[anth-volk]

Fixes #937

Same-Repository Draft PR Check

• This PR is a draft.
• This PR is from a branch in PolicyEngine/policyengine-sim-api, not a personal fork.

Summary

Sixth of nine stacked PRs; it builds on #936. It splits the artifact precompute, the Stage 12 cache precompute and the precompute tests without changing behaviour, and removes all three from the file-length baseline (count 7 to 4). For each source file, a script compared every top-level statement of the old module with the new modules: all appear exactly once and unchanged apart from the renames below, in their original order.

precompute.py (931 lines) and stage12_cache/precompute.py (844 lines)

Both are split the same way. The original module keeps the manifest functions, the functions dataclass and run_precompute, and re-exports every public name in __all__. The Modal precompute apps import the step functions from it inside their function bodies, so they, and the app tests that replace those functions on the module, are unchanged.

Module	Artifact precompute	Stage 12 cache
precompute.py: manifest, functions dataclass, run_precompute, re-exports	276	226
precompute_planning.py: cohorts or baseline identities, state dataset targets, plan_artifacts_impl, select_work	271	350
precompute_datasets.py: national and per-state dataset builds	276	241
precompute_baselines.py: computing baselines and checking determinism	207	134

Renames

Private names another module now uses are public: _cohort_resolution to cohort_resolution (artifact precompute), and _scoped_planned_simulation to scoped_planned_simulation (docstring added) and _baseline_identity to baseline_identity (cache precompute). stage12_worker_validation.py imported _scoped_planned_simulation from the cache precompute; it now imports scoped_planned_simulation from stage12_cache/precompute_planning.py.

Tests

• tests/test_precompute.py (1,835 lines) keeps the schema, planning and orchestration tests (575 lines); the rest move to test_precompute_planning.py (428), test_precompute_datasets.py (443) and test_precompute_baselines.py (417). 71 tests before and after, with the same names; a script compared every class method and statement: all unchanged apart from patch targets.
• Patch targets, each checked against the modules that read the name:
  • artifact precompute: TestPlanArtifactsImpl patches cohort_identity and state_dataset_targets in precompute_planning; TestBuildStateDatasetsImpl patches state_dataset_targets in precompute_datasets, where the state dataset build reads it; TestVerifyDeterminismImpl patches _prepare_cohort_baseline in precompute_baselines.
  • cache precompute (tests/test_stage12_cache_precompute.py, not split): the stubbed_planning fixture and test_states_on_the_default_dataset_add_no_state_datasets patch precompute_planning, which plan_artifacts_impl reads; test_state_targets_follow_how_a_request_for_the_state_resolves patches _country_module there too; the state_build fixture patches precompute_datasets; the determinism tests patch _ensure_dataset_file, scoped_planned_simulation and Stage12CacheStore in precompute_baselines; and the _use_store helper replaces Stage12CacheStore in all four modules, each of which opens the store. test_planned_baseline_file_is_the_one_the_worker_saves_and_loads calls scoped_planned_simulation from precompute_planning.
  • tests/test_stage12_worker_validation.py still patches state_dataset_targets on the cache precompute module, which the validation reads as a module attribute.

Environment variables

No change: the moved code reads the same variables, in the same functions, and no workflow, image or deployment file changes.

Testing

• Executor, full suite: 1,873 passed, 22 skipped, as before.
• Pyright for the executor: no errors.
• Ruff 0.15.22 format and lint: clean.
• python scripts/run_quality_guards.py and python -m scripts.rules guard --base-ref origin/file-length-split-stage12-runtime-tests: passed; python -m scripts.rules render leaves no diff.

Only the lowest PR in the stack runs full CI, so this PR's CI complete check fails by design until the PRs below it merge.

🤖 Generated with Claude Code