Finding (Recorded-only — review IMP-4)
docs/explain/sdl/parser.md:51-53 claims file-backed deterministic expansion; no test runs the pipeline twice and asserts identical output. The design (lockfile-backed imports, no ambient state) is plausibly deterministic — the claim simply lacks its witness.
Scope (one PR)
- Test: parse → instantiate (with variable substitution) → compile a representative scenario twice in separate passes; assert canonical JSON serializations are byte-identical (excluding any explicitly timestamp-typed fields — enumerate them in the test).
- Cover ordered-collection surfaces (multiple agents, features, relationships) and at least one module-import scenario so expansion order is exercised.
- Run one pass with
PYTHONHASHSEED variation (subprocess) to catch hash-order dependence.
Acceptance criteria
- The witness runs in CI; parser.md's determinism claim cites it.
Related
#262 (ASR-514 Determinism And Stability Verification) — this is the narrow witness for the existing documented claim; ASR-514 remains the broader requirement.
Finding (Recorded-only — review IMP-4)
docs/explain/sdl/parser.md:51-53claims file-backed deterministic expansion; no test runs the pipeline twice and asserts identical output. The design (lockfile-backed imports, no ambient state) is plausibly deterministic — the claim simply lacks its witness.Scope (one PR)
PYTHONHASHSEEDvariation (subprocess) to catch hash-order dependence.Acceptance criteria
Related
#262 (ASR-514 Determinism And Stability Verification) — this is the narrow witness for the existing documented claim; ASR-514 remains the broader requirement.