Expose observation metadata through simulator interfaces

[AlexanderFengler]

Depends on #339.

Scope

• resolve explicit observation metadata from registered model names without sampling
• expose a fresh descriptor through Simulator.get_observation_metadata()
• attach validated schema declarations to HSSM-compatible simulator wrappers
• resolve supported *_deadline variants through the same model-config path as the simulator, preserving exact base-model response labels
• preserve the existing unknown-model warning and fallback behavior

Compatibility boundary

This is additive, out-of-band metadata. It does not change simulator(),
Simulator.simulate(), hssm_sim_wrapper, result dictionaries, result metadata, or the
three-value validate_simulator_fun() contract.

The descriptor's obs_dim is semantic and derived from the schema. The existing wrapper
attribute remains a legacy output-width contract. Consequently, a response-only
inverse-temperature-softmax model reports semantic width one while its unchanged HSSM
wrapper still returns the historical two-column dummy-RT/response array.

Metadata lookup remains fail-closed: unknown producers and incomplete declarations do
not acquire schemas from nchoices, legacy obs_dim, result keys, sampled values, or
the HSSM fallback labels.

Review hardening

• isolate the built-in metadata census from mutable custom registrations, which may
  intentionally omit schema declarations
• preserve opaque producer-config extensions without deep-copying unrelated runtime
  state; copied response labels and schema declarations remain isolated
• cover binary DDM and LBA deadline variants so direct, class-based, and wrapper
  introspection agree

Verification

• full package suite: 1,243 passed, 134 skipped
• focused metadata, HSSM-support, registry, and simulator-class suite: 158 passed
• repository-wide Ruff check and format check
• git diff --check
• actual deterministic ddm_deadline wrapper execution
• HSSM consumer canaries against this worktree: legacy simulator and random-variable
  construction, 2 passed with candidate provenance asserted
• independent static and ecosystem reviews after the deadline and test-order fixes

Deferred

RLSSM ModelConfig/AssembledModel adoption, the broader downstream compatibility
matrix, contributor documentation, and any consumer switch to normalized results remain
separate stacked work.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

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: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: e9faab02-b4f1-4a0f-a8b9-e786e60cb814

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

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Flag	Coverage Δ
unittests	94.39% <100.00%> (+0.03%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
ssms/basic_simulators/observation_metadata.py	100.00% <100.00%> (ø)
ssms/basic_simulators/simulator_class.py	83.06% <100.00%> (+0.37%)	⬆️
ssms/hssm_support.py	96.03% <100.00%> (+0.53%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.