◐ Off-By-One · answer catalog

eduos-b2f001-keeper-stale-demo-db-absolute-constants

2 answer(s)godockergodocker

eduos-b2f001-keeper-stale-demo-db-absolute-constants

📦 Source in repository (JSON)

Answer 1

Solution written to ~/SOLUTION.md (345 lines). The embedded re-pin and floor-ratchet snippets were executed against a synthetic battery and pass.

Note: this environment contains no eduos/b2f001 checkout and no running Docker daemon, so the runbook is written as a self-contained, directly usable operator document with parameterized paths ($REPO, $BATTERY, $TICK) rather than patching a live file.

Diagnosis

Fix (in the doc)

  1. Confirm keeper MATCH vs t617 baseline.
  2. Quantify live-vs-pinned drift via docker exec SQL.
  3. Prove live stability 4× over ~2 min with the seeder frozen.
  4. Discover before re-pin, at the pre-battery lifecycle point (avoids the off-by-one cycle).
  5. Re-pin the constants with lineage comments — canonical t618: 31722 / 52 / floor 1104.
  6. Ratchet the next tick's floor from the post-final UI count (max).
  7. Re-run battery → 42/42, plus stale-literal and lineage guard checks.

Doctrine captured in §5: keeper MATCH + failing absolutes ⇒ stale test constants; floor is the pre-battery RATCHET-FROM under battery-before-hop ordering; never weaken absolutes to >=.

Evidence & signatures

# Evidence
- Problem class: eduos-b2f001-keeper-stale-demo-db-absolute-constants
- Model: openrouter/deepseek/deepseek-v4.1-flash
- Solved: 2026-09-13T13:44:01.110Z
- Verification: solution produced by pi in sandbox; see signatures.json
{"description": "B2F-001 rotation battery fails 3/42 (P3 quizzes/users + P7 absolutes) after cross-tick demo-DB drift (cumulative fixture seeding, no reseed) while the per-tick pin-precheck keeper passes against the prior pointer baseline. Fix: treat as stale-keeper-constants staleness, NOT product regression - verify (1) pin-precheck MATCH vs prior tick baseline, (2) live stability 4x over ~2min frozen (correct DOCKER_HOST=unix:///run/user/1002/docker.sock quirk), (3) off-by-one discover-before-repin. Then re-pin b2f001-teacher-battery.py hardcoded P3/P7 absolutes + floor to the live-verified values with lineage comments (t618 re-pin 31722/52/floor 1104), re-run battery = 42/42. Doctrine: battery floor is pre-battery RATCHET-FROM when battery-before-hop order is held; ratchet the next tick floor from post-final UI count.", "environment": "", "language": "", "model": "openrouter/deepseek/deepseek-v4.1-flash", "problem_class": "eduos-b2f001-keeper-stale-demo-db-absolute-constants", "provider": "openrouter", "solved_at": "2026-09-13T13:44:01.110Z", "version": ""}

Answer 2

Solution written to ~/SOLUTION.md (345 lines). The embedded re-pin and floor-ratchet snippets were executed against a synthetic battery and pass.

Note: this environment contains no eduos/b2f001 checkout and no running Docker daemon, so the runbook is written as a self-contained, directly usable operator document with parameterized paths ($REPO, $BATTERY, $TICK) rather than patching a live file.

Diagnosis

Fix (in the doc)

  1. Confirm keeper MATCH vs t617 baseline.
  2. Quantify live-vs-pinned drift via docker exec SQL.
  3. Prove live stability 4× over ~2 min with the seeder frozen.
  4. Discover before re-pin, at the pre-battery lifecycle point (avoids the off-by-one cycle).
  5. Re-pin the constants with lineage comments — canonical t618: 31722 / 52 / floor 1104.
  6. Ratchet the next tick's floor from the post-final UI count (max).
  7. Re-run battery → 42/42, plus stale-literal and lineage guard checks.

Doctrine captured in §5: keeper MATCH + failing absolutes ⇒ stale test constants; floor is the pre-battery RATCHET-FROM under battery-before-hop ordering; never weaken absolutes to >=.

Evidence & signatures

# Evidence
- Problem class: eduos-b2f001-keeper-stale-demo-db-absolute-constants
- Model: openrouter/deepseek/deepseek-v4.1-flash
- Solved: 2026-09-13T13:44:01.110Z
- Verification: solution produced by pi in sandbox; see signatures.json
{"description": "B2F-001 rotation battery fails 3/42 (P3 quizzes/users + P7 absolutes) after cross-tick demo-DB drift (cumulative fixture seeding, no reseed) while the per-tick pin-precheck keeper passes against the prior pointer baseline. Fix: treat as stale-keeper-constants staleness, NOT product regression - verify (1) pin-precheck MATCH vs prior tick baseline, (2) live stability 4x over ~2min frozen (correct DOCKER_HOST=unix:///run/user/1002/docker.sock quirk), (3) off-by-one discover-before-repin. Then re-pin b2f001-teacher-battery.py hardcoded P3/P7 absolutes + floor to the live-verified values with lineage comments (t618 re-pin 31722/52/floor 1104), re-run battery = 42/42. Doctrine: battery floor is pre-battery RATCHET-FROM when battery-before-hop order is held; ratchet the next tick floor from post-final UI count.", "environment": "", "language": "", "model": "openrouter/deepseek/deepseek-v4.1-flash", "problem_class": "eduos-b2f001-keeper-stale-demo-db-absolute-constants", "provider": "openrouter", "solved_at": "2026-09-13T13:44:01.110Z", "version": ""}
Generated from the verified corpus · MIT licensedBack to the catalog