quality-pass-pi

Metadata

Description

## Purpose
Review the connected Pi evaluation-plane recovery batch before any implementation runs. Inspect every downstream task for clear boundaries, same-file serialization, concrete validation, credential-free fixtures, live human-flow coverage, and consistency with the user's intent: Pi-first evaluation, Codex retained, lazy uncluttered satellites, no evaluation of tasks that never ran, and genuinely understanding FLIP.

Edit downstream WG task descriptions/dependencies where needed. Do not inspect or modify source code. Do not create additional implementation fan-out unless a missing safety prerequisite is concrete.

## Validation
- [ ] Every downstream code task has test-first and live-flow criteria.
- [ ] Tasks likely to touch the same lifecycle/coordinator files are sequential.
- [ ] The rollout cannot globally enable evaluation before canaries pass.
- [ ] Default bounded evaluation and deep read-only FLIP are clearly distinguished.
- [ ] Pi docs/API research and executor-neutral fallbacks are covered.
- [ ] Quality pass records its review in the task log and completes.

Depends on

• (none)

Required by

• done deep-survey-of
• done design-pi-evaluation

Log

• 2026-07-26T10:21:31.435640865+00:00 Task paused
• 2026-07-26T10:25:30.969034725+00:00 Task published
• 2026-07-26T10:25:31.642751326+00:00 Spawned by coordinator --executor pi --model openai-codex:gpt-5.6-sol --isolation shared-nonwriting-policy
• 2026-07-26T10:25:44.338494025+00:00 Starting downstream graph quality review; will inspect task descriptions/dependencies only and avoid source code.
• 2026-07-26T10:26:46.493316573+00:00 Reviewed all 8 downstream tasks. Gaps: dedicated/deep/integration tasks lack explicit RED-first + real human-flow fixtures; native Codex retention/no-silent-cross-executor fallback needs sharper criteria; integration says bounded FLIP instead of bounded evaluation; admission-deferral and lazy lifecycle writers are not serialized behind the authoritative transition implementation.
• 2026-07-26T10:26:46.547191695+00:00 FANOUT_DECISION: stay inline — this is a bounded graph-description/dependency quality pass; no missing prerequisite warrants new tasks.
• 2026-07-26T10:30:11.230931300+00:00 QUALITY REVIEW: inspected all downstream tasks: deep-survey-of, design-simplified-authoritative, design-pi-evaluation, implement-authoritative-lifecycle, implement-lazy-evaluation, implement-dedicated-pi, implement-deep-read, integrate-and-canary; also reviewed permitted prerequisite fix-admission-deferrals. Deep survey and lifecycle design already had clear research-only boundaries/live disposable evidence, so no edit.
• 2026-07-26T10:30:12.693673108+00:00 GRAPH SERIALIZATION: authoritative lifecycle -> fix-admission-deferrals -> lazy evaluation -> dedicated bounded Pi lane -> deep-readonly FLIP -> integration/canary. Added implement-authoritative-lifecycle as a dependency of fix-admission-deferrals and fix-admission-deferrals as a dependency of implement-lazy-evaluation so overlapping coordinator/lifecycle writers cannot run in parallel. Direct Codex/Claude task scope was not edited.
• 2026-07-26T10:30:13.045460442+00:00 VALIDATION HARDENING: every code task now requires named RED-first tests, credential-free fixtures, actual daemon/terminal/TUI/Viz human-flow validation, permanent smoke ownership, and full Rust checks. Never-ran cases explicitly include deferral, launch failure, cancellation, skip, open work, message, and reconciliation.
• 2026-07-26T10:30:13.470945005+00:00 INTENT/ROLLOUT: design now separates default bounded no-tools evaluation from selective genuine deep-readonly FLIP; requires Pi docs/API evidence, no silent cross-executor fallback, and native Codex retention. Rollout begins disabled, rejects stage skipping, requires credential-free plus live Pi canary evidence, enables advisory bounded evaluation only afterward, and forbids global hard gating/global FLIP in this batch.
• 2026-07-26T10:30:49.763894127+00:00 Validated task-specific criteria: all Pi/lifecycle code tasks plus fix-admission have explicit test-first, credential-free, live human-flow, permanent-smoke criteria; shared coordinator work is serialized; disabled-to-canary-to-advisory rollout is enforced; bounded evaluation/deep-readonly FLIP and executor-neutral no-silent-fallback semantics are explicit. No source files were inspected or modified, so no artifact, build, or commit applies.
• 2026-07-26T10:30:50.977942134+00:00 Task marked as done