Parent epic: #273 · Spec: docs/superpowers/specs/2026-05-20-codex-pivot.md §8, §11 (#N)
Scope
A short results log extracted from shadow mode capturing the data needed to evaluate the kill-criteria gate in spec §8.
Required fields (structured-table schema, spec §8)
- PR count observed during shadow window
- False-positive rate: Claude vs Codex (per-PR comparison)
- Latency: Codex review post time after PR open / commit push
- BLOCKING-finding miss check: any PR where one gate flagged BLOCKING and the other did not
- Decision: proceed with cutover / extend shadow / revert
- Decision date
Kill criteria (revert triggers, from spec §8)
- BLOCKING miss by Codex on any PR
- Codex false-positive rate > 2× Claude false-positive rate
- Codex review latency > 30 min on > 20% of PRs
Acceptance
- Log committed as
docs/codex-pivot-shadow-decision-log.md (or attached as comment on this issue) before #H closes
- Decision documented with date, decider, rationale tied to the observations above
Attach to
The sub-issue that closes the decision gate (likely #H, when the Claude gate requirement is dropped).
🤖 Generated by Claude Code on behalf of @cbeaulieu-gt
Parent epic: #273 · Spec:
docs/superpowers/specs/2026-05-20-codex-pivot.md§8, §11 (#N)Scope
A short results log extracted from shadow mode capturing the data needed to evaluate the kill-criteria gate in spec §8.
Required fields (structured-table schema, spec §8)
Kill criteria (revert triggers, from spec §8)
Acceptance
docs/codex-pivot-shadow-decision-log.md(or attached as comment on this issue) before #H closesAttach to
The sub-issue that closes the decision gate (likely #H, when the Claude gate requirement is dropped).
🤖 Generated by Claude Code on behalf of @cbeaulieu-gt