Corpus Cleanup Status — 2026-06-01¶
Date: 2026-06-01
Scope: Post-session state of the 318 solo curated scenarios in production after the Jun 1 cube-rule cleanup pass. Reference back to scenario-corpus-audit-2026-05-30.md for the original n=45 audit findings.
TL;DR¶
Six migrations applied to prod today (118, 119, 120, 121, 122, 123) — total: 51 scenarios fixed across earn-the-deadline (all three tiers), advisor-without-stake, Death Ship Memorial redesign. Corpus-wide earn-the-deadline cleanup is now complete — every flagged scenario across Tier 1, Tier 2, and Tier 3 has been audited and either passes or has been fixed.
What remains is not corpus cleanup but corpus expansion: structural cube cells the corpus doesn't produce at all (self/dyad scope, trivial stakes, they-know-more topology, favors-action default). These require new authoring, not editing.
The cleanup unblocks the starter pack v3 work, which is the original reason this session started.
What got fixed this session¶
| Mig | Date | Scenarios touched | What changed |
|---|---|---|---|
| 118 | Jun 1 | 7 — Ethics Board, All-Hands Address, The Pen, The Terms, Two Sources, Sensor Drift, Sixty-One | Advisor-without-stake → either renamed to oversight/collective-vote authority (4) or material stake surfaced (3). Per project-player-must-have-authority-or-stake. |
| 119 | Jun 1 | 2 — Treaty Marriage, Prison Pipeline | Earn-the-deadline anchors added to scan strip + body. Per generator v1.8.2. |
| 120 | Jun 1 | 26 — Tier 1 vague-deadline scenarios across 5 themes (boardroom 5, court 4, frontier 2, rebuild 11, underground 4) | Scan strip deadline rows rewritten to name one of four valid anchor patterns. Per-scenario review pass against full bodies before apply; 10 of 26 revised mid-pass to harmonize with body. |
| 121 | Jun 1 | 4 — The Pen, Two Sources, Sensor Drift (back-review of 118), The Runner (missed in 120 due to title escaping) | Cross-check between 118's rewrites and downstream fields (options + cast block stake fields) surfaced 3 real conflicts (dual-signer dynamic, editor-in-chief vs cast Editor, navigator vs Navigation Officer Khay hierarchy). |
| 122 | Jun 1 | 1 — Death Ship Memorial | Full redesign: body expanded ~245 words with named voices; cast block added (Maren Voss, Station Chief Imani, Chief Engineer Kade); scan strip given earned 90-day-decay + 21-day-memorial anchor; question compressed 365 → 92 chars; option descriptions rewritten with parallel Cost lines. |
| 123 | Jun 1 | 23 — Tier 2 + Tier 3 deadline scenarios across 5 themes (boardroom 1, court 7, frontier 3, rebuild 6, underground 5) | Scan strip deadline rows rewritten. Per-scenario review against full bodies completed; each anchor cross-checked to use a body time-reference or extend body context without conflict. |
Net effect: 51 scenarios brought into compliance (some appear in multiple migrations as a planned sequence — e.g., The Pen in 118 then 121, The Two Sources in 118 then 121 then 123). The advisor-without-stake violations are fully resolved. All three tiers of earn-the-deadline violations are fully resolved. Death Ship Memorial is no longer scenario-incomplete.
What remains open¶
Tier 2 + Tier 3 earn-the-deadline — RESOLVED in migration 123¶
The Jun 1 audit heuristic flagged 74 of 318 scenarios for missing deadline anchors. Triage classified them: - Tier 1 (26): clear violations → fixed in migration 120 - Tier 2 (17): has SOME anchor but soft → resolved in migration 123 - Tier 3 (31): heuristic uncertain → resolved in migration 123
Per-scenario triage of the 48 Tier 2/3 scenarios against full bodies produced: - ~25 PASS — scan strip + body together earn the deadline (The Draft's "commissioner walking to microphone," Twenty Engineers' "water system failing now," Spymaster's Loyalty's "if she learns through any other channel," Fuel Dilemma's wormhole window, etc.) - 3 SHARPEN — body has concrete anchor scan strip didn't reflect (Two Sources, Radio Frequency, Winter Stockpile) → fixed in mig 123 - 20 FIX — no anchor anywhere (Blockade, Court Physician's Dilemma, Fallen Knight, Heretic's Book, Merchant's Justice, Silver Mine, Alien Artifact, Immortality Question, Ore Still Runs, Broken Promise, Monument Debate, Old World Library, Outsider's Skill, Sixty-Three, Encrypted Server, Exit Interview, Old Guard, Take Split, Ducal Divorce, State Response) → fixed in mig 123
Corpus-wide earn-the-deadline cleanup is now complete. 51 scenarios total fixed across migrations 119 (2 scenarios), 120 (26 scenarios), 123 (23 scenarios). The remaining scenarios either passed audit or weren't flagged.
Bias 6 (authority closure when full + domain expert) — clean in sample¶
The Jun 1 sample (n=16) found 0 violations. Likely clean corpus-wide because the v1.10.0 generator rule has been baked in since May 31 and most scenarios predate or were rewritten through it. If a future audit surfaces a violation, the migration pattern is established (add a one-sentence closure to the body).
Bias 7 (parallel option-cost concreteness) — clean in sample¶
Same — 0 violations in the n=16 sample. Death Ship Memorial was the one scenario in this session where options lacked Cost lines (resolved in migration 122). If others exist, they're flaggable by reading the four option descriptions side-by-side.
Structural cube gaps — corpus expansion, not corpus cleanup¶
The May 30 audit identified empty cube cells: self/dyad scope, trivial stakes, full reversibility, they-know-more topology, favors-action default, seconds/minutes deadline. These can't be fixed by editing existing scenarios — they require new authoring.
Decision settled this session (the audit254 conversation's open question): Don't fight surface gravity. Author new scenarios on whichever surface produces the cube cell cleanly. Per feedback_governance_is_just_a_surface: governance is one surface, not a privileged category. Career, health, relational, financial, civic surfaces are coming anyway as part of multi-domain expansion (per the user's intent flagged this session). Those surfaces naturally produce self/dyad scope, they-know-more topology, and trivial-meaningful magnitudes the world-themed corpus doesn't.
The corpus-cleanup work is "make existing scenarios pass the rules." The corpus-expansion work is "add scenarios that fill the structural gaps." Different efforts; the second hasn't started.
Starter pack handoff state¶
The original reason this session started: get the corpus to a state where the starter pack v3 design could proceed with confidence.
Where the starter pack design landed before the corpus cleanup detour:
- Starter pack reframed under the cube + ADR-074: it's a cross-domain MCP demonstration, not a personal-values onboarding pack. Goal is to give a new Claude MCP user a felt sense within 5–10 games that the same judgment pattern shows up across surfaces.
- v2 pack (6 scenarios in
docs/starter-pack/scenarios.md) is corpus content, not the starter pack. Heavy on relational × integrity, life-defining-skewed, only 3 surfaces sampled. - Curated footprint inside games 1–20 decided as Option B: Game 1 specifically curated + Games 2–5 curated + Games 6–20 filtered from cube-tagged corpus.
- Game 1 drafted: The $20 Loan to Sam in
docs/starter-pack/v3-redesign-draft.md. Lowest plausible cube vector. Thin-but-real. - Games 2–5 not yet authored. Cube targets per draft framework: dyadic-personal, workplace-everyday, time-energy, financial-small, creative-strategic — surfaces the existing corpus doesn't cover.
- Full UX flow (overlay vs skip, MCP-connect placement, when coherence% reveals) deferred to a separate session with CC.
What unblocks now: with the corpus genuinely Bias-6/7-compliant and the worst earn-the-deadline violations fixed, Games 6–20 can be served from filtered corpus without quality fear. Author Games 2–5 next session.
Methodology notes for future cleanup passes¶
- Heuristic flagging is fast but noisy. The Jun 1 pass first flagged 125 of 318 (39%) using a narrow regex, then 74 (23%) with broader patterns. Real violation rate after triage was ~12% (Tier 1 only). Run the heuristic to surface candidates; never trust it as a final count.
- Per-scenario body review BEFORE applying is worth the time. Migration 120's pre-apply review caught 10 of 26 anchors that would have created body/scan-strip conflicts. Inline show-then-apply (migrations 118, 119) caught prose issues but missed options/cast-block conflicts — migration 121 was the back-review cleanup.
- Migration scope: scan-strip-only updates are cheap; description rewrites are expensive. Migration 120 was 26 scan-strip updates and ran in seconds with low risk. Migration 122 was a full single-scenario rewrite and required a longer review pass. Bias toward scan-strip-only where the rule allows it.
- Title escaping in the DB is a hidden gotcha. The Runner from Fenn's Crossing has literal
''(two single-quote characters) in its title, not an apostrophe. Standard SQL escaping (Fenn''s) parses as a single apostrophe and misses the row. UseILIKE 'pattern%'for safe matching when titles contain apostrophes.
Migration index (Jun 1)¶
migrations/118_advisor_stake_fixes.sql— 7 scenarios, advisor-without-stake fixmigrations/119_earn_the_deadline_fixes.sql— 2 scenarios, deadline anchorsmigrations/120_earn_the_deadline_tier1.sql— 26 scenarios, Tier 1 scan-strip-only deadline fixesmigrations/121_back_review_fixes.sql— 3 back-review fixes on 118 + Runner re-applymigrations/122_death_ship_memorial_redesign.sql— 1 scenario, full redesignmigrations/123_earn_the_deadline_tier2_3.sql— 23 scenarios, Tier 2 + Tier 3 scan-strip-only deadline fixes