Define canonical response-domain configuration

[AlexanderFengler]

Purpose

• define one canonical response_domains mapping keyed by the physical non-RT columns in Config.response
• normalize legacy choices through the same resolver while preserving established DDM, multichoice, RLSSM, and aDDM behavior
• validate categorical, continuous, and circular columns independently without model-name or response-width branches

Closes the 3A slice of #1265 and targets the cumulative draft sink #1266.

Contract

• Config.response remains the ordered physical DataFrame schema; this PR adds no semantic alias layer
• categorical labels are exact integers, continuous bounds are optional and closed, and circular bounds are required and half-open
• RT remains first and retains its existing positivity, deadline, and missing-row rules
• raw canonical and legacy metadata cannot be supplied together; resolved single-categorical configs retain only a matching derived choices view
• resolved configs are construction snapshots; unsupported nested mutation fails closed on revalidation instead of creating a second mutable compatibility source
• multidimensional responses use separate scalar physical columns

Complexity ledger

• production delta: 8 files, +397/-145 (net +252)
• new architecture: one internal typed mapping, one pure resolver, and one per-column validation loop
• public surface: the response_domains field on existing configuration/model objects and the optional register_model(response_domains=...) argument; no new exported class, enum, module, or registry
• removed duplication: the 57-line choice-only RL validator and the old global choice-validation branch
• retained compatibility: legacy choices remains an input and a derived view for exactly one categorical response
• no model-name dispatch, dependency change, predictive-width work, or simulator integration

Verification

• 283 passed, 32 deselected across focused configuration, validation, missing/deadline, HSSM, RLSSM, aDDM, registration, and sampling-smoke compatibility tests
• ruff check src/hssm
• ruff format --check on all changed Python files
• mypy src/hssm (83 files)
• pyrefly check
• git diff --check
• python scripts/check_docs_weight.py

The existing aDDM five-draw sampling smokes ran only as compatibility tests; no benchmark, recovery study, evidence artifact, or scientific inference was produced.

Deferred

• arbitrary likelihood/predictive observation widths remain the focused 3B child
• synthetic 1D-4D lifecycle proof and public migration docs remain 3C
• dev-only response_kind / response_bounds become resolver adapters only in the final main -> dev integration

[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: 556728f6-330c-45f0-9247-6bdee71cf140

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.}