qa: real known-RED corpus fixture for narration_no_ooc_leak (replaces #977 TODO)

[100yenadmin]

Follow-up to #977

#977 closed the gate-corpus coverage gap for narration_no_ooc_leak with a TODO placeholder (the canonical builder auto-flag). That satisfied the coverage audit but left the anti-goalpost corpus not actually exercising the check's RED path — so a future edit that weakened _NARRATION_LEAK_RE or the n_leak>=3 / dm_text>=MIN_BEATS fatal threshold would slip past the corpus (only test_assert_behavioral.py would catch it).

This replaces the TODO with a faithful minimal known-RED fixture, modeled on the real 2026-06-17 craft audit (#972).

The fixture (qa/gate_corpus/cases/narration_no_ooc_leak/)

6 DM text beats (== MIN_BEATS), 4 leaking the exact OOC patterns the gate bans, 2 clean in-fiction beats:

• as the player character · continuity check · inciting incident · here's how round one actually went

Isolation (so the per-case corpus test's failed == {expected} holds): a clean roll keeps dice_used green; no companion keeps dm_voices_characters inert; baseline state (day=2, 2 visited, xp>0) passes the world/xp floors; no chat/moves keeps the facade lane inert; 6 (<10) beats + no companion keeps structural_completeness inert.

Verification

Gate run directly on the fixture:

RED: 1 behavioral assertion(s) FAILED.
  [FAIL] narration_no_ooc_leak — 4 player-facing beat(s) leaked OOC craft-scaffolding/bookkeeping/system-vocab — e.g. 'Now let me seat Dal Lightspark as the player character...' [pervasive => RED]

Exactly one FATAL, fully isolated.

$ uv run --directory servers/engine python -m pytest \
    ../../qa/test_behavioral_gate_corpus.py ../../qa/test_root_cause_analyzer.py -q -p no:xdist
39 passed   # was 38 passed + 1 skipped (the TODO is now a real, exercised case)

Builder regeneration is deterministic — zero churn to the other 18 fixtures (diff is the new case dir + the manifest TODO→real-case swap + the generator). assert_behavioral.py untouched.

Summary by CodeRabbit

• Tests
  • Enhanced test coverage for narration validation scenarios to improve quality assurance.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: e7a78d77-ebc0-4302-9d64-ba3b0e99a592

📥 Commits

Reviewing files that changed from the base of the PR and between 5544ab1 and c75d48f.

📒 Files selected for processing (4)

• qa/gate_corpus/builder.py
• qa/gate_corpus/cases/narration_no_ooc_leak/run.jsonl
• qa/gate_corpus/cases/narration_no_ooc_leak/state.json
• qa/gate_corpus/manifest.json

---

📝 Walkthrough

Walkthrough

Promotes the narration_no_ooc_leak behavioral gate from a TODO placeholder to a complete corpus case. builder.py gains a provenance entry, a fixture builder function emitting OOC-leak DM text beats, and a _CASES_SPEC registration. Matching run.jsonl, state.json, and manifest.json entries are added.

Changes

narration_no_ooc_leak corpus case

Layer / File(s)	Summary
Fixture builder and corpus registration qa/gate_corpus/builder.py	REAL_RED_PROVENANCE gains the narration_no_ooc_leak key; case_narration_no_ooc_leak() builds a base roll run and appends DM turns with multiple OOC-leak phrasing patterns; _CASES_SPEC registers the case so the manifest emits expected_red_check: narration_no_ooc_leak.
Generated fixture artifacts and manifest entry qa/gate_corpus/cases/narration_no_ooc_leak/run.jsonl, qa/gate_corpus/cases/narration_no_ooc_leak/state.json, qa/gate_corpus/manifest.json	run.jsonl records the mcp__engine__roll tool call, its result, and the leaking narration turns; state.json captures the pc1/Dal party snapshot at loc_camp; manifest.json replaces TODO__narration_no_ooc_leak with the fully specified case object.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• electricsheephq/WorldOS#972: Implements the narration_no_ooc_leak behavioral gate in qa/assert_behavioral.py that this PR's corpus case directly targets.
• electricsheephq/WorldOS#977: Added the TODO__narration_no_ooc_leak placeholder to manifest.json that this PR replaces with the real fixture.

Poem

A rabbit hops past the tavern door,
No OOC whispers shall seep through the floor.
The corpus is fed, the fixture complete,
The gate stands firm — no leaks to defeat!
🐇✨ The TODO is gone, the test is neat!

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (1 warning, 1 inconclusive)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.
Description check	❓ Inconclusive	The description provides comprehensive context (background from #977, the problem it solved, fixture details, verification steps, and impact), but does not include the required CLA and licensing checkboxes from the template.	Add the Licensing/CLA section with required checkboxes and validation checklist from the template to ensure compliance with contribution guidelines.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately summarizes the main change: replacing a TODO placeholder with a real, known-RED corpus fixture for the narration_no_ooc_leak check, including the reference to the prior issue.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

---

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