Task-corpus audit — 2026-08-01

State: open/active · Genre: analysis

Scope: every non-closed task in docs/planning/tasks/ (144 files) audited against the live codebase. Snapshot audited: 758baf72a; edits applied on 081937cea (three tasks spawned between the two — T-08HW, T-ZYBI, T-X8NO — are post-snapshot and not audited; T-B98V closed upstream mid-audit and was left alone).

Method: 12 parallel review agents, one per thematic cluster. Per task: full read; git log --all --grep=<id> commit trail; cross-reference sweep over milestones/decisions/code; cited paths and symbols resolved against the checkout; verdict only with code-level evidence. Medium-confidence closes were re-verified by the orchestrator before applying.

Outcome

Verdict	Count	Applied action
keep	126	last_reviewed: 2026-08-01; relevance_note where the codebase moved under the spec
shipped	5	4 → closed/done (T-FOJN, T-LWEH, T-2CXH, T-PTDG); T-B98V closed upstream
obsolete	6	5 → closed/obsoleted; T-TDES → closed/no-repro
superseded	2	closed/superseded (T-OB74 → D-WK7T, T-KIM9 → T-CF7M)
duplicate	2	closed/superseded into the survivor (T-XC32 → T-12HH, T-0X9M → T-0AM0)
unclear	3	kept + need_human_review: true (T-WML6, T-0PQK, T-WWNY)

need_human_review: true also set on T-BLYF (leak risk now structurally capped) and T-UWRG (every bun→pnpm driver got a Bun-native fix; the item encodes an owner decision).

Closes

Task	Close	Evidence anchor
T-FOJN task-work-durable-task-state-commits	done	commitToMainViaWorktree (lib/services/git/commit-to-main.ts) ships the whole mechanism
T-LWEH entity-title-accessor	done	PR #940 / b9b74ba80; task file never flipped
T-2CXH scope-release-roster-off-moon-app-vitest	done	sdlc.yaml roster comment cites its own retirement; polish-ci.yml (PR #757)
T-PTDG retire-pyyaml-emitter	done	PR #1100; yaml.ts now 108 lines over npm yaml
T-2UYF / T-7270 / T-LH85 polish-*	obsoleted	target tree frozen as apps/legacy-polish (39f911d6c); live polish is apps/determined/products/polish
T-521G pilot-consumer-polish	obsoleted	hard dep T-75UD closed/obsoleted (dist-repo channel retired for D-2APS); M-UWPO annotated
T-132J worktree-move-creates-parent-dir	obsoleted	sole caller already mkdirSync(recursive); migration spent
T-TDES investigate-bash-multiline-arg-glue	no-repro	single 2026-05-22 observation, zero recurrence; its own AC-1 sanctions the close
T-OB74 stash-roundtrip-worktree-guard	superseded	D-WK7T (accepted) removes the stash roundtrip rather than guarding it
T-XC32 task-acs-phrase-against-clean-fixtures	superseded	scope absorbed into T-12HH (Approach step 3 / AC-4)
T-0X9M probe-credits-session-observed-tools	superseded	same probe-false-positive scope as T-0AM0, which links it and carries the residual arm
T-KIM9 task-define-batch-skill	superseded	/sdlc:task-auto-define (T-CF7M) already batches; KIM9’s own gap note concedes the premise

Consolidation groups (encoded in relevance notes; merge when picked up)

Group	Members	Shape
Touchpoint cell-parser hardening	T-I8UZ, T-LF7V, T-TLJW	single-function tokenization fixes in parse-touchpoints.ts / _table_cells.ts — one PR
Paths claim-resolver precision	T-O66A (residual half), T-6S2N, T-HYSY, then T-L7XL	same collectCitations section-provenance plumbing; T-6S2N and T-HYSY pull opposite directions and must be designed together; T-L7XL must sequence after T-O66A
Files-to-touch enumeration rules	T-KS7C, T-KY0Y	T-KS7C’s own body requests the merge
AC authoring guidance	T-12HH (carrier), T-N0HI prose half, T-HQ56 substrate, T-MNGR	T-XC32 already folded in
Preflight probe family	T-DTAQ first (only ready member), T-0AM0 (+absorbed T-0X9M), T-NUML separate, T-0JTT pairs with T-9J63	three distinct problems, not five
Heartbeat lifecycle	T-IZLC, T-P6E5, T-S9SV	one combined design pass
Baseline capture family	T-7FPF + T-G6T3 (deliberate halves), T-LDH4 (same incident as G6T3), T-OLTA (+B-YNEX)
normalizeFinding extensions	T-BCNP (best-specified; promotion candidate), T-2HDX
audit-skill-runtime citations	T-4ZZX (keeper, fully authored), T-T8S2 (5 dangling doc paths live right now)	one regex generalization covers both
graph-scheduler chain	T-ROJC unblocked head; T-7OJL + T-BR6H natural PR-mates; T-8I15 premise fully live	all 11 shared the same stale “greenfield” Today

Ordering and collision repairs

• T-RNHP was open/ready with “Dependencies: none” while building on T-NXAM’s unimplemented doc-edit verb — hard dependency now recorded.
• D-EQGC (task-state planes, open/proposed) pre-empts or reshapes T-RNHP, T-WLNF, T-9TYJ, T-U3NR: decide that ADR before dispatching any of them (each now carries the warning in its relevance note).
• Same-line collisions: T-RNHP ↔ T-BR8T (start_task.ts:589); T-WLNF ↔ T-9TYJ (lint-state-origin.ts input schema). Serialize within each pair.

Capture gaps found (not fixed here)

• The T-SITE post-mortem’s third defect (an index.(md|mdx) alternation read as a literal path) has no tracking task; siblings T-LF7V and T-0O6S exist.
• B-6AHH carries the real spec for the templated T-I4F7; the no-lock finding suggests the true root cause is B-O8MK’s (tsc incremental-cache corruption).
• M-UWPO needs milestone review: two success criteria and one Wave-4 deliverable now point at obsoleted tasks (annotated in place).

Highest-value dispatchable work (post-audit)

1. T-OLTA — confirmed live one-line defect (run.ts:235-238 baselineDir default) breaking every worktree Step 7 run; B-YNEX corroborates.
2. T-DTAQ — only implementation-ready member of the probe family; BLANKET_EDIT_MODES still lacks auto.
3. T-B5ZA — clean spec, zero stale citations, no gating decision.

Systemic findings

The corpus-shape conclusions this audit feeds — spec-rot mechanics, the task-as-only-unit overload, ceremony scaling per instance, competing answers filed as parallel ready tasks, intake/triage vocabulary gaps — are argued with this data in the companion decision Corpus planes — work orders, process registry, product planning (see docs/planning/decisions/).

---

← Back to Notes