fix(rl): WorldModelsAgent network sizing, deterministic clone, pure predict

[ooples]

Summary

Fixes the Generated Layers N-Z shard from the PR #1563 failing-CI list — WorldModelsAgentTests.ActionSelection_ShouldBeFinite and Clone_ShouldProduceSamePolicy (in my hand-off lane from #1562; does not touch any files #1562 owns).

Three distinct root-cause bugs, found by reproduction:

1. Malformed multi-GB networks → weight-streaming crash

Each VAE/RNN sub-network was constructed from a NeuralNetworkArchitecture with no explicit layers, so NeuralNetwork.InitializeLayers fell back to CreateDefaultNeuralNetworkLayers (auto-generating a hidden layer ~2× the flattened input) and the agent's AddLayer calls then stacked the intended layers on top. For a 64×64×3 = 12,288-wide observation, that default layer alone is a single 12,288×24,576 weight (~2.4 GB at FP64) — which tripped weight-streaming's per-tensor byte[] cap (int.MaxValue) and threw NotSupportedException on the first forward.
Fix: build each network with its exact layer list passed to the architecture, so no default layers are generated.

2. Non-deterministic Clone

InitializeLayers does not wire explicitly-supplied custom layers, so their lazy weights initialized from the shared non-deterministic RNG — a clone re-initialized a different policy.
Fix: assign a deterministic per-layer RandomSeed; Clone() reproduces the networks via the constructor (they are never trained — only the controller is, via the evolution strategy) and copies the trained controller weights + rollout state directly. A serialization round-trip is not used for cloning because it rebuilds the layers and drops the seed pins.

3. Non-pure Predict

SelectAction advances the RNN hidden state for sequential rollout, making repeated Predict calls non-deterministic (surfaced Policy_ShouldBeDeterministic once the crash was gone).
Fix: Predict snapshots and restores the hidden state so inference is side-effect-free.

Verification

All 7 WorldModelsAgentTests pass locally (net10.0, Release): ActionSelection_ShouldBeFinite, Clone_ShouldProduceSamePolicy, Policy_ShouldBeDeterministic, and 4 others.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
aidotnet-playground-api	Ready	Preview, Comment	Jun 11, 2026 3:04pm

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 9254acc3-f590-482a-bafe-a2697fb06059

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/worldmodels-agent-clone-finite

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[ooples]

Consolidated into #1565 (single PR for the PR #1563 CI-failure lane, per request). Branch retained but superseded.