZhw}s1+3zgp z-~P9kfou^x((sfh95_q6)UAGqzF6&B&xY56Y+4&hc#9dnAX()eq@&Rs>J-FBR>@WB zd&!bHpJz-EwAd9fSR14hd7{?!4#gQPC8+|(b2P?qCx45j`bM$Yb#7-?pA2K-CUb7l zhQ=B*ys(qk^~<-c1hDB5!aUz3GT!?6HSX7r(vtoXVPQ9R|iZX?Y`F82D|Bk4*tOMq0HWa9(lgze`)fL74 zb&p<$V`@~AQcxA|Ff25pG*@AP7XJNwD!BY0#xvBD3h^Y-^`wf!Z4it7H#CDN7R7=U zUee5rrT>=|(~3ej7oFC-_(XsW95e@UT%I4TM2}L;G}l2>`+TtHem+ajJ3UN8B8K=q zeoH~vS_&Eq2U6*W>97_#l%ge(lzO9$b9|F%^%eSK!J+QtPm3swN6;UhRAr{pR%EXQ zE=(fkGTS7=nae7N`*eo(L^_4hDu}h;T(!y7camwuA0GHi0c+Wl_3stL7T+cG$^_e2 zeWOqN%d7fKMpTz8bpGklu(HhyT<&&xUqjR+Htr9+3?=XDxT%SBY|dYy--c!k8856^ z@ut$=!s>cq)eP>8 zQkVEy7Pa7+L9(v60XBL z5k~hgZ;bJs4YWVAphwn|NwUUk-B{Z``3_)okoQprA#7IR3D)PZz>8`+geLMd{0cn} ze^j#(#%1EsVjZJFD`ln!&usM6P2NiS5)5w}J(}Gt{=SW?rXJO%se!T__FkHJfRlw_ zf?ku4>QCs?t@iFa#_eU4ch8)1z;&y4w;-H?5SJCse@YIDKJ-zFO5x^}# zvC4%p@{;RkPld|%=fgr8>bdEod-UwxdQD2AbNrzKtX;nu) z_Xk_ElpHb@X}q#75wphn`JIi2K;y=WZy@{1r0%1Z-tqGyd2ptS?8)*`@>A=RZ-hrB z%~Sm4{qk4J?I@v(zyqFUQmtF{a-h@4!*nx{GWF&zszx;LsBe!m%AM}n$MD=%QdQVe zpaLH<8D>Rzkw4EXV4pO}7>+%2Mnme$-y^|sVUck1(K&}-WkP?;QLry5pm(vH?gkBq2bm09%V;81!Fh?7wD=dKHVkV$oz=DMiyyu7I0O=3fg5 zQBk3)DUobGG)kBkEQ5DeoS>dLC|&^q73x}~NDfl^1Pde5x` z3si6O`zJ6HK0;%vj3fkDJOD=a&M7kH#V93Ljl+o@x>OhM#BQYI&iWehv`mFbsz>7g z=o<(dQzW@XK2DCs@gaU^RzSUoM=x32hsFh%RM`FAY@FifH*t!y>Vve46m*%;dgv0F z#A>*e#qj^yZZqz!aI&}lNaEP>9Xt7>_U8@BfsUxx&5f$S;8Q46U;wIY5TPu6k*8(8*uZFuF@639-{&?a?vfT)W|w9x`)eZV_A_~=-6O; zBgx2~IM%-g8^ug?{qFq{Yh$baE5K86Q%#_swV>L%G0xZddkHWw={SU~l?pEz&a7=% z_+B<<+GqA2O^8qmT|O9+%SQU0N_gGu&{}$r|5oxru2!A&Q%r-&oW((CK^L-oIa#aJ zS_zp1$wZzQX&4P0uD)xOoD2ZnZpsjc4V70nQvH!={>7Ui+yWRoy>J>&zWI7M} z`vrD@iA|R$A_M+bjbM!(cj>TKB`+ScLSdW$OY97Q5(8<(cAjU_kv&rZOB2%3s}@0| z93^286PXbU^{}sT6~qhU7FqE3i|6f!@buFOuhWf`TbiKutn*B%sn9%A0*aNV>-$i( z%H(t9!b!=>LuiEOlk2b9C2biXF$a6iqM|Mt@UnZ&_pXN!J%U=Tc4GFYeM6yEUGy&L z@?}HEz_F7iZ9-r=;hu!`kOJs9h1) z@BT6Cqz{-Ol~PXgd7XHFIy#;-vA<@(;DuiwJZ)ho+t7<{Tn8mzD_P|iD&bA^nu$?f zOs~BoS?%ww+@E$*3T}@%JfF@~jL`G)hVLE%NZ8G!@}{9XeC3byK`TNSTA%9sSrdM5 z(Z?JHtTt645{E7PwdmiT57=P>lWn9==FH2dqqn4crRa3v`O4{W&utT84=IW#Koh`~ zwUa0k;Czdlfie=tNI~r;_wx*D)E~b9fiwW2s;b%sndJsLwqb?B4D)ho_sJt~>nH_a zcUY0ih-GzDa!2GTxW}{pA^W1;WN-gqzZi~h)Q)FHN)C2M%8#tH>PS(y&}u%?X>Jf? z-DP$&W)7H6{Gr&g&rVH&^F@rchj9P~h8Ku~6n?<|}!b0jcG9U4p&) zEYB#ES({K^GU8@3+K@5gQ-ihY~ z#^$Q$TTF{f|I8qsd}02Bs`50AHoZ)Dl$oRF*=1N5nK~5NGvIu@zICqd>T$Dm9k>hC z*Z|8g{i?hQ(o=T(>8|KT3YlE2H}mmd?O#;R4=B}eiyfL3moB|9i+n`)m-c=s|2b}t ztU0An9XQ#%(3u#%AX*RDoM+53^1I$#yy(6lEFKUL@P%SrBZpuh6JCK7iiMjNRJ_ev zn#%b9VsMh9mfl1$hVVkZ~;p-k0YJl}2v+olP1 zT9i5`DFD_SA=j7gM3kxd*ieL!I_tJE@EVYGZ*?}fHEZ(wpaL@jv5{BAnYuV0cwi$^ zdlU95Z^w!McZ*2!4Z;C7_-gXifE~nTXvhUIHsm(cXNMRY7#o;Cj5$Fjg8cu#j(D^? z#P0wX=>KJjaLu`R$g5ExpTV1VmkO~owV>9+m|`>^q{MQW-=ln!>Q;5fslE-ua*w3B zk(S16%q;X(MJ|ch-OrLw&x0xymOKllyI+jLuRT-Uy>4Aj6IUHr1}oA8Dp{&WHAU4X zgB|IYUeSqMIp2? z`@h^`Tg1`23bp2^rkE2u>hL|&c&{Mk?|k^>jS-i8yk++nLHYc{0quGE$tzb{gcrnC zwQp-$kwjno)-$dI2*2Vwlgd$a_Wc13c#kK7Oh93&pEg7E=~pQ&OeV zJ0M^gk~jiOT|Jyu;bt@Oa(utPJ3bC}`ZU_ORu-z*mcNpMizW!njmzWgm{~skH(0TD zL22hNKL0$=Hs)zZyXv_eXom+ka01r_9H)13YlHuLem^T82#_4Kg&(f%FxU-#t-EGj z;tqL6bSVS7MEEy$tl30#YOhe1eoRHUyjQ+=U)PQJ?MXTfXHtkr_q|-p;D_;$k|(~8 z%xyFDi1)FNxmbK0@SkFF{)Y9-3>!GtY3~69Sf*tdds)NkZT;tK{l?n(xXGT+M4e!1N>^8`Y!1BzK++UV+QU^{bG*$c!_w) zcFliYOJ^fPG8SO)w2Z)&o)IsuDgsAQCiKo6Z`~jT0*tBFF01u5ia8Wpc)J;C@E-U0 z#?W3^@q~Q;s9$ztm9f9%4D6OM&PWx+VOUn>4(bonmf`mVG~tSsuM7+-4PSz>3zoaP z1glUyRkBBPlulSK2HxPnmGlni6YYlc$BU2_N?2c&gK0}m`LT(co)FdSK zE;0IsIg!?$uRqqW!suULq`Qn1A#^90e~_CNSZ7XTtC-u+WEKQ|e@xyAh zJ~x{h8V5R&Ff^kH$8Nk9m*<G#jj4LAtJJ$$*Is@ui} za3R8IMD5lz&C~W-MsFv#*0?)T6)+daL02colHqEyu&y z=9yD*pfg>%^Jiyy_aBT;U{ZbpbcSWm=vNVVPZvWK$1&B-nM7_=(Hu2-QeER5^GswE1*#kdoH{Xc$3*I?rLcKV?|vmmhI;A z_Uz~*w~q281Ko){d}3DYo~uxl@K}Xav+(H@ zjsz_f7@N`KKO?I`UDJ){^rkMT?GRPH{s`dw>uORentN9Sa+x{^2L(2#`!=hWK(iay z+_uCms4G;fQ7l^)NJWJLv@VADBCx8jeS+yHemOSfALkgteDcY1E^UB{{M2N-K%K|i zWzu_hAAOtO1)FKIxfAA@7n$lj|aiK>lH zXL^F3t1%v+aPfpBOEEB+^~NVh79IKpfX{*(K&cU)DM2V}-HYR4+iODxePBvJ?_r)S zDHq-6p4)f)ZVWVhswqNRG!&Z1NKoSaXj{@56}gYDj|)$VSQC;!QYv2O<@Mh%$}psB z<0(dIz)@u(8*S{~)Gok5PpVaC<1O#%(%O?~kaA?dkzTI%t#6peN#{D}Zb&`(%O~#A z?)t-a$?BkPm1&7v=-cIbY-FV_mKfy~4 zBoo`Eo*Z+JLfP|FC%SqYrs9dolqTq>lq?u_887G)lPee!lPVaI`i2-RIFCF}}i zWu3+u&@oa-$+|9}!=-2TB!-6yUScu#1nZ_7r#c+@UmnGyY2em}w*+x%=$S-nmz%Sx zPY~*Me*~sy@BC)ZmPK1bb)@uebLEE{PjQ~SxNs7HmS&;6yZ7_e%v=)zTC7hjF`DR1c{Are~W|%wDW{WOcD}9H_hk| zcADvl=*Pq~1L($$780HHpQzkIXm!j{k0lx>$vOWt=yl3*$Yt3)&xGFO?pI+rU6yl% z>192c0h$r7H)8+I44X-c6!L_KB3=tT`VpunWA~WUOK@kx0)nHh0yGx@Kedic-y)Hs^R?FOfYgd!}_uq%9q5s1mdm zCYg|!66kInu0yo`rZpkF5P~%RbEdUpP5{zC-=${$Ni$|9^F%EayJux8@Ig^%A@*T% zhx0$ToxbL2x(S;nyvZkGQ&u=Ykr>YlBOK(+G(5jkZ=i%;11%=K8U{{yCIwOjxI delta 74907 zcmZ^pQ*b2=5T#?AH@0otnb>wRv7Je7Y}>}fwr$(CoeB4^-L0+Nhkfg+uKK#V`{|tS zu?Rq67$BYt$jKup2%M*9ihC1^_hq2K=)`cvb-sf-Ew?4qsNYLfi zeLaL{d(`XD5UEbDqW1bU&Vm0wwOixbIFj&5ZYCiT(s-=sj~Cc`T6Pg zvDkVY2b|A}?VR7?>Gt)pUAYy6zHLRmeTmnf*lhg(3L%wY#$ zP!N_O(ATV4>Qmf4U)gEKghT#$e`y`)+HTrueB~{k=vo)UIcY&Qb-(Rqi}CQKVo%Up7vH8sc;Cq>0f+Y7sni7o z3Vf9uDfF^hy()WdrVISisl5?D{w6D&*Xy3@wi6xtSy!e>m0vql!IboWEcTOEb3MxL z-NcN_VVhwIm4bZ`DRz;x716|&vbAR@Iy0iw}->MmTJjEj|`SqvTGxc zW_RZnsroHze@Wl9|CQkD4nvV6?%t|-Qo&mN@>&T}3}UoAWVW}JjfUNMXSe(O0AjTO zf9`@!4kQ0*)gES;u!@)_L~_w&9|%wQ7%!`V6>T(YWZI6TJ#WY1qE?iSGgkJEclYJE zd=&Z|d7i{Ruk)YygV1(yR4R~#@SbgJWB!3XifOPIem2w*g7`zp1x*9d#LHYKshA?` zt7#dpLQ~#k8PMB6uZ@kgD*6^ecUuDy zjGMbylf>Ao?D;KPc-KJ{d9q?Ci(y{M+woU!b25`rM5BO@z1lshpfwLlqP z<1S*LA6x9o85QZ!3g*yxz{nq&7595%rx5Gkb)c@z(dtAemN(_s$|>4drGZQoaN}mj ze4_s5gDtWM4H8sGb-Tnm6MupO9_E068#6jPyQ16tYfIWpt>MhH#RyegN9)KjSK1_d zbUHHB$?fs%VO8j?I+-sRwRp$%W?e(Z@VBKbciDs)Le-f1 z#n7_V@x=YN^cBMYMr8}wLaikQMT!E_gULAjK5dOf7R%8PUE2B`+9it(Om)pT__HH+ z5;Aa}WZN`FCDXVjB@(cEK*?c!$|P|OuV21H{Dh+7HQm_sCchkFwUt0L^HJF48G7xO zfZkx#!ul|0mMc0^r;>&55SN1f9m6`qzCVrCP&BZwA|X z@Ra^I=XZc)HWyQsG^^hUOm`hb_uU~#X1PehrsqRzOS@(cXP8uyB7*`yqpO!$)b;R2 zOICX6VNOYDOofWJXgE}rfaHr`x)t4zaKGX;E;G3{G32{-JxQW2o>P23_=fiAFy3}b z0B?VLTeD-tyc+)?@7TXiYT=);Yv&p;qvnm3$R&{NH}F&8$&8%`A~?xMPG-N|HX1A8 zON4B$7wt{^5Qan{7#lEhT91mBqE*{lf6rxLMYO-dcEw#RIr``YM*RB5k1lZ5yQs-} zq*`&UZu3MbCAoB)+H_I}BK~t%pGF)YYGi{nCetA99ErB~b_R#7LnU5AHB40_xb-qQ zZZkHp2PoiSeH`-w5x{ftM;rM%PomQiFv)Kj5zKKo6*I8zFiU6LP4T)zHh4?@S@EZ! zIdmQjH1yFh%bS@Xx?SP5x|iy^q0YFs{0HJ!gU5C5=z>s~Hf*!LYPwn_+1pY^m~kj6 zb6&H|#}*Yx|8Ok2NaT9g#cKgHPnjRDL)Q>*%-32vpv(w>7o_Ltem+c*?XFd{7V9G! zD`LfO4X?q=p*xESYzTAPFg#+&Q7 zr(=3A6@zpdd~Tj^C+cBgMc!dy<32YJxhgcQ`iIX}y`x|@SfPrnA_)|O_~)0tsug8= zb)#q8v(`hv=YHu{8k8)TFufR*H{IGz=%iDto?&k|Mvm)!lZKJ7YF*kqJ(P3G^-}=VOU(B-l0dUUOGY)|<-BClXhZ$gzvPZ97day4o^O2OJE5W+s};lk9$m2Zo>Y59`bH zh759K@lWSvtZp*%v9k9ugu#}L4=U%NW@KcVd>N9RY z{>qG((du93W{xJjs-`XhgYa3^sN7QIlhafhuSd2#?E<*P9zJO!B79#1+e`5ugHHS+ zwHq4XMe3)yNm1>VzA4rH6#|yz94HdO>2r!?`a~~fbT{R*Cn!XshX+_AxwCCT$Go!X zv>y^+YHZKHj?%hIihPoQ3~a-HbL*+er*ju zK&*i}T7Wzj72OnR@}5aQM8TN)xp~5`qRH64QtKFR zNB}Cw3c^*o7-Bg)d8A{@4lg0!JbHfdaXZ){J6H#4+#66yWPmhUVC(lc1+M(X1+*NH zJ<-{&l7Vsg7@!Mu$2eZq(BHXPar0Ie8#ovK1yW5~&@ zk<@qXNVzkpaU1iO(^n(O9TYW|^kYVpIq1rer!qn)>TYz#IxPca$6Zfa8cL~1^~6aw zce)bxaCCVfdyis_!U1Yr*hjXd80)+@CgLMK7ShKf!TNFCrjY-u`7r}uVqg&{!|2MN zyYTdNKs@BU08ZkT-+K9_Z38l9a@W`wl;;dazD7KG$tk&F`?eiporvkYuPc;~=R7;H zyl3u2RyL-flGJ;Vjzp&ZQX44u~~i~kz50oJ2Z3FXlfpQoJ2fkTrW2= zdm)T};{{#ibD!i`EnLMQ)QD>EUU7CTGn~Obdd?151Po+fv5(V!z+GuT%{-!}&b$Nye_*Gi~>tY@%9K8N%I1G8&DY|J2B zdZ>@Gi1>PhVobZadBGufEuh-lx#FD}MaB)vV#552j4}tgtK(3qn-3x{_b|3#CK0o7 zX1BcA^t7XP$00&!J-Z#J+Ldy95bT~DX@>H^H0Wqka*=&A(DI3yys2-#OW>$-ug|uL z*EAN1vC&w&yIP8{x(gM7;9X;74GJZKRS|iXESj0U5pC_!+!O|@lkq=YuB{4&S zr2|fiyq2MJlVJ10T!?~CT9|5Dmo)Tc2tlTs=;xeDYrvh#B4$T^sr}KzS{f4#vD}ai zKXd|5pR^LY;m>m49+x$*YEKMHhRkHNX34q6&4sAX%jlodvgbMg_&y5r_0iO2mHEdWk-V@ZN!9f7ZV#&tiu=&L!L9V7S*LJ|%caC^_ z)A~fF;Qlr5*d*^A1ZVZ&@HDCgjr{4R_p;q4)R_XuqO2uvSCK10=|xIpEQvtXlp!uXDQ{1r-TwJL8`?&|y5BNexv0|DsAt-uM8v+%O zjr;%BBPDv{u}5vF-5+!p(6rOOpuxO=I>a}Ne1}lvoTMa|hyEX6{g{z0MoE>esZzQ^ zgbniwwutZN>!4rym(WMTwv@uNhV1nHVe8M;?DyKw_xF7o5fSh^>#J`6M67|s_tW!c zR;aptS?gjnZnT%vG!VG`Luj~j^`qeS!!{!R@w`sdcKW?n|MP>y6lBBX69)Ww8x{2a zv@FW7mc=RRbZ#6j^&k9UANgL9d~^2txz1yjU6Xpgxm;Xc(n)!N*J*)lYF)b6(Mg$a zZFPJb`uAn{l@cWJGQ90{##!)%h&RfN+bYS@H?uM2E7#olH4AL=GMalK)EE=%xBk>3 zQ>622{Yk5x-P6C^ho%sS#(ZDnHrRBztV~@ib(r2*cy|Er)fJ-@(S>SJ!1UP0T7=c| zpYY6&cwvu2= z(~{~t8RtsclL1NO8wz=|h3uP=O6|n^Uic??Q0s2%r*}st6_~wI$qA(Jvx$OloL62k*gJ22Q37$Hu_-6|l0Joz0EwDmu7c?e z*BGM0_6SJ-g!AstPke5p3(Xx5shfQTJg^wGup=Rw>~w=q_1Qx#U?6v!f-cDY5KJ_j78Ah>r3YhFf^G2PCrVvLFO)lXh2>21;$f6!IVqFl_hHwxgLXMkzd}n=hJ6Vo5 z*G1vkNq`2_v;a(P*pw6xXTcU&X7E`&F2Qroh9;@&U2eSF_z~}i^+M8XhCKEgK;td! zsegTH8RhZrs6v48>bva4`2ADrwYYQOx3E|fSdi?zd_%wDI}O}@XnkGC0fk%k7YD1g zYMy9+AMArd12flmL{29Z(p<(hmJFvOsIxE;G|-l8g5SoI>8c1j@p4JD+w{3OJws+` zO2Ku+UF8b`vm0;2-6*dWkTm;PZw^lhbH14N03pAc#u6=nn{Hg%H(W3I!A-OK$7$qx zi$xDVz7qy%F>N>^Qz6{o9AL@5K3+oA{NcibJ=^FH_6U7_T`bCe1=~D`<1o$?D0a>m z4m6d_UU&OxRq?eaq4o@q@y&R-@?^L)T83JNYJ_?LzPK- zZqCF}0XoI|!Gvx6WpUZaox0_2y&I2uA@KOV-$PMO+SwF;sf5{C2zh@U{1H@J6!!uu z#-7`8xV?~K^h_{$H>f>3CM8D{nhHaZW|)%$J@E@Z8mz-Yb+AMJz1VV4>G_RUO%Y|s zIHz08l1XB@5zbqVx6g{NL^F^odIcUwZd1tiI+!U^!B~jt*g?rf?eG)+6P8X+9QX*^ zXw=FGPGvuR%_#cD{x+4ghY*6e2=qiY!61Op8;IO=trQIDsP$-Vrh$e)S3ymkJ+j+Z zONXz>WQ8Uwg%Q|QfZWD!u%lEYp=`kygoq{Kf3zR>WYoE|pY>N!fIy_IbU!|}4NJiK zmcx$peM)}6VBnj+nYjSHhfJ0(y`WQg$DnepXi?{`z*pU80Xrm|Bl@#=z>P>Q#y0gG zIiCd#heCkFFR5WE#`uj^c{4l}lgq>>bc zCJoFfM4GJhZz1{4P(PQB852x&t1QsHzCdZ}Ibn(PJ z)c2-=z)UlC$C%U}LK3 zZfQY0>1+U@+ISOBI(1;**_$!)ZJ>_#5u{c5+dK zp;B2S=Tr{)_2}B=E!0;!l8h$|oEAoigt;L8tKV71B@G@n8+3^6!|ExD66^Jatd{sd} zA1s^_0BODv^aH>AevKflV^m z3I&U5S&~G?m@dgJ!w{K-I-W)-0oo9wgjcZ_>A=k+tr(hA(BzX)M4*sxXDeGSFq7YM z!ar-z`~*}S_iofTPd^$6f2(1nG8h*afVEc8$pK8W^n1GS7Bb`M0&CfjA)oW1zJgje zS&aLyXYiA61I_-bIX0mun-|0GKyh@ZD5Y^R;gebRvS#0%&`wSU4o)z;N(6HTh}4cl z9u+HS!q2&=6qQi`&Ww~<5qRdJXhX^=P4+6#C?BB1fWfyOh~?!^ud0QWY!hfm%AvN^ zI7@W5iXZ6&$X*c(K+%?GTFbi?7-)lYE}Pz%OA{M3(M#sB))zGDrtHIhLs@h!K2%f& zq0`yAU9OTzK~$773MLU*x<>AB{=Y|9F(N%gSCqNrRY5YXU3{&((pnylTy2g~oRq+$ zJ0>tKJWLlfVcFeabgxI6g6qfW6BGDqQWg8rUs_lN|A<}0fXV4XV+lwW`GV5C1j$td zAa*cTOLzrKRN{blcZ^h&iD-%^b^YZ^#}G7<4sG)Opl~}BfEz0V=;Mj}N|EDnVqf0h z+z36!CLD1d1%t|pX~jHyKPs!XDjBdJl={Vpe_3Qm<<;E&zo>$q$elSX3-uVWqr0Dy zUAEuuI)rpDfSe_Vxwh=d1M0)t`7_&+1ry5BdeN^?lfhG)kwB2p0#tzQe+h#R0avx>UErf@0G1<3NdJI5d8cQ%-Z+|+Hl^T9kbimWAJ++&f>Bmm;~BQuax3Q{s~CVbsOt@l4p^zw4b z1@ur}tmT_E(B(ctFQ!7_tPX&h8#Aa#hQ5 zd>4M+QV$_)%cYKKjeHmmjQLv|+x2W}HXcM>q59<|WP5 zzr%y7nIe}){eoF=eWMv|L`Op!ai1lQ<8awR(NE43)1u+p%O*OP!w}5~J^|T`>3ZWf zfW4eJC(0M)&p|yK`F4qKx6u0j%|;*sW!YJs->La9gPFMb32I9WibC|JtJn`?n_%cQ zq&Y0eW=`Q163ANNjMyp6y`-zhzhRb{oh74{#SZ9U#Hlf|XyD6 zY=X7KHn>~Zhjv<8LOdA?ydV%Ggq?j!w_||HzCzeSH+gZdWTmGTzqROH!YqR+uy~y6 zB5W(kA?z0>&$Ng9o3p%NWtLuw`qRpT0u9qKuF3RUbvDb;G8K+x^WNA>x`axaAhEn`_YLG>|bpp zc0R?pM#`!Kb8HdYU%v!YE7!2uc~k84vG&`Vcjnx(WZ0Buge zTXZo0kCKHv2JOSGCly+$!tQx6)*Ecn;HV+)tx9HcAuTmM4r7m;#9$WrPQ0qJnW1g{ z6Vg#kH@WXthK?FXt`_iP$ba}s;>0N!VQg|Tt(Dl&06nzWYb&;|`V$ba+(r%FWT|~knb^!(bI0XXhE32}o@?AgpKs)pkh}~D@i_d`IIrwhdC2)r zhLFxF%qM@xgGnHL^-EuvNi&!8RO)`CI>W&hicW}w&3(=Hau0|F0;f;srswL9y&_&c zedgN`Ulq5Y1?R@WK0J69|7A$>_$6}s>4d=nb6B=Yj^oH^^jv7@az75 zR_F`auMt@IzZ@2HB1#}T+y8s={m)_{>qhK3rFn@KLwh8W83GAgft3FJNR~p z{3G0hV)?mr^(%ewQDpF?RXor7Ru)`NeR&zz&z?pk;*R(Z_)j(-=lCj}#^$0y>B z)z+S0Rjhc%z&qDl7d}jgWpxK$JfQ1qt&}x7=n3fLmx1Rz%Vh9r)$^z@Y8>F6V)r~K zJ1VSKb*L@tY?X7#t0{Cbmfce_x~`o?h~tC$VOTxr9!$2(*}k#EeLu+ig_qI-#pvBy z6@3>@ZsJO3pnbC^&SwXvBapb~xe8<6s+AUZwE)e^zD1M9+Lc;aO31v43T)l~AiK4> zZ&G-TAEfo!b91H0MQA4mygg!dt%uiqZ9H4TEOD&r4YQCn-0UP(TD(HL5r%g{kE;K0 z#V}N<%9Bpiua0Xynew5Qor_NZsa6U-YuJO2gl(nXEsq#Z=&YCDS!S2!Bdg2p$iB@F zEX8!-En8NRPUJUEG8&wqHqsk((&^Qtjf-UOtduGGmIjkz6V7aG8k1z#eRJr;i|oKjf!w!>JcrR2TDY!vTRaND{6JulpCN&GykCc<%L(L345FjzbJCT`eC_Y+%x=D_KXQUX4(7 z%=7V0I6(lG-SCoLhITqz+BX8KLME)gKN0mYzY4C6t5(O8Dow5nnjT^SY_jt%zw<@} z-?Ao|}(sj`QcG5tz(|i$r*QK3>bzT)v zeJT&6^U_(h^Cot(F81KWVFb%^RpA5$&<$32vZ0E$KyzD1dq-v55%qJk*yQ?}>B~WN zb?L{(iFJ1(VFE!2GTym|W%;$(sDFL<`AIKZn;l>5=OF%5hD!=HST_UA#VzA))yx&$ zJq|O*LIa0aAR<7nTW~q@0Ek7SK(nIz>lbDx>uoz*?GvgsM>CyrrW=8LtQYFDNP;yKv=--#CQS&{G=E0`<{Uc^xfZLM-aM*~s$;Z1>2hrd~-p%%%+K;FbMUn1XSoXW}MGfFQ0OFf>_$t{)qbe^R+wKvb ztua8AbOQ9-u2P~}Tig$23A$MF6d z5?+&=T>n&VZ65dDHlbtP^zRGHi(BPiZDnHVM|LJE7-Luq3Ca{g)T-8|jOQ=O{A(wJ z%*@o%sR9)Q(V}Al1@-PtN`D0zKT4NdwKp(C<6hva{{do-2F9*p(H;y7Z#TT9IYsFA zMddk(_1geLO+CiX+@e&Y@cnUeGv2*6i5}N;o-(8!j@$tD+HB>ogU3W z4T@IFfD1!U)aXBOt@W|%VkQGY@r<(SkSjmV_+1OH68_vaG}ba2;rtT3zawU)xf?quoUSy>NQN%FszU4{idzw*~$YGWg7kJ2y^0 zJC*+T7-~B9wdhajD_?$-CqnnFzCB~*|4xuYebyEgC8`=@P5iJzPTY^x52|-wh(zOO zk~#i|@)Hq%S5VSazRmSw7r@=2X_7aG_Hhd)Pom+X8^}{?@?3%|!v=L~Dr!`1v%>`z zy$UwaPDh*MBu(&mhym<&*0B7+E@vH?bXnc>@VxFXKC%wwHXouo;edavV6%zGu$n*( z9$oKCTMsZM5fSb`2*lL?+D!rz-7E53<*v~!XC1P{GOfVc(74q39|eCQ!j-OfXmOPN zf>YV@31l6^aPQX-S+R`GbNPUu4*p^sY;R^NsJYY6^|m#rGBRUFDsM{`v{m*`0qGcS z>7!C96gjz&URj|xu__08UGi!KS#H>aJ|+0WAt}YYDrHV z-)^o^B;uZe&|eEcDUMpS$__@r$DCPQ#fGlEaY=49WadXC(^dS3blF+w2AAypc|f zG1lzeMCgw;{qvKRJsle~q~Dxt7$rF>O$eySa=abUCPL}DkK**;qxeh#!|EbB(G#jf z0v^ZGcqF4#-;GADVZD6Iko0^CRKKV1ge|epROH!sPpEd0|5O?VBQ(!T`Gaihy`f#p z!b8A0CU_5X^`Z={Bs*vALq-B!%!@wK5l+WIy$q^9L9IA!GOy7a1~2vMCarF@%SwWv zM`$>x{@zN$?$L){e2UKT<|a)Bt&d81FmcBkR{9(L1=C@xBsenN^u7v_pMVp6Ov%pM zFJ9Diqfy9H&7bLqU|iiBU_QvccX5=Z{hbr5C@ZR;hsPBuheInz zP}l2S)Ws2HoWCQX?V_Tiyg0DRYetq-*!mL)J8>u6gyZ=e6R3-q4Oq zKjsM@R+4!|*RGI1X@CQc7o4F@+6o@fR1i72*D3tXy9PY0cWm_A2PSH$x!UQQra>%D zOwi-IxPv4p(7EWxT}BJWDxAf-_o#Npb`kT`_E4*z)4g>7%$XWgCf6>OowgJcMo7g) z8-3y#B**VHn}wdIXY%*(BI?U8W@)*sq>~Mp)X@LE5wP5ci-iNdobUSH2mg_ajt=p! z>zmCgcspmqd1RMZ3KDOHutk!yyKNxd2H5dLu3d^!c#Q=2_<%#zAV?2 z;GZsvns(}EmYz^BgrmH!AP;rSq(>nDwaHApE^b@d60;)MW$3DAFh7sXLEO-)t9QdP zJ7JBR2!ILfJCp?$QX?0r%k>v=+F&`K&eVei>oKTwuVpU#=?#)(2SNx8L$*&zQO#)H zZZ(qB85^ejMk@I27WpQ)NKF z17XBSzPUc=11+#W|FSq9xw6ijafWweG|R1|JWYkyu@DAsgl zKB^j6+TNo~gtM#N$gp+-=H%Oxq*g~ldKZB=%mk6Oz_x0S_tCqw$;bNcZGRR~7pYm6$UPEaw&t;y+O z2vGKHyg7X=y$_W@C7q6wkY6!(>qxw4$96=!$bT&aI~6(XWxY9kBNminB6y1^TZrZgvxr9?7~2N7z<9}S=RWgPeRtYtCJc)N6BbU1F(JxFS~2LO`~C! zaOs+9?k2++A!0&$Kcp6${!&I$uy10~IhdBV3o~2KRttV=_?|)W&$YQ?pbWo-G5bJ& z?XuW+!--}^NVC9o^Yyqj&vm$TetR;nm?Z*ET&^*MBe+6tGMOwP$~)0M3nvF*)v86G zxe(z1F1}7^OC7oANRT9_h9JS56GwzfjrNMk2z2R4AyQg6vVHLH(xf_m4B3Eg3!thyir`aOi9P^N zv(h@Ge$K~kG8Di+nHJ*zX{x0$dVC~f&i8CXg3b`^lsEs4pH-ZiwE6Q=Vx5xl!!dem zmN9NrZ?)r5aq&URtq2izHJh|}4Vl7#Y*@Z;0^VuxzL*D%Xv@85a`*fzf zNxHjr;{GNvLyqnR&0<2Er0royoEa8abB-KqnS#G~$=Q+qnvqdcfR+h2bzc2MRB`WN zubLjE=GgV8Fma|KH;#u`n3)gxcUOJ~mKKS(9x17re~GT{6~J<~>d*T(;9Li&too(Z z)GgKUv#5{DINUagAB=E0?P0Zw4}n$C7B$eOLD`T*i*L>fG_gGsBsG^jVvxX{biSXc z)KkuOW^qx8B1Oj<;DjHd^2YX>0X-U_>0Lj(y~i)SQWqnnG{?zSbb=PEZV37_^*Djc zc{i%xM{9c}u1oBm)zOO6sN(c9b$7 z2TR|~f;z(@uea=nk06P$$|_>;G8|YT&)jVN$vtFv0JcEZ)pad^LRD|PBsL;1n?Baj zQiwz}f!qRUofKU%2Xt&TE?$JQ1H}xW{b3RG%R2OS8aCY9K}xR(+YG63ilDd)$hTbF zNjuy366L;MN|Y9waUp>N4rhORyf3n@Y${)xopbH1urwi~;IgoJD3F@zN%B@Bm3u#F zf(jjC1dfj{Z#8)umc*Z7` zu_y(gXcYjdYJk?!my;`w%Aqw9#n=9eua(qcjV5x$`*yWrwwoqNW9OO1rN%<^*+y5V zO?=#C-WR}H9cK11=ka^`66>{Q_sLg<@d2%+WS96%^PtwR4F*{P84435wm4-nRu7?s z7##s;;nubB>lPaL#^km^l98s&4Th1##ZsRn#1@vsz=qbi#>NhsG{T0KM9ATs1kZsE z#=@DTEJO?xh~+~ACqWSNQP?K~`8Jvnqf9b!`Qc3TphENp&|8(urU^I>OpINtPCP+ds@Ame6$!WwB>&7xrC&L zzvF67>GJnRi04TIM&TAz-O()dy!JFng13>~DD$C_f zD_+~OpM+C24dW@NP3SRp*8ZY&N4B`xXG9e(X4>&5x|X;SoUs$WUT$|t@2fe8UutG% z-rE&$yJ^o7F5c>LKsXt7+ce!Y9R4^FKEnTPOfG<@rB~MP@d75})>n(Q>~EOk&++r1 zjk@tpvT~{>KT6e7=tt5)R*^& zCcjHbn=?IBD)_ZpnvRFuwHGu#iI0tcTX1wPTv;|@>-S9MCGNtG^`MZi$-m53HR(oF zXmG?VPQ!zM5rYKYetqA2C^8WN-}eT;J@VXchs!3h2`+mBfvaC@K**n-TkRhp$hW=+ z8Na|~6E5`d{%qE?p8JbWxI(FqMNBZ0ZLL1ZmpRvO>?tV>fN+y+WBQ51^rNwZst1sbl*wK1ER6c@d`EwsiVUPegMoNhxL~R#@$aIrF&Q$8VcyYX0GS2>*WC{zHBiKd>FxFMZ8=5-*8h zk3SmCQ>+~1%uIlgDv7Kjo5BnYRWlXfAI_t4Cdag-T1=0NX@SHJJ%t~*ZmecHAsYxP zBFTt;A{_wlLTqVW}RkkM*jf5unbUQY=$2GpSQBBVEDz zFo^8(l_9`@8#^dmx2R%JD+&QN|E?ZCBpZI(DpR8m5$ns84o6sdQz1OgsNP_laz40PlmveuCvQeAB3kCw2MUW&!v#{Z01<|$%J+;ovbamhU(b*J)gKsW#*fF;Q2{~)9QL$avhHJ^hG z@ONMXkr_D5K}`-coVYg)p$V1>|G^*rgU?Cr{Ut>C|H79>`CBfar($6KH{4MVy`HVf zTlUS@CS?jHi@MtOq(*cPC_XOEyns@K*A|#t-xZ5$D(T@c}?U7*EPMetwef5p4$(WR4rWz$~Jkd}n6`fw5^{26ExGr?L! zia3Ii9N<%c0VTVxQxR?&+`}x3 z-jUKY2hR*CAL%Udp2N`r;!Y4`1qAgW{s}B~tisU&{dC=Rk=ig(&B|s7^g3jwMl@Zl zSZJ#W;HW4pP+(Fc!op2ZCW@BL$#?H96F$>l29veW^fzP}zdm4M8g$^(iwMQoLHf7O z)Ykfpn?bDAxNYo!v{zDb6eQ@{$jp$gX1MSlFmlDd5L5?(f%R<$?*nyiT3TD9JX^(c zUM=-y;VEbuhFNu^!BHxh=!_G)FfoC=3KRj=!GpY@s7OKBW+pieCIpv6;@tG=sQYyg z2K`zZOuohdQauRM0gC^F0J839@_*-1?hwf>0;y40Cx9@Rq1C? zFf-GS=rG2$5F2DEQNS^{6iQy&z=pYWYRpLsKriqcMF6!GAX*!6$J}UhsG-L$!6x_k zzyjsA5kIsw{=y+FL#aH#;Uj$lnwFFDz_ISdkw}-n8T|nyd8{Ti+9q2n3dFr=u!(_m zK3hz@yS7!t6Uu)%eN-l5o-Wa%uGlZ~x)khqT;VSN{Qn>Z+W!+`y-H>v+<q%(3dcrq-GBQV zchEt-hju@(X82n51NRL|^8x+c{?P(nzSGMGHrEK-YCnf{_Lo8WZSwba7;@*UpoIYX z*C3Oa-{N59&Exk~>!}eA|7YAyQP?&W38uA7fao<#Ig|}PJ%c6IT5OOb@c6Gh7xmI# zhi9hKI||zX?muKk)L-jdpriq{^oJEfoqVr<%ChS-wZ}giHTNNrbf7%tKs09UX97l> zTBfAulB$G!j(QU^=>P$9EkZsxJ<745H4sMSLOvIU{kwl$b|}7f67Ap1QTLlbEkGr_ z+;ift>xp1*LR%izfI)^dz!-+CS#!ZP?0r0oY&R2ELQ&|~VeNEOBFs860A=CaD>7kU z0CKgdN+#F|<755YzMm6P)jUI0z&a>B{w2m`yKztmCP~vcKey}`oR#+c`F_UfIT)HKQs{b-w)x!3AQbb8M5QSN|MjhY`*!3Zb zD$zvxH6Ft}8`8ULJ~;L^)^u-uaCK{8F9nVn6ip;&$ZUkog`@qODDM?&h19ezHQKm^ zeI>$HgB4Z`{ca*jn3R8Ibde5DCqA|u4e7lPkvLN1o3+xnQGbK}`ro=-GinpZl!;25jcuUA)QAxj4g3rLE)x12h$BJ_LQix|T{aAl58to= zlEt<&%^v3EBsl$L4d&lY+ot9&vOflig5wEp@Q=0Y08-GFq?lR6zvPP_IXS>>(g;64 zfgeoZI(a4V-*7JC(Cb1;_b0<`YV=_s!t?o13=BBamy|G)5*wO^6!67&kCYij*xuRm z_rBKWJiHJVke@rS#N)St-l-VnCpQA5ud|fouO5=4j!ayU-7dIV7zR+-B5EhloEb=~ z5a)@*WX*p_wQ!IonE~?46_!M~(7pMf7WI)lWZ2_@Zs-Z$8L3w~PwVP680FR-IqZup z)o))A#&zAHzx&H4O3~lFNlVe7cARMI{wS-%_J7XI0S_DHHGOe)pg1_dIC26;P9AEbPw%}YT2JwOvVqhI<8wGCYA2Ce1yu;u#InqN1c(~rlCJYX^@WPUc zYZ~QnU|Ld>e|H`03Pys`a;WQ!S0$IYuVNqq;!6;NU}{~xe{Mh=M>;ot@sl9Ysc#aj z(v@Ety1!Dc!k+p#b!B0vF*8QiQZx5i8&P?_p>imwYOw0LApd48bA5)(?2X=~nMHaX&ORZ@80}qOt z&!S{FkEk5t0!#+czazmZQr1qQKX{(*aVO*dDQ2_h_w-CuGfOk_Asvzzx$?Xk>!!r} zUdvW_VTljOb^9!c5~xpmkv_U!MGAKq4!)Xl-TX>zwvcOY3%92NRND%RJzWN$b13POmDyHV=`}UGLGFxr)>L|=F&ZRn9RZu@<@UUO`q`Gv9Y++b&pp*unn{+ z92^O`5HF(8($NCtGQZC0)&}nt@4zBI6mI#i8eY$nt2{&K;SS)SihJ0SwtS5xn}?C_ zCIqHmZg2Om-lN+av2=)S*cGIID@+8fPs%sH-QwzK<J=jpwt)xC9I>$l=*#v%hMaNZ}HGUexPv#Z+YVLFq^k8 zj-G2@rqLV^P3pNcyas7Y>#J(6zl;aeUOt|zdttw8&oKX zh0UhUu`p`1XxfIaV5jCvO%b>1<*h`Z{k6ENI z9D)uz)2*cbxzU_3%eSc@9$YaXM<6y;W=NIxuzWTyVLyyl6XPrqInz&k1CbB!`g-39 zga2j`?PU^$TLv(-lDrcPOsA+B50`;?jI1JJO`Tjfs+%bjyCa2eB@UfA$C#RvN!0i& z=h-!?Q}ob(ThT_k9bce~2ARi7InRKjv!8-5zv{+Wd>PgB>#ZT=e*k4bn!jeJh|V8r zM>)yJj40tL8cJOg3UnC-2-OH)$dX7EVc95wH$eib5R?#n;8JR#l3hY>O;Ud=DGyQQ z&n3I6tA%t*MqlPwp+PI+S-DM&`7jk>0k4VvL*`r>LZ03qfcPwpA#*;O30N>R$yDh4 zz2X5YUceL{hdDz==#wOAr8lU}rct^t?)4>X|T=M0em9BnI z5AR4IgL2F`e_AWNqyzc`ajAkkbC_J7S^|DSsq1dZuvj7e8K3hi z!M+16D%_O6S+n4Tl7j+UnFSM0`xa{^9Cfh~!|A4KI&=jSje7j*zDE)s`^}Um>C?C-q_vHRSGMko=Y4*nRDJT=otsgfNV0@?! zhz(rl<4K#wQa#v0Pr!5lh$(9s7uIx7T$7bi7O}*M0)p3jKgZT>( z_oHJ{jHV`#|1=9^7_M;wHsT9*|2&V%*inL&tH)(jf0%!9Wvw(qvC+(Kp0m%kbnwGf ze+aGgII5-d_YV(u(^T5T)uvUl*@xJ&?8XM=^-m2pB|B_;Kbu?S93=e+Ti&Y9-bB!pnR0!%njUDWjLDzj|BdQO(YK?$`tVHS*{2@YRP12Pg&fDXOfkuhFomk zW?R3jLUw;?O>EMEkg-0=J}IcId?$NVS4X59G0kQNukTIH8Yf^U1g#qX4WUKW$X@E# zIHfQ$ullQGFXebtQe#REn=aX$H~3l#IV(Q@5Y`B+D;=~HN$tI8V1(Jxg}SbjsFeAV zX>K~B=OMTa3YTe+j*e3vq@EDUrPZL~*XJGBU|fH*(Sq()x~Sf3BPk_O;gD9-QNQV& zX!*@bAEb|2Qj`U?;?r+{3f}o)e6W5m z)PBP9q^#PKc3#T8&w6^2$>mdG48TNS8tWG%ERtU*wC=sB>{q&m>MLI%}ZT3-afI(K- zr;zDVPZP>M>IztC;6k!C3EfLcK~#T0Rr6u!M?~-ya`U7fk|XYEFTX0Jg~9G+%}kz> z!`2Vla!pDJPF-zX(%BUly(+|8)+JdC#ZDsv$wzO$F{=kKLPo}ReJvM-{N?KQ zA~&w+N#qu97~PmIY&_JdFBhv{>VVS|TX_E^A2Ce4LGYlNh52T*1oM}p-_O7O8o~Qf6Hzt|Fh9nXhEJ4`HGI;n! zJi@Xib3(cSnhfR^*zaHjRu_Dp*4W_Ozpk&=ru+utSniJJH`LmwAI*O*`57~@(Ny+^ z`X(FDmJOc_jvNbXGPTNqITencdLu#%@UE{pSj(;0iuv9hjvjBOJ#3~{Z02=rCa7&T zwKa{+l#BhW(Lw;;TU=8bfv_M%nAGa$O5s)e=?I%?1pB$KEqMz4;jH4#NgPk3-Tc9u zM=>YwO$jVY;|o>7a%q3oBcW818eqK`%Osc6nPPAEIavMH3KsTrsUo>aRZTZJVa;eI zNGUUi6f(u~S>6b5&@^gfU`7buy!CuNdZy`6Y-U8=k^L(fOG_y&Y8Rdq(@4KlqCCk%{0TF!kD$qy#aYqFuq8Zp&I%Av_xyQTcC zZa;sk!Gn`;QX3Ls$a2{78~e&L-u`Iab5o+8u=1!ly1tjJwfE~3BIvGeQ;)}IDUOQA z)M}hU%=BC>)x&?)_04V?I&Ldn9ZZmoH@m%ikayd}TnSngtbD-BFTRr%o2LSN*kB%q z&l#QK&avQ#Psak&5e%T-8`EpxVE%*;-VTiCAVbPj-x9-@%r96Y+)fPtBU%x9Ck~$n z!}G}dqy5)+TkFoU$om(0to?Ss+{B|aqhP;g#!dQ|^>lxgHdRRp_<9#-H)&fp`zT6l z>i(pJ0uC#KNf0hSJ)xi)#q4Of>VBe_|2mh(^mWfcxyo|Ah&(zc{LQi1GQ;yZQnrDF z)LO!0^&oXi)Z*YVq=><<=ceGH{uU4G%bD#O{ePa$ z1cI45I_~e?KKsvLSo*Y810S1)Aad!zB8}EpcbR`JX~=hV=QZ)YH*6gWM{&CSt=vU} z*-v&44SVmVE#hXu2biC~s|19%tWy4}j>^00B)f*h{DvC3WP{Wa$;@cjk`K}l{-Io0 zsG{4D@84D2f4(N99+N}~=syiWgRrBkW@*NslEwdc#;Dpc4gEv8ZRavgE zL+^hN?$D`cfC`D0zm3x+VQI#FSY0S7)Inyb zM*^yj+6F2X8m*=%RC0Z`B8+CKLf&Xnx0!|&6QZNa;}j#C6c=t<4#y4I+nOT3s-Hvu zh9-@HNh`9fZ7dM}xs)=R-Am?6lshxkBPd2;8;%j)!|~;ddbGE=lr|zeFilvQ(>{N% z$=bgM)9bX;v-Py?hQM;e!SwoCC={LPszqn|YSEeQcyPLA=$}?seB)%TuY)h`B-8mI zm1EASqmk#FTk2&1x&j5>sooS)D}g~U)1ZIyz5FRDyr*P*6b__UOES=t!{n7b;P9rz z^_b7Zu_3jn(fWO}Q;WRU(kXwdC1ZbTCm{DQjWs>ljZ^jck=t(n$ZgWr-QF?Tu6FSC zYUdQ~a%*OM&WCcj8!~&G9?8uEeLvn;h|I!f#i?*5)t(c%;ea7=!P2k<^}(K0DmZuH zm(E%C^>dcF`@Ix=s+LqIhbSA)OqM+I-3b+wDH=9|1^#69XhzNtMxV$yPAPv*O_-;C zw)w~W_;4u*(^rEqT|Ed>4K|RI%nbpR0{Kn|rL}~wu6BqHw!Rg0qF+}LS*T#2kG?pv z!z-iVG;za1py8p0k6P;M6P6?G-qFBvj+dKyl=7#>p0|Qo+E?8NR_>jE{+vOjElO5v z!(oWrx_lmPK1@q(b+t`5O%Q*k$y?QQ$(Ak7>p2qI5DmL7?1{VM#p_g!GMuQ;<)37tWV9y}c=x?g?Cp40CM)!2uWb+iyllv>2y7_;@Jr-?0d`jtf zHT>hnXOGtpKVH0l^FY@nZL42-h>Gcl_-#rvl?C=7o$I8XtsCI@WlrgDy$5V;@;ivh36I;Z0+f6XrJou-r5=_`3R zR|I8b>{HGReuegP*XR-TWtEwhj_r6vKifZ|uwQmCyMwZNW95IYOZ47utQ4^018L#x zzH_1>v-YmRf021#uIl;UC0}LDN!Q>UqfQ0VO8o*satq;Wa!I{E#SVm!-tG|PLSfx& z@EBg#v*7m}UmTW{sq@w;_3o%syHp0OZzuy+tKCTZCVWt;Y}Ot28#i1G*l;6cVHI{2 zFJjXw?C3q1JJ)|oLi=^P6OL>+IB;>owXPINJ`NoA&3@Bm6f2e6Y;Z|8P^@VNBw&-b z;v=K!c#96%{OpmIV8}4?}Ggu7eWq3RY~8lTd5`@cc*O^j~6YZ{8LOkBKMK%VN6dH;t!7Jjy0S?_?FQ0 zRdt8maL?&UkpPOddY)0$oAG^!kKig*7Y!@5c<0%hn7|FI$S8mF0QV{vctQU%o-p$! z9ZmOlD0zQ1%yZUQdD3e+#V|;lOxQPVuIKKn_^rO9y^GUy|oX#zF$X z$+5T1Duh2rVLEe9m;K%2o!y%k6CS_zaTDj8Hw_-YF!}m$5y{QC=JD|5W>cFn$D94| z<(pP-P!9fbs&EJJ8)i}cu4qv~zXpr0(K4_u6!T8W7K7^d4F;7zzQL3X))a{E7BRLN zF+YEP75DK=*^l%4&vA43J)-3JCFGkEJTr7FfJYq zC>@~la>QwWe0lZFu4s^pG&J{94f0RA{{@Cy!k3YK0~52q?#qOKFgQLw3UhRFWnpa! zc-pO6&5q=_4Zi0o>IJ$YQL=0Sg+2tE1jucHz6Cjqr&2jBf_?wXihm*{yQ;g9V9=Pd zBvTZBKT^K@eu94*m$lUEw$*jH{QRwr{ipr?{qpSFunC>0bpGi z__IB4ow>}bmniH9S+U#OOIw|5McR>PVP|^jfpo&nZI6VA=pDD=<+v@fgL4_j18>Hx z3Cpskk!~d6CL5(TBeTjH>@&rFm<1*DzWKM2hfxv&(?K+b7wKgQ$_Cl6Y7mE!*ir3QMKVtKqp#6nm7s%TQTy$=WBoKz{|LMF#BvO z6H`#tLE@3M=tvg5z-~LzS;6Mg?y2hvoC~c@~qM?UE(jc8`_hEwTvj&Y5(7HWxm7Uj-(A($FrNrPVl+VYFYe z@FY=U@`{3QLJ&JfsUU6Q51iJ4P49KZf&rGM1 z5WsJy?oR%E&15NL==u!j!MD4qoJUQu8$fd2PlMad2cRx37r;e(=gQN#HF_Ki4=UU& zit|(aslQM?YZ-;t{oUmYs8P&d;uoiqI{lr0UqIJhZ)3ns5_F-~NwzafEEHo&0|kw3 z=Z>vbFNb=(M-m`r_7s)-Q3m-(`@`2zR+00*I-jyMGpSLHqcgLjMxVN@Cn>$Zk?5WT z&m;kkRJE>R=UJ?DdEg;*roaS3G=R)`fx-q6-2Azj{0}pHaT}SzFr*HyuE#T)g_+)e ziWY|M+5TWy)1^i0O2&DSXg`3e*M|>Y@3SPV#bq>}KL!tWvAmn*d)UM@W@97+z1uVD zoZuJ7^&hM*H8$e({yO(*jMuQBy!7xDVel*sFqS*iz;!L`BbDT2){zWI?oatZ9scid zF{UXv^cY(FtidcD&0=D5uw*{05h2BY^_${Bw??;BH^j-mj_BQ%;l&$H`CmE(wyj6tJ5;#HiopviX3d|HKN^DB;h?(_#Xj2$` zJPv!t+cK+yotSoBl!e46)Wny}UvW?nji2+m<))2E*Fws0HMOummCG92(C z&3(Q+c@{qzHz-{Kb6a#i!AxsrO{+eQO3-$Rz2WSblXQ8~BQdRakj5qHL6-4F(}8DA zN9T6pgOl1ZvDP@;mh0Pz^V} ztnjB3WG+2{u@Z`vN0N*T)x^|O;gubDYj@mVDC*C16>nodK&9@Hp2q=*>WCz~W7)d#QG^4)7Pu{P5l$_g1WDLP$M1Eh}B^fO4^2`xFCb?9vGC zqAx$BSSd=uJH{FB+MWv4W4=%@-Xspp;@Nu|@d3wY}33pw$a)GdhLtlCSy?^OKK zEL}V5kv_o*YX9m#y?fi&9}~LwP3x9*>@s!XhTdkKTLQ|5Z>Cs(Nr1DPDD<>hl*7wr zk@TKBUtM^;7-JzgJbmGgju^Zyz{Z0|XS@QixOGp;in(~7@M@rJgkq2>I2VEo!O zJB+CLIaa~x`RY;1qaT`09kOXL zs24vV?j=-Q`712%E`jZ?`CZLexJ$Q8?WR1OuS3K-+k%bw^1Oicc?BtQr@MvSk`IXy zlRTBb-)NEaSA;cwdSdMw^UIK5igZa8>svW45J-43V4!k;8{ktY4sPeLD#T#=U-9i6 zmLci(7hSAP(p_3=IBChM(a`x7`1+tc)@{0pdZNeAKA=pH=3IT*^C6la@4P?)7--t1^Oz?SOvI zqY2K7uFr~u)6Nbx!oOxlxO(xL+YMlzW#{oNZ4rupB-?0tcG=y-mjFEEkqnn9ktE>j z=mvZQlo#uW2>25A(3%EKP~9zac6XAsFZoVx5pf(E@K(jbZr@S7zQDEA5FD<^V9li* zz1f zFSB)jR6gu>jYD9nb7VPLKku-HNZ-ko+C9n#xXqtv7=Fk2;;(^DVCA{yz7z7uDpR^0 zzZ&UZ9{7;t`^*XEFHsVF7Wil<3wD+@(~D1^`n6Dk+bfZ{Zhe~XTgNkM=+?aLbGMx@ zou2RW{z(Un0C24_Ea~#LUePLxhRLemmi_t%r#}G5mRsDHQ8)qVm za%Ev{3V7PB96638xzX1v%G@<7hYa38q74CB0rD0aPm<9&H6=%DtflXZ3^K?dldNM_ zXf(PilSeS_i+%b1A}=ET$cuxEvtjc-ICJ^-TYUEK`1{w(@0U#lZ!Z6h->#bs0nFvM zi#Br8A=vupKbOB>e!cwp>wo3TFVbH87OYjyUcTNhW(&?4f4TZi*x)W-Z~xeZFzN$* z`YYqF7Q)_y&|V1nEq)I1Gvs&r_Hk7j;z)JGUr0~#Um6ARxrL8^ef`@-8j+y#F2b6( z(VuCVk1K@u1@2`9pVm}qcU@jR2jjseR_2Lmfo?@B(keLXFJF61N`JW-V{MPi`tSv8 z%CKp0vVfpISICJ+l~4|!;7L!FL%7Nt4F%2kO@5;Go~TzTwqSf9J}tCO zThmDkN!y%$ii%U5&VR5Wb4E-XgXl6Ditg-3IYnejW5pWI_%Pzms*kI+z8oKkc~thK z1u{VsjAA@bI>$(Yv_9P9h8EGpSSan0c7&h`EV>|_c18RYND%P|8yrn+;GiA&PQJhr z4?2YtaV;7rSwS-EplJ>qVR4145^;eLoQg}SEoeuoM$3ZrqkkccP;gc?X$zDOQJx<3 z6eS~raLW{?k-yvfhkukviz7o{2Yt(QMpf*eF~n!nSXF7nHSab9T*%2-B&yMj`#Z@7;X8>m zj)MHg6&-QLzbdvW;vd&?Hc51352bD4QnVL^fFrpHvxxv1XaWJZ7o1Z`@8xeRy+^Q- ztoo+IQTiZnBCU{V*eA1Tf;8=53y7!56jj-eqGg>e41Yp9)~Tc(AhKo%VsAi@Ue{1{ z`$GmQDuqZ`#Ejt08yECP{04?!lz|g&QSBBLSN2}>mAUh6kRZ4fE?Y1%Q#x9uBU_t3 z$$iaBSWwjf=LwfbjUKQd!aTqk`r+^M5eV{WS5xaioC*H2UG))u>aYpd9wK2zvNgTW$YYzC_R%O z&4LnJ+iGjmG!pqldy<~EXni{<2WQt-l+;6($vlspC;m!5kX{~*SreE^RdJ6}M&v(o zRL~$O3j?=S8*(%`ptI4E%}8uY2OGu{C2N4cpMS3iq-f3?8Lo!V(f1VW6eL`v(nFD+z0YV=0n3I#yHF5Lg*qEU>~VmNO&uwJ@XcDG;V8|h4;k*=C$tJ zxX>FiFiAd^$AaF28x+Hj@SHL5$d(zS$iYqvm^#5X9H4?^{MU&lf}9=fHTOkttl91~ zyMH>OfvFhg4$l-?V`1(&HZ_R$=eY;NElFcyvLA>(Xfj6h1H?+QHTlz9pn^D+377Ru2r|WJX~kuLe6YF%;b8x1*Qbp5|4uLzJH;zTgS~js3mOMS^Q*0A}G| z^Q0P;TLNLJV97yj2d<3z`mR4}1^#@JSn_JD-l?^!tW(;BQVSiE_Aj2!}V)>-sK&zr(|OWh;i z1zCtwLwMiGSR4Vf7`D!wjZH`w^hp&A7TI)Ge!-^aZLP!wOgE|eV6%fPkHM9B6VazD}Mu~{^MjN zVN(t~45raSv#}+XEPK;JF)}trT2Bk!(7Q+b_Y-zNS}Yh(eEKkkyXGqd%pINVw9`hL3THmN&uzwF3AyidgiDl=`-I`lR z{H}%?4x<7&lC}k^ok$cn!OD?6gM6zhSwq-25LWge_N%+1)~f;WsFP5zYH;1fZ~>;L zpuGuYBl&B+8%OJ-uR}VHGazKELb~KCuu#1tTPyQY6IPbin2S$csaof?pPVg?Ak1D%%f1KChEL9 zeW~kZ0jV=J=P*92>&&=u4xK^33ME(R#O8f;qhVkzO9614g zjMMWt=gMqd+=DuM8lrd}+mHSiYMR?5q@=h*!k{6djEs zX>G`y?1JIdX?7BxPb_DBY7la zeoV;9x?ITfS{$Wib*8bnn}qd4rv?caL&mP!h?yXseXNmfpxrQ!`3Uj66G`(ZvZb%u zRx$_o-l7+|WUq z6|*-v5dgJ`_k<{#$)#}>1M!SYO~dobzf#3t($K}!<*lZ`0zXkclXbSzs}8(8con46 zq8SlGew?RZO{bG<0o-w|eJ2Bdk@htL!pq#2Hl@t>)dN;xvBYZJ4kb>tr*_QQ)TL zJ8VvXtizs%-B$Aoh^?Gx%|0uIo!f^CO4KYcO>es%RBO&oh(sKms1G3Fw@-kIm8VypDeX`in7q3h!y@{awEJq^0r*Z5UgM$WuUqbcJ#LD+n zn^4g+eaZ_adWi9-jzd;9qQzo6ycP`x zg9giq8y`~lk@qVN%v++|3oTJf=uy)Mv!Cxbox?vqvPm61eIDGLZeUG1NQB)?WTlpJ zJ>$>U>uMiH)P4g_VO`qPu(p7mEx9;SC(JUZA6V&ZOi&XFg4L5MX|ri+Z`S>9r{1?F zl`VrW^6q?wU3jKf5Pxp-CUU|#h1+r!c+@iN{APbD1LKpIbhICBa98VS zj@T4o^wtfnG&K`AX0cPE$HAn*0}-cl;B}2Ebn_8Ge%|Q9;eWGjeJVXEYp20(hJGYq z3jDfql;|6nr`bO1d2Wr>VdED+MN110wRCMr)S+)rIA5OWukZW4*^!!aV2$k6 zhsM?pBPgjKrdRjF-pfhnqTpUg728y7TRZjp0JLho(to_SGG{o#*evfd4QICFe9p&& zASi1$Yd`1K3oorRt3R>pM%PV?RM8y$sq}9FME$n_DKcdLUn0^>4Cxnei#}c&OsZT% zlbbwA-m~|M#a~C`U$IDK4Hdsr7loLH_x;?4bdJpKXNhJ_>Vma%Ev{3V7P3JUOrAypg$o#S87=$vlb|(0G9X+rVZEOt!nh^faC$2mAk% z6z@G4TBIW`FMd{`dYIUdGs8 z#>9fZ+nD{C`}Or1vVGLIU!P#pL#(mCj}F(3#+YQjqI;~?sWZd=269`c?VKt!^^J=U z99!o^N4wOS0lq$c-JG_drR&T3JE}$M+XnQvTqOO(aRG1+9oIkWs5-B%=)Y-of0|qO zOn}Gr{h0g8|MTNN8hZU!Af=ApHrC+57#(p3KEVTbZfhHSyCbBH-1n-h{q-N=qf*Nh zlNE8**PA1Fw$gHo_--+S!8U z6}cBZAsR606P=evN#-ET;hr=Cf5g)Y^PtB*Gk1(wqfz2g+@K_=!hpzd4v+605q%e0JjX*j@(FutAF}zHt%DOjfVmhN7NM< z_a0E>T;;@_vs|8}0I5wQD;Pf2Uz&FPu!i0&II-o&vmL;x$s}&L)p2oT%09J8eBpDb^Hw#%gBlfHLNq2#)B%4E&LQ!Dt{-<= z+V3cj*fxJH`D~~ILCPTn0=;;e2IoxB$~`Hfwgt#*Y}7Fh*J~($;#noP3A}dUYEVCC zVoP>l1>uG^uL`u2&k@#|e>C}lHHf_TjxEV1p{}xzjA8!|G?(^CT|zvbr@!q4T?@a5 zxOqsK1yE^1AFJ>NGlXwwmHO67erdlVnidlJU^~eOtaz#j{07$gG?dT_g&Lr2DHIpZ zgqnEW8R)iNDbk|Oay}0TAWENh%ik_%xWd%4rpJgKiMrR3SC!8JP+`9^AbX07l`JCDm*{& ze~9Si_$Y`;<7{P`f3E)kZUOuzs&$)XI~eB6k<-(ppBIOh1hlQ;3EvF*PtCC%Swk=2 zYQp#^eLHDu&oi|3{`$0!GYj4!KOrD1gXYOr&(nX?{{7tF^j2c1K_?uj1{tn#Dp%xQLl^~m{(-3c179ITXCd+B1$*-9N`u{l|vp)o&2e+8+Jpmo`vu;-7f4{VLq z8@FV$HuZIgJZ^oBi&jpz4<_xJZ2kDe@c>en<1ZCBfW7r`(6sM@@(`-847Z_q&M-3q z8Sut3e$1MQAYf|0+#4!=%(-_|naw{T<~ia4JNxa74`MOpDMdH7P_Y^wzbcjT7@AKg z?o%OajF`+jf389swPI7Hw36+7Kqk%eT#C=yo38Li`~Z{b4tdLES6OVjgV$*B?hY+2 z#;khzmZC2H5sI>#q_|hiYQK4X8o}NU8T2UdFh`aH4>R>BlB*U7f zxx$+T@B;_YO<_r9U-}aaJVZ2Fq)gPe%PfQm!U1X8%v?vXw2#CI$um&~M{u6Ugon;) zxbGDNe|Jy{x|{?sChkb}B7i|*#Snzen+oOO5OM^^%al%v2 zGe>bavxp3I-B;3@~UGHr}BkYcM>nq;p zIA&=vPIQ6~Ta72mU|CWHaia2;-2M;8{rSIl5dU(r;%59#c%fc*kYY15%Y~h5KMJqx z;bdhSx9FoEu!b{^l(0TX|N2RX53}MP6N@%cFe_VQJMk$6MtKeugI8zR8R;ywEz)b` ze}NiXnwn^szmQv`J)cNL0QO)_FJt>EyXjyngS)U9g=-a*~E2w1`n;SN*tKsfPf02u2z==-|_1_mhJ8-9;7NU)2hKF8EV0n#M5{6X~f=D!?x^rw8C0ZxagL&36YXPkpy-2* zn&&bAiuN=BVlX%We~NA@_>Z(D^?SIf?uSp_sQWcn$a>gP33A4?YCD22 z5&^9s)-6=}DHCXOGD0e4X#}X1?SRn!gfWxm)kmn}ihT@=f7q1GQ#3x37Yud_sklS+ z>kzqgPqO|6=dN{syRO2F3J+xe&`u}LQ?={1qj$Qoi%q>$u6} zEToHh^WdZZe~msZQK(b4*)|Azmx_{}au)BAf$H7UMcL!F9tBCnPP8^7)w(PD$u#!s z{NBh6O6P90Y0o{-_I(3L!#TiW&MEXtahQGAXw)o$_9(yXjf4tu5c1bd3Jqc_#m!dxy(a_uceSybFybs#a8_H4A8z2v}1jc3%DJF94q@?r_ zUWH-9SEZHi+vkgFb1y*_m_(r0x@n^vfxJU-l4?`__EW7$%^gK1Rnu#D;m1C0f(j-< z#x>3bxpSwB+(K~@p3ZXAufv?s#~unIHoc2ye;Y44gri)m|3xmD3W~DaH@Zb&y>@bG ziZWYMQ(K48&$j$E@G_rVEvI)n^BD2%Rn-nRQ!HcCL)WOLfZ@h04&Si0+xxkN*Hc0h z-=(SCf=@Z6WTyW5_O>bVIIMeUT@_8(_mIFojcx3!yVGpiyerP`An$+01hDGv5AQU> zf68P#&Oi+FDLJf7KWPLDeQ@ve@K~QV2qlmH1v+rl=CU2Nc0{9>`t_NT?B=S$cj9u< z&*A>|Anv5y%52fb498$B_kw;ao3)PdAbQKfL1a05&jv$gpGBPE1N#q89d9UZ4MrNN z;MjkVP2yH`(I@D^SO}Z3-onZkS8~nOf2YQcStvs?53_NV*B2q^i!4R<(DSbWopkkS zzu&lF)DaKnpl)76(Z8khmp~g3ZHzOe!u6km$h3e?d{%ISE$Z6^JDw_H2}Ga)v$YN5q8Ap6=01* zJ{f{8>z_m5%o};xKI%L#e|wJBC~}|DW{4Z}mrVeKFX^SL!aumPi^brDrr`^@`1EG= z*_ZCgqe(9f7`*I;#T`V4rSGka2rKk-SG|N-zc|om)4f;dMJ9&+_cT(!dbh|@zhki9c2n$lMR(--rzuQa8f1e9@)+|CCqu;6BT_MBWCgruTe^-LX4RG z8=p-;DB`J+YYSf01JX+%I)jw7_6q_TM2yBn;b(*^pkw|P@{VL-I#S6zcy;$~bB7x4LJ=pC3xL6V?0TP3?v%vBi2 zypF>Z2dZm7akS@+#7~)`2l8vZ(`brL^oI#S>75$J$crJ141XV?TU*e`2frl?Hk^C~ z=Y4Baf867u@KY$fi?7MUEr;*9wXrKcKjCMZGWX^6rQJk5onC-LrUMpN2A8G7qCQkp zK=hk6Gb+Q0#Xho7s@TL#*(UDQiR!aUcM;>k4M(Rwk6Z_SLm=7dgA|4eaQrD2(aYq` z6!r8cvu>1`k#WnonR|o58iCWI8aZ^0Ite*xyi5NVnViE_hpF{i*q=B`N{oE6e8Hr|+!f|8fIgQo zwy-y)SRLbykbM$J{+gHc^Y3C6!a>7);VKuTpE=y* z`rr!dlgl1@wB%Msa82sl!oCZXexvgDGVJabmHC@rKXY=4SUDI8T-^5ec`S(?b2UD; zrfIYwHgU%DPZc)H@803`D(wR>PC>%&&=G@R!}v}z!CYRezS3R+b*~IJgbcqQ&$q)s zK2b#7RRedQ(Wgu&Pw#S+;UCV~R-HmJdqh#_5(modpIc%TnEleqCxaHh*io^LSVayR zHdg~I{=#~GAhoYZUpXFT>8l&b=lq@X8M{j}@_n&5iA>yh4LjXv6RZWSE996JswtK7 z?&eY2W#eCrRz`ayogVc}PSNlrpat-$6)M<+gr@OJEY5!J6P+23=vg>=eVpFzmGYdsp z#SG_yPx^cH2zuRd{Uf(7Ia#Y*nhVNkk-c1^VE<9!Cbv>zb$dk&tjMB}Z zib{ChHiPfeN~Ijk!066#^g(~Z2y=rM&xVR%(8)8mWUxumSl2(mcfZpz2H@N`KE@+5dh@N?_)Pe{n3#{+n@v^r|IJ;S8m1hRN0?9|L2|VEF?S+`di1RKM5vB5! zobIJ)sb%IhydN32e2EzClM ztDE4#b^|Fc`>B=jv?UTB#41_bR9I?H{1o!XKf|?yt|7{($tk?08{!?9D5?_0>XL`d z>PdC6>o2J1I|1mql;mD2jD}P*&k%m;8*3&MOij`lAj2!|k|jow>uFrMn&!?RPqnwP z=e+Qo#X#6c+LnFi$ZL={Dc>1$VW2O>P@at-pDipF-sPCTw6hLhV+I}O1s|A#=S(AAg1%3gUq;OP)d{! zV~2Ez7s^nV20y4wGr_p5*TqihXdbF$>-F?Hcr;wAn<$FhQqmA+h?K%=FLqoXPA zv7S@;{79@8q1p$|S?IjWTBji7n1A$xnQx^){{?m!3Hyot$d*88ShH*r8y2I6fXT#J z=!E)_pN$kN%59we{Il%*JXhI1@O1(8lGzkE*`V=3BJsDF&~_oQs##%x5MAAeIhw?R zjJn1BfrM~t&Ca^%bO|rY2=Emt2hs`> zDj)?cAEXqb&5(j!xHe5fA&H@)!Evp}Brby&LQign6}j+F5p(K6FyN#|^~SkGJrYO} z9*iy1M9b|)DMx2-Nh*6zP7)8f^xx=bAnyREkpoHN^RAmI8cEba?^6;V&A~(&B=6|B zszM?e`UF~w%5+bP7>*vhbSnh>>W`Q=AQOtr0$cMUqP%3MU2M5K*<-5Z#U%{IQdWR&9>EKl`=o1C3eN3}{c{KWK> zQ!NyJ`>W+@4&ra-BU69DtSMdupH0l=&WuR>{g;&?M6@3LU=I0^a)<;UBE%8EI9of@ z`{Y!jIU5JE3POPWO}rH6$xf?Ho)z`3*I<9$-H|y&m%+qky-UTyv#~CrS;0!3h9&(5 ze)HpeaUbijQ9LUVqa`fomvio5{kPKRf+J2zx zxN8hmf%-fQu7wcPWMsHwjZ7uL^o*MYmOj+UEm$ZsEoAKjHF?nUMC(2z_imgZxLQ0) zkjK8#qmjrjk*EgWG`%If0GOL8j_`TpM?(Q`U{;(+DOD*T--aOHeps6nqjY$kJjQ#cF9)!#hgx2NH{RlOI;v)Q1c?v zz7nn-pcf8OVXe9w>-;lrwHL{4yvNfr_L`apR5tRFRfJ{NKdW^7T@95SP*XO?J3b%j zbH6WLWvV{oMxO1j@|Fp}KuFN0h*=JiXHc1PhsVGua$>za5C^uUr%c4MG#`)S4EEi& z6rEPt?u0c9U9LH)zmjP$qZDTS=v#3*MDKXbhjIW0P)T3GUMDijaPJ>V5Tcj=3rC3t zJM<~hX~M!~4=^x{Ov0zXiIE%S=EP9J&H3#bAa7{VN%4Tu$-- zdd{r*`Q&0VmkEq95|w%({0;bdk6fbAXi4@!eofwhmB%7y*qBE-N5gE1V&%wSFc?AQ z4vjxAqF0#ZEQ{Yx-e}Rf7P@8Z3m5ok?#9O6G8|m2iH1n0AI;zESS8FbnXo(-fWGTM z5N4jiGdRT0pXd(g=QxBWGwTykj{@q(4e}5m3&`J=C6a?>@KkRPZQiF!sWr7z2CLWL zIO=_fFjmg1A<;eoi;l^^NL0APqUl&M3OD8l@iM?G<}XZ2^Pi4yTGtZ(>9^hgh|`;Q z0m%S^ODGlk@W-+Mb&~=tTVLdQ zo`O5tf3*P&U?9xRc0d}+%eAAbnvsYnu)ZWSh@OwA#XT=XmJKXSy>Gokb=&chhgiYd zPn$RmS0NeBbJEWAH6L&zoi&9C8)mZYa~V3dW~B6FMZVGfjT^mkwhIM%&zxH68{Y>k#!EhC!!Z@= zxkaOLW;^!kksD1^rJyghp`j`r$sV zGytI78kUAW(&I9&!J6G@fTa>wK1vaO+8XdeNz=!5(_ekwu!uZv_cDz9fS%E4$6@tW z!S!7j#8Z=D#%Z%$FracZKu(E3UbY_%PaV~IZJhIDaE1Vtlv49h90}hN)BQ_2b|*xvXR(Ahx5yJ+UM_VzGzu_E z#>s{DwCKd{xaQTgHBMr5Yo1@hz-2-#(Z(gu_e60&8u!XFlmH7;5>E_#8Jw!4d2NP{Frlt=NsG}cfv?OXCM?J9se+=H>J zvUn}f*q|k}mIhj(2IVqkse_4NHANDVal~U8hmS8l(|$e2V;H1+2VHwSRJv}Rc0It)E&Q9| z9_I;Y9l0c_=SlUtdS>O3vx}}iIdGXO7F6yvJ8CKm6hutl1$(L#zrx!JXFK)!_04qd z*lwL0THrlIBz`O6TSv}o3*S8Flyw3bO!!1>`%u2Ob~%l63s&l6%$lTPgND?pzIq#w zpiyF`^zR&$(ID8F#?>E~<7vQ3Q-dD6-@`Vy#f*6QVxw@M@B~Ynq4(iK{0$u?eeC^W zG4F)e^dVW>H%&M1)u~bh9p_7q7*|#gOE1dp8D7hPy2CNy(y$$q4J7SRg>p}6lf3r$ znEEe$4t;&H+W@`e2+5RIR9|(!hps*iGA&#V-f> zc6K_G6FF?J%-)K%F@03u<+f8qx(KO}-}i^TQ<@%g(4M#5;CCYB$iiFOINPI>xo{zt zZ8~Ut_yDbathJ%h#TtNS@3E%?nhSI{?8!(y+|Mzi+T$I2H`cX`Hd4+U$4^x4u-P4Y zSgWuTh+f?{+a426eH!|E+;OY7W~|7vzXWy8qEv^Ly^r@C1~Agv=%<0nS+L9^`uj@T zdV?1{7h?he9X6(xzYu`=a}OM_wvm2jf}srl(P-Tg-5`Pg$=f(p?cB|f(*z6cgPNmr=VBi~d<`6$sLXNnxXAum7pYm0Ta!K8?X;tlw00F+ zJ(PiyJ3eYDBps0Tkx*K4e#@u!#5N1*2k(w?_tJ4k z$ajgXS<|Q3*f!+46t#$>Gb_itzTpaA)NvRsk75(wO|)NlPvc1Mt@Ikb*!7UONTd=K zIB z^fRI(WVbNqf@naE{)+aIE4#?l9{dhOL=@Y{{GF>3#{89v{In%_Kj6ASo7lAao36fd zV#+nCb{yb`qwgP}47zCeV;)P{x98Qz^l0j+Kv_#c z@{z{vJ{39N1qO!v^sC2kR*L)6D=^(xu+Up<{f_qYbX@guFL0K2UnQVpPr+S8*6dGc z1@xrypQr&j&6V7$?$4HWX4SOQGbq}ch&)jrKSn^z{ML5#_DveAw0tq3dvyl~b!Y=l;3;M@J zBC7)E>>f*sudY{wJw|aaL8JMJG&vRzYG{R2jHlG_>3quaQqyWmcrApBDI_yy2LVob z{SEU;^o2}p%F6YBNT}VZ1@r(F!17<3SgsyVJy|EpKPH5J1y!H^49ad`^xO7#X3$s< zQkkS;9RABUl|z#jbbMM%$iN?|zx+{VxrP|-@7wa*i?52?MV-S5?I|t~?L^-K`Jcxh z@9&rO+YbGoSB^UN+UBYAmmB+jU&?*&DURD$-|ML^$~1ubnf`EBZP!)pVbJ>5_6ybNeTn&R+cKUcnorg+xD^2p<7D?#~nW5yFx z5A0pLISFe2K&Y4g$*7s}V-e+Ct^q>?ZoImpmX=zII5YZ|D{Ex?3tnHF35Mj+v57nQ z$bmG8ssw-`l!pBdbW7ESXSjJ-j4YW5;sfn|hv^b7SNC+yc%$Ff`-afj?b=kg-TLj4 zXIT4e7;D%$&~~H1i?IZCMZFEwXK#SiU8+$rpJ*!;^9deUutbYpd=(uGe)P;;=lL@z zhXA>RHHDU8VcK!uvuuM?gDv*UsDa}OxgKN{`U#+X6^7k;sa>d$&arKYsdFuH>5nBx zH9Z;D5K>{~)g)Mtz<1z;!Q1M6<^_F$-bC47%U0g@mV94BLm$^8r6&45S*|2r7nkB` zfh7__1*aBk10rclSC>M|D4&DOYOs2p%%uByCd>md4EivgQ_ds}gR z<)o(aEbS~Y=U(~1+pc#W@@Bc%$gEKM%+l8yMt!zIU(q0jN0f)2-_+Y^&lF>BiUHzP zCHStd(ILY(x2tRD-B8?h_?VWdlvvoIUj>l-(@Yk^r#Zd;PHGd2Ath8rf%fs#9oLH! z>&&NnZW1&T8SuJq_es}+e{uOb%x8+vj(?HaBJvZ%32Dx&Em~X{lh3c}kJJ2l&$_aG z-TK-(<~F;tRj(8Sk$lI5cX!JkSkb@!+o=zOq1j=2^1?oW)#bEUes8cgvWE@hq76Vz z?AjghG&;69Tb{&sy(6Pse93Vcq*>ZtU3>oU7bVXa(}Mr%2^ZQwe*GF?KKcIF+kA2* zR86D1Ih4V>C=d5m?6s_UuldRRd1?qvm2czA4@8vEHTB~mGv;P6BVP$qxv{83<4>8D zI9y%=OGonvc0`r*-y(CPB59PL;oyKUl1HxqE}=Z#g>_v%F&a$iwv@ z$&s`*Yr0j?56Ds?&~2YiZqA68GlFnvDG62hFr55pL9hEMCjJ}F>V73H%1#Ce2FPn2 z(^DgT%7JDi>$8VM$9&OjJUb=(l?#md65r^b<81^@o=rrzIl{id?@+NrO+yqP_?UB?FFJ zGz<7wkVjVb^GznLyAT2cFO?di#GHfR(1{LWVV9q;{aU3(>Q(oa`z{{N5V9m`|9~E8 zSBNCh<(N@HT-`;plH?15MkW9Yl}3OA+tY-~yxY_v=(j0?f{ZK|_pB9(g0F4T&2H5N zp%l>*v3RCau7w(hfU;4YRJC!qOwfXh!>oAxs}H1^UQqyS25p%XOFv)M zW$Te6wJ1m>2h9bTXJF#5^(;w5xgm#|Si`5su)-u-e{+0#@mA>I5wRc;)U!eC9o z?#I_-?50OQT-O02x{^@%1cg07Z7cZ)e@72Uidx=$1?;1kw1QgaCV`k2kV z9dshvHDhpKN~RPw{VGaife<9$TUbT4@PSE~o~L8W;b%|V`1RH0{jigsnWId32&-6A zAlci_*hyfKy2-b(sw_im#p%L*+HxE|nwT>yX5US_f!@xA85sf42^2vkrv06rtw(YE z6*KHsVvc;(>Xvu5+6HRQ>?qDVn-;}APXm-5_w3X#dF92txcf0-$|ZtD@M6s`MX}8s zul~48i8bCCW_0`1`a8Vc$zKZnM+a9BSuk~rTkks8B3I&L7#cPqwp^54N%0*;d>jyM z*XKvU1VlPadY?GJVb-vbFmVpU78A>iITQ&zs~?U$;XAr4Pja_oy57(5?tb}96T-$} znB*z|^jf^I*~i`h(U3bN-U&+b-~KJ9*m1*lL{zyTe=~>b4ssADa^Tj$KGFVWXsw{A5JjpG)Uy92Z z1%5v5jSyfK9AV0gV$|YT_%Qb(-=2%FC0l}vFJYlU%vkJbU282G9-?*WxJ?{X?znDW zt~1Zm0+9#XIdn&|r_-kuk^wZ%7)Q0~@_u%Zfyrt)EeS#Uqs>Ms1ox8Xp@F5)AYJTr z_g=iMHmwa%-BEg~&N@37wlFRT&uT=w#3XGc98~5SJdj3oxT_E@sCUbgtf+ESW<#2{ zRuHx_b=YtO@5KR=^omM=iFr&L2#!0)uKgfmQ&QxYf=k9pS_5N-9lQrRuHnbVBILFOmY__ra>MbXdV3g%XeuV!{8v`iZjvJkLR3{7l96&tN75DfZ`fkPxpjx zZV?NJs8;0;$^$E_GavWXBDU<~PV5-3ejsk^lS@+ZC2OJ}Lz+WKkI#-fq<>?p!?P?# z32=>dE*9%ptsLgcbwpN7VNdHJv%dh>LrigR(vl#gwUh*l_06^Gmudk&B%Fq9&KtTF zt=JGJP`+ke-+6eFqiHluR%9D3pU|D=Z1Mx7wL~#RO6R(~L+E{~=b=mJP<7;NCs7K2 zAI3M2)ML3ijMGlE-5xxaD&@5a4$rnGvyyZg$E5_O=iFsKD$iuR<|4y~yK-K-+Q{Hm zNMG6a-7r#^l5HLiV&@lRFjR~NOUD=H98YC7V&J4YxS{H1|B7s6xVWzTnYO3yri=oZ zb+H8LeblgPi^8ZpJ@fcERH2D-@Hx(Ii9>|TG1EM%t0crRq$h#<<~iIuMO(NSO5^hw zC&Fo!H90_v-!(U)hj?g#>4jHUL3`SzFn(3h>3-a%aKu~9P;$Uyd)@bpxkSi{=Kz~j zm2Bmn#{yBnU~S&K)iX7%XdEHO{g(zH?k@r14MIQ+;rt%D)iNpAS2|dPtHbp~$y8vo z+Z+$oUi`_lxR!z7W&uC$Bg5mG!`K=BR~Xbrp_0@{ zXHB7xBv8TThP16w)mXhcty&6@vpCDtl0O%)aD`vE5OAT(E;aTGmMso{U8NVrCL1R% zRhZ+UG^%-$)FM809qV*l*;KssHAfYjT|-BPFjoeDNZ@fUt<55;jq0tk)+CghX!khAS=f^V~`&hg~)B9(<- zIPsxa4H)$=?g=AyqJs@}>QuBT_A0|k@Vbc>)yP)>NtH3{Y-dI*<$oE8ahsH4u}U~< zZD2nOlID=v#ndJ1>AnM&4-$?DA#bNcjQ6B85(^cr5sRjmE?tV8c26dB@n+=4DM}gM zi9wRaKU=qot7i(v{ZSAza<))u>}Z0xyjU|Cyt=3ntP0uizVXAwDtfqxcM+|4j5D<1 z^sjq&DPw}+{6~9^S@uvBF@M5dV=f+p!q0v&&H8m_NvUspTS5U$2@_r&Q0TnJKPfKg zk}?rmpUAyB5w^g>@{s{yuooMc)kEU*dqr$qAj)dsQi&mv~baSE8GPvOsmob zmUC-AT&x6^S}6cjXrk6NtHN-3>7og66=dt>9Y)r>+oPkhh2bok2A3Ganu9Qoh8iuE znfLDy@&fYRxgOl>>f|4HHc?$YSw(NvLMH+UV;`+MCpxxQkif7kFS95Ki%kICWNo*04V(4*#ak2R1>cm40tT-3fqE6`;dQNdm>7qGInE=0U1- ziVp~K^|!j@G!0pi8k3fZ(E%nfgg-4?1xiN<1J?do zxnUMZLF6KW&c@db{G1Syq-vdUFj?Uvy>zc3_*^G`3Awspgi~*GNK`!r5>t5+;RCpwS-{~)@Yr5X1AB3M7|iNu5REEG zVKNrVr}C-(EpS;=EE3m)2qzZexxfP7ucY%UvCV({y9}qcn67@^@zof0fWU*X4 zDpm_AEX*mXgUdV*%F%+H5F>H&(j6-?%?>a0Rl|GIGqV{Ec@2r-AtYs-4INLIHZnlk zwySOfm6yLnlh+SJ9=vux)#qy%F=@iRh)0_38NsX}dqDzIW;80cdo-kq5Nk-TWGj8y zL|-VQJ;i(p{WpqhW0!y7W(}K4g*a%J{|sp1!a!mJ1uy@R^e;~=F80#VLk?4i>>nJ* z;dg20i}B5@)Yxf~6oply|LRiTMCkzCr5JS@=!PhXcb2E5iWqrAD?+>0^?6;{tTqbP zzx^>_pW3on6rY^Hxcz+GuLyb#^tx`e^e3SY};v2hBCpfL->VfPddm zU6xffiyeN+^_YKo2HMW;{b~<SZ%ApAj6BT){UA{X9lPl!rQ^n1@%J!MY$H0%r&lVhZ<>7LiCEy^pr7qikRL< zhKl^GC~D`Ip^;Kfc~Q%7qecvbz@daXCH1&MGgq=h&nG5jp610&bv&(8l5dcn_+!nr ziJL8cX%Hh~+r?-NP#>OI8nC9`D^pTyjO%?XWvK6=mi@fvFjlm=$OD)#>7k&D#)EZ% zJ2!SzSdu3IDv-{!lq5T6%8|81yN8TvT6VUhqKYqrP&7P?c!`>CBR?=o;00Z6C|Ydm zAieyxr_Czg%!)L0R!_U?(*{s-fR9Xo->3`l2k1Jp!fm%R6{;uDrJL^Ibui}aa$TVm#2QLWz-g}M}H zu6cOs zA4^tkeZ5<3Pu}MH%U&@IyG(SJ+&t|IjWh#@zW?dV$JdTRpZHrfofxWe(~jHuBlpj5 zdh`(h%%d$D6g z&)|CAv971GWF;HL@AgaiQlS5bgw72%lJChQOD}#b0M1@$drSZ8WJsyRc7kCa%Y9#b zvr%$%`|I)3(ee7_c4Uyf$vMjV`dRh^ z{8LOAu<9T;zM_>nOY6(P*G%}TFO44QQvsdZB5E=CLB|$QyYrXg1pG?W9xS-@XC}@ExoRU!&s3!S znZClSmVHCgp;Za47CpHXKQJ;>zV)YMuYf6>8``|F4FNCCLHF5!Fw#r5c4I)$N-rxl zLMEF^n>01RD%CTvZQTp;sL!)u>xx#jtGX7&F zz?WeC1p87_pTWAs1Iif?IGQv<0$jxK?ReJSEbv9^oJP9*+xs+I#HCQ%#MDP%+1YkX zr;YrnAYuL7`3T6mKs znjFjCgYEbUWX%BjM_U_<4vf4mTrTPdVA6<03QSf4k8Cr%%aQIrizhB^0b#ridTh`> z9Jz6NE;y7j@TOw4$0+-#H)cuG=i{N-ecjNta1)6Q@WEVC5(gU02i4H|(61B>s8z(} z+E-GJhuBTCZ-a$?+ZT!_1gy`TIs)+=M`C6rbNt3!;3;rNF46nD9S&|1AZC&QejJd~ zL%0EODDp%cxH(@#erPpupIxU`eM#-7m^IJdyNu3AiFTQ6^S_+_$lB#3c8eN~THkx_ ziV8w`8pE^VE~YQ?+jq%8xAmx?_x{3bi{)2vQ7IDA@vcPe);UN;70S%a-xJPG63Evv z-`hp^4CCAmem&8BR*2U2C2J)Ev?Z*g%sx#mH^$Sp8nW78=4qmqO*I|<#Jitb{egO2 zwjPyGdS@k>QPnGP1p?jf#Jo6(e{KUg^HhA_DRm35p_?EyY5g-S<-Qy+=lh^+%`V+O zR~IMGF0nG}cTyhfL#h_dU*s_v^FVx+oV?KTT>M_4!05~J&C|cbZ3F-VHB2wd+letHA!$B z7&F0SLIOGb*NtmOLE)lh(cfEcSws-)S4-MRbH$DzPh~#o2>f`7^4;YMIZs(WEbRXH zoC@*?IiS=CLP4917tQOsq>__$ANRZwD}}ca;`}>ecPmOT*4A@?WnO$p94C`x-6pTC zP*`(oxTTnxx!I80#iiz$J3DS|$K`Nr^S{FgDpsNIQpf2(spqlC_D?H>@I@FgowepT zk(clYfeUNl5l9_qL|7jS+pF8O!+VgqzTdKFOeV{L_!lTD2G!k=C>szZSA_QzoYq_5 zc>06b_WF~FBIgBQ5I&wA;`q5r-fsBC6LCr7WPcUn>(RXe6sH%7yYFO+%G$uXdm3d- zIN9OOqD9Y2>R%a}mNT(pABMt}W`i=Os|s-^WK>lIVkI>rsOH2u{K12_<=a0+3OcM$AvPTDb zhb>3fuxH_}E72X!4!1X8!FX9LLLhJ5r&lOW`pas>j~+K$Kjc+w)@vAq{~iOE5#%p+ zedQ7Z^mdNZJz)ToA9SU^Gdl2~h9MTGpHeWaJ!9L%2_jb@RvVEQ_~Y7{Nj^;ou`;C^ zk`w&ENp%8nTsCBPWnJILOIW6ga?V*Hl-Tknqdqe*_A0RpP2WZkhb3Jh3387Mt$m-( z2t!?82J}aC>`V#2%#v}!q!&h@0Rz*?Fy2yg>Sc#3XVS3&CH*$}BVr*-FrC@g%pIasE!Bog!;+lqe)xga{_@zH-ea=>F;!`)w-709SU`t>a zBDM-ZCix-KWtG>{QMec|_O65If3V>RLWoJv#siBDS^-}ac2gVp7$ZF7$b z15v(fNgRxR7Kq_X2(_Ra=10~|=HtYgKOWeYo2Ff}-HMYGjGv_8tatm<{TeSzc4<=+56PfGk2(Vo81WWrA%J&_+` zFfr^@Ol`MA(W#>9Y^!JIJZ!=Ik+~?0l;M=|&n~c?!V3^Dt<4l3G2nr5eSd`kI*Ga< z{ShX!aeVl5Q-s|67!d+^eZ-EsOk=I(t+q&34m?$lq?vZ6Dy6?nFx_r*s$j0_`lJlN zl{oSAMx6B@lCW+JNJzzTmeBZsA?K)DSTfcl{Y>E8AB%DFJ2Cr(`J}jp8d=jAd8wR> zwSk|-?4R1%2#t4{GR%88T{3_<_g5z;L{!O5zEAW%Es54iTDv6sc4corUbE$#0wSSQ zFsY!UksJ+Fp-$8UwwY%Tvx;_47&8tKNsqp5peY#6na=X@V-0tG(mUUvey3DdaVIfd zyMAfm;>w?L8~aFZruGHqP|LFGH3Nj2zA0H5RQ(xJAob-c_s20IiMLp1->#t6z`M53 zPcQ{jQZ34&6p18V?af`kXrw)-aC40G1##{T_KCnyn1R(!Q~weLvm_7?kWDQBU)y5; zbW&k=Xc-^ftnkQcm@aJJDo26-xrZJnsgR4iK@-q>?qMr|P&YsiK_aQG4U^z2bmx8q`xX2QOcE$d~){8iy_>!qyM;%pBCKh_$YdPhIZ0rmRQaVgezB`wCsd3_A2*4GbKxCVF$(M7n9L!{sZ1HGC3f`O zq~VSD9u|<~{^%=dX-(jXJ)mUUc(H`kLnZ20yEs&W2TrhJuHgmrZ z>W7%a5ZP|ai5l)ntzen!nbFfJ2Av1x;sDB6+dPP}4QK%vpt)PnF};T;}boh8`mS!6WSNa3v;Qr(jxRU(ep+cFr( zKnTUIEEx07VkEcHsN!{qZHQ=>|;Dh3Tf!0#*X^x2!U{4FVy3{B_8XFJNN&Ak1n{!>? z_OC1kK7{+AdcZ!!CoO5ZkKWm|(lMOXA)e(~@cM^#ACNMf-k#0_I7AMj=av+F6lF2c z7G5Iyy+$|RxQL|+fe3J>lR>+P5rVS)6+UJS>ZN2V4S$F_^BRrwfP9#y&+ZNX2JUq_ zAUE9uK=68Y!kxS!4ZMq>hT(P2!Y~9+-LBlI>cqNar*~msEeybOs~9VlzfIdSCh{z= zqcTw+bZ77r-U^b+M%7I)G9i#!MOT_3aY7}zQ4kt*n*ex_kTVZ+#b5~tF0PxX=6?5D zj~CJnSj!PSviY`7I~IwWR8aeMH+pG}<4q$1h;f8)bY0KDJ)kxlEn~0E4YK1jT=Ydt zS97;hL&t4>{3j8Uw)A8&2ng+%mMGeIDEt5n1d`>9#PbQQX%ATbfcPB$mEadBtz#r> z)jat1ZNwTx{%jd!unQaNiK@isif=j;atoU^_V_m)>rDZAVmyq!XU(^*$CHxk96&w* z+o`r;dU*o=D#@oS>Xl-9vA*wA4`WK&aJFR*3w$KiODQyI02lnHAOiLid>rEW;2AK( zW!H0D?Fw@=HL}6KwMt;w#2!j{^8iEnDw_P8C26+a{hqqSa>i|6f3K`rwky147- zunEyI`p@Lc4h@gnM+$64`hOnXJ6k27Im4i3x1<3zJ|K}sfaI$?wviVda#?0rA+{U} zTi7o!SkKD4u)R9w&`71L6ND#5cMD63TZ1rnDUMm#20!uWngK+U@p9bpZNyWMWxyp3 z^U7EN?$BWmW(G(VpQAKmBjgYj_4ZCVI!4c>`f*;tPm@5o9&QV96funTPU{4a`32wr zrM85W{9LNc_i2&3vte!1v`#Y{dKB?G8KP?I4+~$!H}8##)ujLHC;k4gfi5ngW$toR zI*~Y8rUd;SV&kgcElG5c;V*hY_xo1h;&V^a0pEbqQw^43&P;}q(Vc$yN{MA1jIM4h zrt{F?Wo=faFs@uz&(W`=5~~Ezf)4HB1qnHlrn#-=B*5_}S{2tf9prD?a7yLoPE(i~ zU?GJ4At8-|2B%bJ&c~meM}l5RkXZERICe6OVo#u+Rok-cg3Rdp{+>;Z63`FMMpD#b zUiN&1kKrBf*)pp0k~Wp8>Yjl?VZ4Xxfc$)MDM4sY$hsC!zN5@`-@^mI#OuPw_~^1x z)O=E5^ErRrr?3<^-=JDsSGE#OdTstGrDOo-l_F?ge~2~ymR!CX=RhA4+DGneE88UF zI;Y#ho9dLO_xs01g5($aut7jdglz+-XIt)vhhDv|nWl0_+ zGV-VaNxu%L`oL~0a&6ue@I_A2*Xto+ttOG(^ND@uo+gu?@W-;+Ip+87S_sv-8l4G0Iv04VEL zF6JABsDquDQ1IduTPUscMQ`qD&C~sA+rqvT=It_*r%+wbc81eEQNs|2iX2HH5>2#` zO&5Aq-70IUlE*c_P82+#`WT(B<4XOk%-ie!Yb&VH{Yn5LBvO|>wN?7YQK##IgbL`Z zF)^&2GwkdC{6vQQ2H%5rP7YEE01=2$Ns);zs`%slF!RtfoZ_T*VFmeOttIoLg!vi z^Z?Jd`$sQddR%|?d&Jjz9-wDLO=sfz7V(K5G(Y{5ENtJKfymc3zdauC`Sva>LrCp> zs;m|Y?o=isfG`#1N-=G+1BlMEbwNEVST;n z*syU=uGY|ZVFciysOz}YT-!*DC}-~LQFALWyejc7-;9QF8`j+JoZvEHSdualPMG>~ zm}{=JdR}^Ed_UNL>QtbUkP;#-s}b%t*4jl)5E)*3>nn5HRIY{`8Ihjz^wZwxF*`M? z?RliHZv=$TS@mvSqQ}(M3MaE<|GFNS`#G-n>V=F4s{%ln6eC1$3xtw3UhVrW2bMNa zpG@LzT3}11-IZKpbR4VG5W3qHag~)2N{KUycx6dW)?oIVb5ls9)ZD6a(Af*hJ*?ny z4`INExBg$lDoq)I+y1n5=6<4OrfJzRu+;B^S3?&sJ?;o}b%VNkQRvqe5wKT!dq}NP z-pUS%s~6z<3u-@Tj|=(62>M=KTJwY|qBGkJH&3Jm9ErdyKV!QEO^kHH>pi>&3K1(Z zLrN92_csh-N2Dhfg_jHSkYO!Y^Z;DomP3|HXV7Lk=c_=;o0t`83$5l6Lv#4M;5sm& zO|!ibdAVIVpT$9#by76h?JzJ2L#Asl+2#B$?B86B4Q}zvqDy4(gaoLbT!5sCvnA;! zgdILDtlD}0JpcPL^`wUMUMa+v%4&nx7(DHyFSe=?<56@zw@hd7K4yUglDf@WwjQtw zda|%0UCj44&s{H(C!t!(7>nyFBqssIO0F!A15p=L+8I-Q)Cr}Ndz8QCs%@Wd+3S1h1p|Hk7nr?mFYnz?d$qvyB^!5RdCU~| z`|c0J9P6!ekGnB6rw{fMFxERdFk)7n6md94fFuku^HFrgep!6L09x6rXH#+|F>X7m z(m5-5sd165*%`t6Pok@aZn}t&SvEea3NFBfJgbotDqYQodfMU5#rKf7`E<45>0Y9 zmBa3*7_Sk+8f?jJscNTFlF{cn9!)b>lZRZtb?Gy59MP)0^e_8JEsqs{WWg)znel&5 zkq~Qr&+1h$7I4+bl|KHbC`nLBxKKbD^ZT7Ccd)=J?jC)y3I_5+SUApm4Mtm;==+8L z`Qhcx!5o~UjF`j1+i`Jso0~yx6YRv*66oBmsLNmbi?sCKFkuj`_8Tu$0xl(ofj4AI z-nw`>GvFAYZpxty^1@MbyyB#;Ua@uXx%$=cqPSg8_q3mk<^XEnQ_B(*;C@N^)IE{E_tDx^tCVj zFQEhcIp|5D6bT_YtIP)OCYXEvo{!6x4pLlno-u-)>w`+wDP#0^%jfyLgRHP;#uLWR#gb|2bM4Qw)q4 z;vH@-L&DydIEskqH*t|xA&&xE%89B4oz5x%Je!o8WWpUPrCBI8;F|JBHW(< zNS|P8uf8TEH_ZwuhRJU98B1nW$@ZJ~Z0RHUMrrpq3%#GLFE3R+97BBb%ll?H$32=X z8H$wB@sDIV+JOb^Azjkhb{)aJ=K8LJv0O4pAVcWM2qnbXObQiHap3h9WX9@*nce9v%U_nkB`MeTg0F5wI#>5>nX#F&gZ*BjMUK8HcohbeF0Td6w z{YOSAaeMf*s7?Nj-bReJoA(aq?E=bhPLft&I(*=q=rBwL+Xxc_XF+$FXV0}vy7zy$ z`lc|=Vx`@-t!dk~ZQHhOf77;Y+qN}r+uhT)-F;^N=UnXnT+~~Yto7t3Su05;_0}^5 zOZy$t8Zuu|;UN$lGD%G^{9eD@N1HYx2%09K5Qz5v`aSfX6GGxz9(dslU75j=wmA_F zEZVqnxepyffZQr{#f7wbU#Tl;DhnAv1!zviw3WUX?&Co;AdizDxr|qtC$fqc33VP< z7$q4-fz?&#;T_cDe2@DYLD+dgTN=(aRUTFW8c$M)NSC0;?P2@lbA~56B4HMnxMd;uvop66`9tc`Z3SNrVE58RtgsNGE@FQht8I zb03K`ux#ff*tD1mwpNdy6@H~Mp&lF);u`y# z*~(dF6~#56Vylx`3^e*O|AHBizMe%{95<(kxhHck1Z(0OSQba!XcUpq(&#&gg zeU#bC+h}xHEr_l7^H&}W-22a^M+oO&78O`%84$7e^xyu!>DIL|i( zo8o`@2EH25S6Lsy|30t~4mW(Fp8V0x%~8Mexjqr3cTl)Ld?q!}AJjzyCx$z z_E+N&L3{FAVVtzmoJt*}$!Bpm_a|&3=%$D zSo*@say;*UM_Yj6{!BAa4fr@-JevFPXMiW+KaTC{Yp+qC6@IjO3Zn8hgeyB%3K&Xy zw0dC3k8K~W4@On}{z{y<-i=k$`#gpKrj?QVdfSs z3#taaEx%N?{Fd4j*j)zW{})MAL_;oyneKgZIu;bl0E4>fRk5cwjWXtMV)&b0tBy$T z7%!s5sJi7TjV`tR#dx#(SfO{x@pE;Picx;=0m)ytqR2=Z=xf5)cD|D2lR%_5czU@# z^3a>piL$Bse*0n>K;f}2yyXJ|Cja4ZCvZZ2yFseB$~vb01FYS|j1(ds?N_f~R#YvF zpaGkkfUTX>5hnA{Tt&pvj*9X{c68VW9^pcTB(=L(55Fn3limo~9=*zxiI7gliHv=> z>g70Ds-*JJdv7qMN!8!!VRT2P8wpAQOBp?qvdpIv z#h#9$C61gI74=$^Az#KS{KKHoEA?r_PxfL{$7GC+FvLiv_HAy#W)P7IxGCk_LR&S? zL1sTP@FhkY;3@#V>eBJ-n}?=adc7Y>BPHg#Nf_%34Q{hYtRszp?%bhWdOG|b3@>7X zAU&mD1;hW%cDQ~0Q(;P&Z;?5lJm{T?%O=3uKEj6083>$0a9~RF2}_u<=eD7DAe`1Bp4+G_~pI+E|S4w>8K<=PVr1?f7F=L z`GUhf?K6K$`ZH53LCd}+8dXF}TJ?M%KQdk6X(iAZl>obawEpZ~Ve=3tJQ=G2yo%%; z1?r39Disj=ZR{EN2t_^A^3t$vFAZaJkW|L!GNbEM+B86PWUW-&ga(hy(SGRX)al$8 z#0^ppID<0V)Z$6c63TSy6eQe!_Fc14{`a}}7r z>+KxQmw>*jY*d_A!i*)VdK5M;Fy?hY6$iD%ihJOtjvfr3DXmPkG7P<0nsZIehGpHP zYHz`qcmN?5WTzQ&@Zny?=;hDx#;+NJVVO$^(5>K8#{vops}R$RqV~!bQmPpm6Hpgl zp7O;ld?rQrW(c!BQ+82Pli};ngf$rF>GXK$z;Z486GD0*bO84X>H2OP zKzOCi>c9y|%6bXuoByqnz1@9ugNT^&INdqs4p!)cqJJu!hW2DPiAE+FzW=XtRyOLI z7MSA~U@bH@9g(1ZJ^GKi*gqK~{$1npZIe%xNTSBC9ek=p(a}AtW9Q94^!HTHL2F^w zoi32ElXt?#A_I@n#`EjpxVdWAKDy`!3k103LoYc*UZXrh$Rdtb>aeB zWDw6tbq}T1ZO~8##|>cJod%`Ee1$;QZD9hBy=){@XHBG=e8-QaJYTrbVNE~`fAZ&w z+AP=K+HCPUlIcX@1ggYP!=xOU;Qw)?1Um%MD{Z(&n(6=m$$!mB8WFt8F=324 zA(d)vGk|^@i+cZT1YM!2g3xwMBA0fxOIM~}oV+px-6e5A2!&HX3GZ&NBaQ!2u>N7H zOJr3*ZWV~Iaa{W3YIpW{mxdjo6-*2ylm@!O6`NoR7~5)Gd9nzEhD6Whg+3r9uxe$2v~ZcF$9AtnW| zaQwGJCr4wlc8eWu`-P@^N-vs3sStl;DXLH|;2LC0Z&U0sO<%*2<+a2=YbTubb`6CYJ~^NEkn8 zuDbL&$nX2_Aq!3acOSr;WFLXS8TzjcO+B{~+B(G4xS6_G)sv&A@6F|o`0Oab7o39K zGv`ORO)^yp1Js(q&F8U;y2;tPviOEtDpN#?KLBcMZ?Ra{0g;wFm-Y+{eW`jtRGWK- zS@7ZRslHnb6Bg*CyA=OCy=Wq#lx+0jx_jNd#fNDOqYHD5xfZ}3$Fvn+l~|~$&R((B zPNG6%#BlS~o}Vm*6;{&5^YYND_tQgG7+-(L?pcu88LQ#NXhgF>On)9~yki|^%w5Y= zt4K_zn^JG^(koih5-szuUX8NsD&xGvFKCozy9OvfdL?zbF;s1(;V$gqTd3NpH*kWvW2bD5l^t5`8CQ~YY^y|XCXmoSa+z@1$kAOnJ2d2m%&E0 zzZ)Ph0{)aP{{uc|_(tH>_mWbPl6Ku`x2V7Sh+dKZo3=JV%g*VfaeKE{tJZ!$zb(8lSS)EWVG~~qJ8)TAb^xBGov5Z)2 zZ&Gwa9JH^Bpz6@#E56d86NXC1B)nh>jLnyi7@`3zm?q~QM^VxYXNz$KmMp^GuzQ^!Zuk!@Lf zS0OLP@z8hQtXMDgb?Xe>`<$3r3G|5#WP~Q9kPgruQb;`gAj|{{tnuh#b$<_U-bAV! z^(WIhE#DGBJOR>9fF8O2Savweh~(!*l9?c^L$kY>r^Hk1e*?OLxSRnFa8 zu>&Z%ciJg7mYhqWl7KmVHOw#hE>|$Vu9x@?;K4l4%YpG+*-0~6x@KU*>N+4yOCc2Sw_Vo6t4RJ#;AzW38qUHFYT)cd!wOGLfWrD2p`hzr@oDe468>%D zU*67(>*V{>=y9FqBM-eTR@qb#9ibEm5(Bs#kak&}?zc^HIW&5P9vd{mw_Bl>k&BgS zW0{ug(YF)8@hU-<;35?q9FxGYxRP4vJtq)?T74^TAq~;{LAjrt*f(_XMl$DtDiNKH zxrt}?Y71St-wdO66qjBq<2G`y`Mx{-zYg7F2Gb%H#2@ikYDv03tA z2j4%+2`FUiXb32 zkS3PyGo%<-()Xb;?xlN$`#EtQajbsdJliL`WdN6N*Q5Y0p+L;Chm<4e&urscd+s`c zSfCY9q~jBiz17dK&)70htf4MwAciu76NkmL2e}jk^Qf8jlg#ZDwxb8FCYJRH=fU{Z zd5oy!nhcrN^oh1V)wWo%)CKZ_!Lq?HtsLr0U+cTvrS zUoIb~vV>$m)f9lNgo9-Z0hXV^<8fXo0S0^+Z|l0^(F9$#Y@#rjY4pHWY1UP-C+AEE z%I}Zypth2>j!v(9=<4^c`~G~1By2kIQSiQDmP_K%h})EL;#f;X=Iqer93+-jOVyIm zrv4>m5}3<`GhazT^I)(GuO@=%lIaP6c?_ zaa*U$x#xQe@SWSS@3)wiyQj?PQk|L{!Q5asmE-0yjt#U{!{Fp!TB3pB(c6}k?NB%Q zXuSw`FW3E5y%U*%++uDHg!GwW_eyjxcHOW{!~5Pqv!)p@E~uweH)NX=|IGA^>RgpQ z5X3+At3KbzToT#%w7u%0^dCJ8HwMf)-b{%M4JE$$E0>g{U2+O@vfdVGmaBJJqxPoO z8`?~^;X>km_B{(4nC{D6Y7f-u#1-#L`=q@Oo+lj|<7ahxGt7@T$)910^Oe$n-C(xF zT5H=W>b2vHJG{`;;jSw{51y2#LX?gbBM#U@d=!*JD5;ohG%W2ea_JTe+X1Z43y>CK z{2E)^v*MR~*ptIgWAvx~78}O-%{|2@mv;pezd$&DWGM8e+yMdj7@9SgDE_e^c)4kp z*U%T74EJ69=Car|_lDWe`v`Tfe%c_lCgk{d*+ht)X}Or=XKTB-7&WynZU#?;G0TPd zD79>f(5kuO>p1dL)61H(;gZET(u?q6mz+c&xT%FgJ`sf0?UDMNa@TK{-LsXwy`mTD zYjSVJ%hc&04#zqyX{g7TdO^BmKz4!U-f~ z)2e=09?12V^>=#f9Oc&8asRP(fd&E1IP=Mx=7p0%dULq@|HHy5CP4*)4Ne{ z0!9x|?LObfUU>c9Jl?ed0)0J8W1LZI`a7Nxz%4+H> z^d-^Zz^o)dq$H27T6qYSv^ZMNO=v30Q0ek-cF7I^$$jwv2(Hk>?bIky>6Z{5+3D^U zcWUPju_g6}D&hQjm|OwAkJZ0*0ZS>IO=$#DjTByDG_3uhNDOANZBUb&+EOaUXX$mC zm`v@-LzWyHo-{~Ak0aZ!o+KaQA|8+4M{)+vwq^e7%V3<;p0jdgWopTirT34^+cs?3 zh9Yi28i{s}&mwmi?X1kgNdc{53*I8h-bR{pDxI@Qa~d9_AvSkv_FMMPRkfQ!lrT0V zcxA=>jWs*DpCUss`y%`GjoSPu1?}IBnVI_bJj)k5gXRTSMiCVswm(UrhV9X7|3aFJ z*6VLHFY>v`t1=RUEq70aO{&->=&epUwp}>^k@CR7JE28|enwz99qZZ|%6NLeIy;9q zz)18+4bgbpfs8G7aI zk;)0VwsDQ(j-kI4=tmf_k!_F`r(qpgEIPc5dAk2FFHx!Mx!Z1$)<#*49CIait%dFb zLZoZjbucwG6~64w%3AIkD_wmJHVNKe>}?bsU1)}fNO3(7YmYtS4I7&JM z6*}SY*j{lB73x^-CF}whXUZnOkp?gTT}PXsHnb)|N?f!0dm@S%RjL8m{N5sq*F+-J z;lJIGL#He)Kh!<#dTLpZ9MRU1xtDtA%T^fJ!iMyOv=*sIiJS_R8kPxl2~AksQ;KDEd2)4dYC8q>ehZ)~yV{iG z4(X-XPWb4y6b+a|w*x9EFasqX^i2wQfYe^nDBjUiM-zp>O$0B886*WQIe%kf0t=0V zD5MGDZM^gzUNnJPRo^SQ!U-^zFO5QQLzJ!B3xfhL)B!28JeD->>L1nb(s*u8GM|60H!3kcTgz1a=XNpu}D8(^sjxe`vpvz!Er*G zn!6<`>`^CvrJ9z~Eq_!16FVx>v!pc`oHCkVReh#(A>Q@Uc-9-&2=-uXEd zY&M9mC^Yln0%+=%LbQpL#~jYF;sAbTRvWSud&bg|=H#X#4qwm;OZoyIoUI~e%V(VC z3RU7#%WGpGVljY~_HENXOv*UDYdqOMlHcA%p^Le6=QI%_#x3Rc=<|x@Xi{@k7PHhpdXjw#s7I4 zy@cTlEC9YD3g{aXCR-wtY_7AzXx9b)k_BmgQ{XDSg;)qVD3VP{wQzH>b9biI@j(l+Rq%Wj=nNU>CjN%YSxGkmqr*N3CFsajoA)ool zPI0Wo#3~r1XIikZ$dYB$hWse!#C2dkJi4Y;TQ4e>`P5d^f-|?M$DfY5Y`QEZ$XpR59oCTqn8B zG6OiB^2oOnp~8NhxV(dwge~;eAu=^7532H$<*%Dc3ki%m6#Oo0ib^^h|1|q@{i**p zx;N4%$cfpNe_qJjemt=(k7vIrrxhTZ6qDz2fTD4u@F0a|-7d{vlTh9G{(El9-ji)G zb#Rn)hjo3^>v{KZ4Y*PG=JR**ri>OIi2=)@O^V+_WQ8cXM0kxJGbQL&&j>a?&HhZJNM-0 z@(JiGx(g#Ht2?;?a%+|)zUe<5)h6iI9~%aB^_S(XRkf}w0}5xl;{!65#T~L78wMcO zk1#QX5=BR7)E6*0fm~vu*3slR=c;la@xFa|`@-O0GwcjHjt=vzm87Q+7LN=&2<9+3 zX=fyVXqhQrpD~P;WVu=lcTd$*;1cZHdd5ZP2P=z z6$$8187UYaV6zaC?+iH)4?~t;r2}43u?1oh0NtwPks%nk@0N!L7TX^7??iN3`bj-n z?+B0XpZKm<6Cc)E7>%G*h`yvfG!FDC%_i*HPVcaoY4@bn3f$^t%zs9eGeZr1fo#+C z^Fs$G>t`@2`1ywTKsE)V|Jr|cvW}l`-z}4phnB9my0h*2aOBQEdh;GP#{&vkB*rTu zi!z_6Q}WMSlH9*^OU8=seA&Nl@BJ6&fNUzWUt7qjBx`gNe*G%jc^eyLH^X>0{e5&rPJV}RibUS}}3VT;5@8qUVpthvD$owUb(M|Z$H3OeN*HP_ETLCQg&Nr5E3!QR`n_4%8UywA z@TqP#j@3Fr7Tee$7wYq@)w*Iy6YZR33B+29##64UbfqpDtAkupZRAa7BR1!42Z`Qr zB3bqnO#Xen=?b-_fQBg1LoGQ_>&9G+K(^<EcqjR(T@dS(oyr!}6X+6@s;=pgOc_EOf6@%AUjSVY9 zvnHdJu1z!@+WNe%AhG*zAXwPa$;u=Z=Vtfd8|;V5TDv>v(yOsMirSNx`112@kp-%c zSDVr4!WeVO{$>YIJqbgp*{U<-gTRB_jGjjgnl*`OlE~nOCl-0 z{t<$cXcX)%VuvKiK#u}k0)2p>Qy-qSAcZ_ z?AHy%OneI1Uyj8XH2X%c5`k|2yN!BRiDo{+&SJD~lI7#9=TOAnIzAP%gH+L3y$plO zIDBY{o}xuyFIQO+Qk#NY9U3j&xtk8Bqqj5#x`v*_9Har>bSnE5&PYp3a3}i2QX0`+ zW6&RK2V!-x%K_BffZ9YyWd!|z0INw1K>eu)rd|IyI-yW_>E)yyfKnKYWni%QMSy}# zk%5XeYyRs-Q0J)Pr5IUUlnU7i|F|pxVq+*}hdLzhCWWbNlX_C_i!$m|y z@?%BZn@B-d2^DWG#`{67yIB;NDwkVeJezjpXi;YPZy;f@;6ez}ds|evsZ;?l@Ro(o zILbx@Q>K{ELDkAf1o@%^IFVf!HXQVOQ4%%AO{4SnQ&`r+pXJ1x3v0&0-hvLfimP7S zD?fx*C`{f#h1^YVJiz=!d>^KX+vFtI#t`%Pcsp%imAla5`rEZ8WK*QaA3}{0F}tLn zr)MY+ej8PoTHF^VUP|aIqk#ZO1jo0Nnj1|WpqdgRUIh}Ev*^!1)d(#52rZ3nC8D^` zygTGJ#PGf!>CcPd77Ayifrp~FMTT4!#qg-Y-l_wt+nOp#j_lTGZI~X%m(+>kS=Pea z*-Z*-;KAKY|ICcSq_L|qV^XUoIpb{11?5;MxZ`j4EMrmvxgQuPaVi4T!7qm?1W6vO zU0jpQy8hB`vRRDJeBu<}M>g4}Ws~tdTw=*QpO&E+4d*tP#^JN4^Lbn1KA)b!XR*oa z14=cEV*6;;vltw6P1F3v_r8_8) z=6>cPhhuu+j;?EeJ-z`nNJAgjGS^!gs7?HkZc-v76e`1#q)b*>}A@F zfwxOBZue&na|rx9MD$I`{7SvZShAB4DasC6A`zy=qq$~0(4YaCH5e}0WoXcyQ`^$J zfu^PG5yJ7Zwiu<6&@&6FOp{ziPzwu!A>EDi8E3^UuH(8V1ZZZN)5v3(qAMz)WSgQ& zg;^V=%joCJDR+I_60YjIAzIbFzgnrpkp5Bz99c3<+g2HeHkoq&sA$bHd}$4NKc8mF zX(^mRGzcox>Kp)cKIs?*)ir|@4ZYy`-}u+Bw1SJhv=dO^>9`V}%F1&v#}BPaLbCIT z)RoxuIP){*_WO4hPi9jGzk8$k>LZowh#$nULm@AddPgO}c4i5m>$LsbJLtl^ERR9HfW2ASi zgcjXLxRl5KvB(cJael}s}Zn^4|v%= zog8DQ$KXdhRCV3_w}5_!Pcz2>XB60wD%5V3YOzh4utinQFjtmk89@?4zmG*h$Ka7h zLzRJjmI7Zk%R_6DCxQ$E)ZL6HQye0mlcxrph!yI>#iN^vbyG zK^)fAUZnb(#SZt_ZLK50E>fqn`Za6eII=fu6f(0@WwWks=7fVr zaq}~r&574AJ4XZ9wcRk>9;N@p(2Llw49!c76%pj^POmZ2X_Iz_Qe0LHh`Yuf$JqPb zy0A@N!6{8w4~4qQ&tuALUukr90p*8L=PD1u#U7x$Ms^*XK8fJPC0QA81`drm5t4AL zk@xTvc1bnFktU9)XX>B`yo5vuX%EhMEY*w?C6Yg)3I_qEfN}yCZGYMWb0B$k1v+qg zvuAlv5h^n|{4Dj+cB@z!lx|4KxWDyyfkzNgOGTD1UmiXuL~&w;@F%=@{BwJy2=9%@ zw7HfaBh8nytXF~@B*J{+t`Q9RLoey`Lb zOY2LIbbbIAx3S(^X1W?L!NhU9x$$1a_!B?8zE$F zXBbQsuTrILp&2X#-=UYH6dl>$c>pREhpW z;9B6_{^xpcpJOiM4GOkg@EBwP0&<$S{}S&}WNu`^qN@Yg#BZ)kr{+w2)31d<7odAH z?koe!!OF39{6T)BA$XW966BR&YzfvCoa46RPk0~{23OR=;K?jJK9!9mZGLW=?5+)< zmd3bx@DKLgB}2QK@=mTMjKLQk2WRemD%{$2N*Qu&@t|4s{*&jGG8BG1S?!ET#BcP^ zytaZ7PmduHi6y|}Zr=** zuhciaVNm78Ssuc(bNOquNVMM3<8W8X9XD-whQa$OFti&MZ4_;;`9)@7sUhTXRO0YE zEJ?q1&&zBM!QY_$UbaE`SnH(K?x0p8H$ixLG!!DNFneypvK7HfP1hJ z%r<5`3>ZrV7s52UlP;T1YA0s@QV-GZ6D`pLKAUb3C#a*N(oZ#~^~DzcF+i+VFC$iUs7_x;|;^KB}jdvm2Z>VVJUC4Hb^HR&idG`C|OpvAI+-(1ol)_hbDJ*LI-%E~1uHzHCYb8BTQGmpKg z+#D*EfUIw*ZVEK1WFRw)ebNhu zY?J*XO1$!VT_HQ_;7z@FeYcbJOJFM_AIHWPJuynC4!XUf$I2Q%U5DV~nU(9h9<`h3v+d7+;hk`T*%Kqp0 zTb@sohgyX7wYO6ND_JXrl5UVg_`fo(tQcwM{QpU0J{rxxI`pW2`ktvw9pDXLJytX7Ycu9l}g3Bo&=wy&Y(i9ag&CWWVhugt^Z1ud+K^xzNsP!G3%2>7)T|47t_U4%p5r!@GqA>ML+M1O7bnYe;>2X&*J;nPT{`O)5}jT}d5S1ar!EAMdT>H83p$^`XwxS8}p(U8h5x zcMyVIm3Kph%Rr+ixk}WJRnH9kf)%&@_&Y@e6aB-edC=sK2uK`pe=&(ukNI-unDA%Q zA7Z8T(be4a(+yNqJs;!3|5sbf$26bu z@5M@Cl08s=-ni}ShrTxoPd{Ygz|2`GTf~lNv zeS44lqojjFM93Zfz>cNaTAbYT?Cn?2(XX*xyqz3DB#_Q^m9WHz$v&087*V@S`d(iH z7kq9PBE8OL>C7Q>KI(_ksXExDvZ`C36cMR2N***cJGHuZ{Hk|tIhfzv%l5A>M**=U z0PE8!$84X!ZB z4T4fxg^(>8M!KFd7Yxzo{abiv1AY{a}mjD6!!0 zp!!7D?L254Xkn;5^P;Rh;>+t6?^SOHV2rekx?MKMTAo!H4VX3|(F)vI-KEyi)yg?t zIqF4TIq6N{U3)40ku%0HF5=BwasC)iB#t`bHr|g=Fv+P)a2VY8jX9qbwvG57upa+s z{Rhxalal6X(kwcX&nH68s|s;QE9)Zde`{7E=@h*G2ekhO0RM3QqgA5%8Yn7|FSEDT z`)twJ_fzQbk)yZ|n&3rE!D>d#!#WI7AHAn+F0!+{MwsTjvDH#cPyJU-(d2odo2!zZ zY2KV$o@(c~@_MYM=YiY27j{i~6jRxJWG9kN5xN{vt9VE0KTEq52?zWgY})6@nKms! zuS3yKgJ4G*E%!RRqm^wW6^ z48Uzg+RT6d4`{cd;Vd_nY?Z6ugRCIbQfgoqjwe+5-g{H$btg`A2~8Vvq1te#1Xb6b za!0Ydis{j9%%QurSc+&YEf?D9DrrQuIswG=lyW_Sya&u1vQHzyJACIoFzimFh%Dwq zdy{eOPGLLFE<)B2mhVH>C}(|mNK=zudc9fveE71er!0Bzdrx^#iZGqyiPwKV`6r|( z33vNHpke>wSoQ})VTip4!*UninP}$9Xv#7h)>vMV28heP6732$ks=Lv?E#q4 z(x!XjS7cwc_q5SgQC6L&k#wL^M=MO0GXzzRQ2$g2FP7mhOj4gHPYr&Ol8$E9vDMl+ z**RWC+C*J6)*`H$nSJ#TPE0`PMiF(8{u>y|1!9Rkdos<0IzOK98Dc~%q8K56;x*Aw zL=2Un;Jm~z6CojR8#s+TlwJ;aC-nnW-%We<<{Hk#xO{*nyPa?Qqh`HWaHyP7D19@ zgEJgae%@$t|6``&D%Ac;+gVpggp7Yb^kM(-dDmf;H8?sy+1&$BW%$^pUMVa@mGyXu z6&>uJX?w9znBK_MX3BvO4$dOFZ}IS8%22yX$Q^BR&`7Ad(vp3Ssu5V<>AV((U)L8E zN_}%=a;DjgtkqLWq`Axpv|{zmNUsfLfc>NXbDa{fd8hh~@h>X$B*+V`g7cyIwpb_v zj4U6EmJG^26WSl(3^Bn6AC{YXtCElYDaZ_5LU=l^5kHUv(Q^<>w=0Mn;=)9zE{KLs zl!iO(?LY{h8I)fZ-x8Q>ZVVn9XuXUQY|VyRnW89JP^+GgnJW87tZ5N$juVrH4Zqlq*rj3w_MphDqXIB-~ z15ZXrD)CU{zP{E<-9byB4-7<8lCWJXBqtZIo6mBmtH`u;KKuy9z4p1G#ZKzp@_2dK zWa+Y)$-e*Tz4`g-OHji2mtvE==}G?_wSk1u7m%6NBr*I|Z z*e-;Qn=jAKQLDs^5l2VC!Ffxqva!{z2gr!J)?cVIlaIQ6JP0w>mY!t)qBm&WSogW} zoi(fC)iXvlQPUwQR1?Zv>z}(~q6dsV9tlMHa$2LYrDn>?6_hH^B*Pv_%axfcHkEUf zb5yJ^=}zR_=VK`AFA<;XO=M|H(-o;J(ohjtqqk&gi@YiOs{*xTz!b8oLZM6B0g498 z@l;9C#l%$ZR1>*?vwgQHIZz)lRKbA#!Y(7$xtoZa5T_|JG<4Rq&~$hcE{;6VPtKSg zX}Krw)no)Fw?%BjvhgP1YjDk(Y=`PIjehLRn!QbX`nmtSXUFtSOSFEsO4v5&)$~m8 zkA8QGKoyz7O({73DeWVO*dvwB0-%_F5yT=YtFUso5~0^>mVCzl;t zHXS?PS%bqdP4LW;Tp+2d4#`N+a`wQ%dMxqfmBhXxfIga>CqTr!4K z6s(ccH+0Ci5<)S2>liqi3(#_A^DBV@hX#!k^LB7s?(JDlo%+xk>9N6gW;{Eo6GXbb zMYe9{rVii5?PcPNet_d{qY(M++gWpF>Oug?4yEG*!~VswD0Zfy@D0KqjSp zvA{|6Fe^0Wi(9v~?U{X$gE6-ngn;dl1jB_3sDVet$Kx0B{v|XGdgf;w@QE z&)daC^os=ZE5*sAGPw2JG5S11K9?56KDs%x!*cFOal=6abcqmCgK%vFFl>ttvgyZL zCksGBN$TssKCBF9eQCsokj#$clM{+puO^Oqir7itoiqg*_lLc}NCel_OP1B4rp}Ae z^nA%bvZSt!CI?e^0V+W~$41!yn4$WzVCZk(h3aWhZ1>1PI545@8)y5cahXY3%oHSb z1O+7}1qF2^6%=%I5aYV&z856$haV0GJZ=ZLjty;<{4u?ajz+JDfORri%XD5euwr~M zQt9qu6dI?-m#tmaq8pvc!^zs2b-&#hGx8hd`cdJeX=NVZH-~-f(i0k^hvXMT9F=`<#R@Iz=8iQkj|LIE zrx28wDfme$KV*S~f$z4-77$59y{*{K(TWW$OHUAY0#_GuE>pmF=SO1YNxACoYx=%o??u`t-M` z81tA>-1Gntb#F3c-0F3qmuQ{7fWO6kCL5wI73{|vKwxtn6z+CU{8sbwT9$V$@S|!1 z?m@|>J}-gO@2J~LbP3SU31$w*4JK5ii7OGBIC`ni-|v^q8G6kP)16aYB4zwlz6jjT2J zSMt9Oe0^O&+B<_R2=4H9k&0}3B>tvFtgrqizi$PKG7 zZvtThKwKL5>N6BZUjIENF_6YoG9@wiJmIQ{XMJerx;AkCM zl7Ph@;f=J0jFR|{3)~Fq!pnaJ261^b2W#qn%?YTeZ_>F9!`&>r;YX;AtzQ zUxWa)q_}baVM8?tedf^EUV(2kgM6vnCHbIAgg6#LVTwWiPjU0uic;b>Y zvrf+A5foDf!rbxAHrXB*5*$hH#3Z#$WUUKv0ivBLYvchYdTER)nYmVuQkEL+UW#X; z?uH}_8Gx(w-;pF_-1ywI+epXrEh&*nx zh$T795ja7t(a+J$M}OIvEdlvgpBH#OH(=oN@9?*Cs2L*AAL*QCu+$^XYXlCrbD!p!>(5yA+$fH25GwmF`8z)ladX2NkX`PQE#Ya zrTw&&JMe{^sZA?2==X9>sH9GiZqu~}!vQ1D*ct2d&;f4H`Tqc%$@824RJW_d_Jsf)=C@3aM<8|5x%!#6y_9c|?c=V4-@2K&! zrlQ+Ven6l8*QGk&jC{U)8(JAk^6f9_NBb3BgIZ}GIGgu(!8c0Sj>p3o9F3E(fG6Uu zcnhfTSV(|m;9)!j^4NKg(sTFpz=gcWCeOGv+zO8apKKPkUCH|R9(&Dy_>W^B^6zzD zZ@9!=`zT^@m(cNU{ySbQIVcfD%wURzA{b7X7s8dmx<19T@EdO9@vGSG!33BMZ{L3k zwmcqQrksf|g&E8spza?>r>hS0N2-t3^m0d!ri9cAHKAzr26m9pr5y1Z;qx>YM)LLu z5v!zIbS)~4O4p(6Oy1{zopC6hOHl7u?CmpDUlIl?u|PQe*fZ42^*%$|Ve`z`8->i&V`f}S6cwnGpx1;yY=g8-H{VTgd8M=U_^ z;3-68LN9O%n2hGmH&G|{6~VC`?un^lSl0_JjOE0WY>6JWq+SSrVz=+`ICtOeNwIt{ zp%&_82y27^!=F598O?a4hx8oF=;mPZPcM`)cJy4kWTzUP+r+0=AVW0^AvEKdUTrW` zR`T6lX9{b%>*tqY2ckY!;l%h~QwUd)hY-SXg%& zrhyrl^LzFf37rY$m?JWBP-y7gck`rN**9PM-7F z9KHsl6708unfePJ=*4c=6S=GmZF(-!Riy~vfQKK(o_HFI*ZJXT*bJNVsi}H>E}LGe zEv`Cxq-q~e)U9hBct2yGOWhYT_wK(V`0(FilyEx&n-s8rVJ1q;Bn#!)HO*^1x2*Gy zPh1(b92#7E(Ybq3Rp%b zqRf3od1y}!&uAN~RW-dyVIAF0Fvida0MAb+P2 zpRX^`m0GG5A-G5`DoQ`8sL0Q=EG+vBRdFy!krbDI7@rhpIR#tH#~hx>dY(sxd&Pne z_!IO=@+3Yaa{t=fuA6k^2Nv zXLb_~xGTzzijYn6Xcoz%I%%OY9&+UWdQB0JbI7SdUQ+rQjN|SM+{Ml$jL#8yc8R-0 zTJ1R!$j;K;PiPohIvXoWO-@WIg8va=Rf5BR&!A9?FjQn7OAcVCJgfGjy~1K5R-$Wl zJT#Mn9y`G*HLahpe&R%}Wiaut&vBDNbwBRoPmW)>^;`?Lg7e%?nA~|2Mt5{Ndvh{!DxneCy*^t_J!ZW<6HQTYHnS)$KIKHg%mQw=Da&Qpe3^7cth)vi)+eu8~6i% zZO4jRxG#+FZri2DuZ+zRHtTfyT)kz6UY9{Mo?UuUDcZ*~9n>)e$j~G|H5xxQYHj|( z4Z@+4hK6pIIY%82Yw-^ZiS(QHcsVF}Mn!?wttVfbJctykZ~+`UM55ODLg0-sP6a$u zMFe{^Q6$=g%y)NYTjr#@qZAHLP>7v>XL^nwvSfY*#h2H_?^&S7{*7$f+Ii`;6(>{c zTf1scoeA8Vn$Ooaca*hr_fB}ZFVuK>E)eiW94OzFhI9#R63t{w^AwDksq8xoPy1xQ zWmj`ka(z+nQ8a${_kdkRzv<`e@s76P91q|9>CR4C^;z7HjW^NV6Z{;S=)J0ccxk*S zYm`o#OLvI=7HM5{xIUV5Gby?1W^EQiy4(zbRMbYyK%@MesK9nwtRT0q(J+6&N*-Na zfhxGFQe#c20B?aLuPn?jy2jazvS_S`@8?8W{V0oA^HCOtLDiy<4AE`kG)s{&n5376 zlPJAe!0rR9u&1<1{)mV;jWY^=;DX*p#t=b$kM1_P`!D~`P00S2OrBDhpV7cI-0q=IyURdx=iDIe2J?21v#8VBRX&<_OymY#Rgf7|;F_ z&ryiop^xFFm{2I`)u)rvUau9#;1aBWL-O+k1cg=}7?@{B5!9FHnsrEji`di&<7mJc|_Zjf(?3-emQ?bI2J zSwcyEU1b}4D5gF%B0MpF$WPpS`Od{F7n*nolQzmLi;O1gk$`NMvf8h<%Z?Kzl4w#Q zjc`->SPxk^2j}Q<=A?-RoB}zJod?NJh!QDRu0n3vYk8NQ-aB`2!K-R-?GiH2pnvwU zH_`jWXZVNZH%jZdBi)Jq)k4))#Hj{yA z6PH-c1veCs5U@AAxOm>bhBaRvIRF4JZU<|VfoT)Bw#@~M9)IAVW?mT0n@+S#f`O3_-@vNT(Vz!jFn4O=M?S3)z1bXhZ* zc~_f~F>zZXK6kUm6S4-sC2McitYxSX&Bod~ni6ZF0iF_Tqj&O!EiD&@txF=~`Pea) zR~;Wq$jc{1sDJHxwsJkGz*=?yAbRM5TB5UVe5RBs-(`E^*wgS5RU_O=^I8$_e(uudLI^q1@r_kZZu7YC2{+Q(gL{LVx?dutzd z>9{pI_ z=JaiN^@&(g=n_%7GFulAMYcpVw)V0_gn_z1uzh%$I+O&my7#=?AY$jxb!R`%Iz&o| zrgbe`fvBVJ#b2N3R5z1s5EUR@l1BlxB`s0zcYmHVn?&qLO-UXZOdU&>l@4l_gheG& z-Yo8U^lvAqurERNiNKyIv30Db1@Bh)huTB^N-B}``yIq&F9-@@DfEA z{(m!Xrg!OG1qqL5(>vTRhBB9FgR|*$me`w?eUF_|2eMPLQ>=C7$3y3bo%8t)w!pyl z8R*s=+Z^J-9J(^B#xF499^=)f?5QNpTa2pR-jwFr^<34No86de+PuqIO0}&<*%NL< zje;w*vIlc81rG6G4h47Wq85IGgDz;H?SH`>%z^H79hWTqsr4W5XZr@D-LY2f=5#6A z4qZY?m$0czfrF8Ua>(gp2M2R7heCjvV#czZHG?TkfWG!;08QC!8QL}Y=`Hm6q{GPjv__CaP7NmAMLP`XbW3n0RPSaag&Fdw#9)pTF%6SaCALzUVE|t# zU3(vLtJKh0hc+paq|O@h^nct$LDh_ghJI+NgGhH4Lhz_GMr!3;y2YF>alP4c2!x;f~a4z?RG%mU!k8dpx5d-8BmiI{?j-no3_>scQ=;KmWo1Nu- z7-^V^M{W7G;>jcKM@C>^I-Nz1MYribAp3XiOov*yfVxvv-DsmdE&}}%Z_F18#WG2uKfTY;vpo*frZS-l5MP5f`<*@lA25d=2Md=#5^SC zJBXj!772z&wsNEi6T|Qr88`~aGx-EP{k*kl6g>{Ban#DwJ|WuYxpZ>o$fFA)GR?P) zU5o_8a{&5EK~Ds2`o08@@HES*?uy@Dj5 zMQ&7TOuDOVo9a~d8k@Kx*gVv_+#~`Gn3~+_awnaeC#8uYkD(~t-mu~f`=rLNdXimI zYuk1$2knqrm4Bz{m?gG4_gkAqZxcCB^=(XvTt|Y1AB-f2r^Fo77Y-y5pm4PHHYIeP z`y+j)TP%)-NnJB+HTA_*;)AEyj^8Bk)()c$=~E&O>Dt(CwrNH@oyru6X3@u?vF+AercQ1~v_n2o5-&?BGg>d1n zh+n-V8Ni+4!cpMCaN#AtW1xb}m3Tx@6g+hZ36s+xrW>K!fi?-t`8jmjr=m&kk1|ZW z%wfYN4S!D!xITELM`LP3ZfPaSZ#6`YnO4ReGbU#>EXEj1Sq<%Iwb6Zo?fBsJ{m*x4 zzjtxYdlJOp(Rk4$>c!Kuix2Pq`uxSK+0A$=)#;cPb&HsX`J&nH;s_J8)ex#7NCJbajzK5rHu=aXA(iqq#H z+K#A#zy|jAX*0R{ln0&t zdi8#CiA}eYA@O>|&H5=eRMn{p>0>ma5cz6wfsOCvTDOB1`z2^`wgeVDzq(+-Qx9O_SARVGh)1ddSX>b&b`MU9_Ul5$^r7O8fJ%w~ zv1swzXwk=`<>JRDXYb#xq!Hf|Eq)U%@r%$>OfA;LE?TqzS|UG+<+oyA87)QobA$!B~JEdgdMfyT9zu1^B?jC;{&nr22K3-nJ87l?O&wnqz zfB*ikk6*n0Ns3KOqz?!8VqooJ7xWSlVuc6MbLFWtwOHcAEFJ4b2MyRXpYr=hJl46D zu%oV{Z2KA;O2AVO2Dd=??2-#L^X77JbANp|xxM~8xV^q#+)SqTi@_CN3W50l_v7oq z`1W=_`z>EzH;cv1{l`y(@q9k}_kWxk3_!h|&G}jg1AZRQPu4K%S?Rb4e)G>aPhUL; zLSD?Kvu~R2gWL$ElxDad*bA-xI&F z7{4X!bx;vhKYYy*;agyTZ1l<&>& z+eX4FIoGrNMy|J`?F6Bnj5^qLcK9QCzhufTZd@NX;RxIm*dL3V_6>axH}QDfoIQK` z_`|a$iofL^A#BzkQGQX1iKP?+-}Z1LnPT|PF@4DHzGjLkS+4_?*e^ZeF{MX={n1d_ zPBBg#50!W4@BaSgRR_4=sDDt6o25tmqS8Z3Q!G@q>LMkuP>tz-mP&=H{_2J5ZWDIV z(yu<^G1bRT=0{^?yY_JMc&t1=d-D3j^QA2D8>W@GS$f1TDm}10DLsyy>Fr!J!NTw4 zZ7Nj5*DO6s=IbJ*S9!EYeuqM(!2ejVXt@I|>Zsbo>0a&O%tq~@Tz|Lt&}$W*JbU?n zum6E<^^1$0#fSO>#fJk7dYC9yAYSdNKr~-oSRTNdbbivSK%DJYAoEYNwh9?M8(eh? zhMs1J8`AX+={F3C&D6?}POk0v^4*&^&!1tnS3Aw9zqT30`uV+#kWN&Ir2$3UHel!* z@ITFbHuz=oYcoi7Sr5m8U7aAi5!TB;{TEmB0G9#f0~43a_yrt7Wi>H4VKO#3HZVCc zGBYw}I5IIbHexk6GGt;hGiEn8K0IYLF*so|HaRvhIWaObGG;h3F*G(}H8?V4Vlp#k zH#e6l`2|RSG%`Lu3UhRFWnpa!c-l4CcUToi7Y6V>JF_vCNMeamW2_MYMG+AYu^=Fb zfTBVKqu5c@*n5vr?4ZV8z=pjSRK(tss3{svj7h#Uqv^$Tzx%#B_n+T$p4qv(cc-3N z%d)H|Pi#x=u$IxKbS&LS^XV44iEg7?ZObZaS>lv`&ao_(Y-fou-c5IkW%=N$r`%@A zIJ%GS5v$x0b5*fqJUvMF(@W&;e1`UgiGeVw4gCxzPc?5^N~})FCN4)y(=uYElt#^zJ+P~W-G;$bzC~Ik z%B!!0)B7-eD$MYKJ&wSh9bm70Ff*3E3;XcD_5Bt0dj}5irF^5T-Ehz~I5?4h0<$O4 z-{8<3IE?Kx{6$&>N5;cZFT>HL=o~nv99;;1$Fe2I^Bu?UgcBm+Bt_HVWcKe=)|plx zPLHBD;0*Tc%*ynkX~VK&%^GLoXX!JvyjZilqUP*7aL%hRr<1AaBi7szCay%EqZP!O zw*xijC%^?O;KG@3Q8mi;U+hoM!(6_5?zb?HeX(RHT*~%e#x`8e@w4J-x*V^d(wTtnWFhf8dP!@hA9mUHD6X`V0JxtH4a?txtVLYDAdw2A=q|i5pRW8X)r01Kh0^Rw1dq1=hg|Aa&7$QtvE$ zIhkI7_0h}X%eU}-1RGp}ew(3x|9tv648W|EhCjnb(_vs2^B)?DH1_sI5Dlb_MS_$m zZ{mr+HgPa*Mnh{xZ0FQuw=T(MelfS$&wi&cf9wHWXUX%N8VY}iY0G|{O4WMTC-$-w%Fw^o02vx znIm?!?@dWtmgI=-*WZ*xv1C5uE;N?5qtRkF-{ffrJ2cwVj}bd;t%*C(_B2lH@T;iV zA{(~E9%M(}H8tbKZuO^$lV}3%D0Z7L)NG4+VMh&s?XU{j(b$#jnEzmV?5cKrdCIFD z_>=_9LAxVHi=BwsWG7*N<+qc&QH&G26V^|=GscgdvKV&7s&02{N#BR5yiXd|Tf6%} z%HO7A_SqQ+VUN|YXA{`#B+QJYm`!%?OxTD2wjcH{yMHLfsI~`SC9wyFnHESCJ1fP+ zDKwdO5__;?$~%jleapmM#U3gqH!BfU!PDcSJWD9dGj+PrRGKD#s#310Uo5IlwyE<9 zOBOKhPkYerG+mT0);;BmJx=*yRaE{7=C>K50=#1;lfFuOifV`)Nd@A4Rb%XKstI;R z6*L$&#oniavA?KhSU*(=_6`+_c2;4*Fnkr*YUrm4sp6uYx()s7y6Z5)cd zNkvtltkVv?rlQe*J1PeKrD7}7e_$NeHx-ZFQ+2@Vp%QqX1mut^@imx)zE;WGV5jC3 zBSdvBH1FO^RLU6>_oKaOA5mS$qGmU&x+*n*ZZ|diib{KA;sK(%vo+HBMj5s#A4s$4 zAWu2pVnn{VI+%U8knuP=jAqjzqQ+vsR^!r4{h^}9?=|s%NctKbE^5L()SSe=n4Alz z6vL^QL27ysobd$C#2i$!eCQ!K8*8zNkR__+VHC4ft;m2Y|AebC)L1%- zjuy4SJBG)9h{}J}#N$O3AYWA*`SeX3`J3y&EwxP@_iQqeP7<}X7Ooc6GFK-@z`d?q*S(r_@JS`PIEl%HF>}nev{W^68%;t5lzJRDZ#>>kDKu z>dRU1EAP0SBkF7Kj^?gZbLl)$57>qeIO@LT8uGA;sX1TNBR>-_67^kgT>YN^=Z9YK z$DZ(iCuF4RXSUui99O?`w*H2!SpANjLOtfZ_ybv)`V;$z`V0Gu`kQa@k9XZ(Omk_T zsDC?~^7Zshx`Zwj-2l0nZh-#M{(+|EGSLCv(YT5(rz=D^#OX-~;#8#@WAD>VFn4qi z=8JBM)1_{PJxhn+q^6snf}#Cj7-p^x$6ls?Jt>FBh$Pq&b5=)YP~^0_bpXXVTDRd7 zqv}%Z#X7n*#pBCLw~SVf8xpFaKpJb~3r7i0g@Ct0V2y>N>Ca(XcR5}v`C zQJ-UvoX1?#7m#b|i`WPCCCpHLng8&A3UY6K6??0`R+)Zd-hGGYx4rw$Ub>6!7X1#s zuITF=X*Y(OI(tOloN40yqTj_F)bC+O(C=f%)F1HQ-ooy$Z}UHV$nkjx`>ei;?+p5* zUGQU$gnL+L^e1eq`?2QE2k1e1$WxBB#Zjls)x#VEwW#~;$j!0cyxPs8-OM?ElJ(s$ zPHvv;=EiP5?B>93-s?~|*FDNOn#Ryr+MdSIcngUN{7*Z*XVH8Kh1aq9Z5$~_iLdKeTKTk+<^K~e;Pm=(nho~4Wgbo^3@oV?YrN9Y~3%kqv;ar z<|Fm?N$1hIG>_(rqeK(o7P@I0s#H+>0 znQP)*p7^ne*NC&|iHY|-inF-8iPws=qpXP!c;ewEUMEgbwuujW;+rO3FV0)xCNA>C rD@?q>)8;-V-a*}YdQzOnp0)1(xzl_Im!AX%9||xyH3}sqMNdWwT0M9# diff --git a/paper/pnas/make_figs.py b/paper/pnas/make_figs.py index caa8cec..f11966a 100644 --- a/paper/pnas/make_figs.py +++ b/paper/pnas/make_figs.py @@ -47,7 +47,7 @@ def fig1a(): from matplotlib.patches import FancyBboxPatch TIERS = [ - ("Population genetics (exact)", "Wright\u2013Fisher simulator (NumPy)", "closed forms \u00b7 bitwise-reproducible", + ("Biological model", "Wright\u2013Fisher simulator (NumPy)", "closed forms \u00b7 bitwise-reproducible", "#3d7eb8", "#eaf2fa"), ("Trained networks", "RNN \u00b7 MLP \u00b7 VAE on a synthetic oracle;\nconvolutional VAE on MNIST", "sign-level tests \u00b7 exact oracles", "#e97313", "#fdefe1"), diff --git a/tasks/lessons.md b/tasks/lessons.md index 43e5880..b3a6694 100644 --- a/tasks/lessons.md +++ b/tasks/lessons.md @@ -32,8 +32,10 @@ statement about the literature or the result. Confident papers situate; they do reader-facing prose in every project.) ## Terminology: "model" is overloaded in this project (2026-09-07) -In the PNAS manuscript and figures, "model" must mean an AI model. The theory tier is -"population genetics" / "the Wright–Fisher simulator" / "the minimal inheritance model" -(with the disambiguating adjective), never a bare "model" label — GG flagged Fig. 1's -"Exact model" header as confusing for exactly this reason. When naming tiers, panels, or -headers, reserve the bare word for trained AI systems. +In the PNAS manuscript and figures, "model" must mean an AI model. The settled term for the +pop-gen construct is **"the biological model"** (GG decision, 2026-09-07) — used everywhere +("the multi-locus biological model", "Biological-model tier"), never "minimal model", +"analytic model", or "exact model"; GG also vetoed "(exact)" as a tier label (the guarantee +line "closed forms · bitwise-reproducible" carries that content). Support level formerly +"Exact" is now "Closed form". Keep "exact" only in technical noun phrases (exact-match +verifier, exact oracle, exact equilibrium, exact recovery).