Daily harness signal · September 15, 2026

Green tickets.
Missing feature.

One implementation lesson · Planning and integration

A correct dependency graph does not prove complete scope. Before delegating tickets, map every parent requirement to an actual owner and consumer-level proof; keep parent acceptance independent of child success.

Evergreen · source August 21, 2026

Close the gaps between tickets

Use when

A PRD becomes separately implemented and reviewed tickets, especially when new writers must preserve existing APIs and UI behavior. In Matt Pocock’s repository, a reporter describes serving-table hydration being excluded as “another ticket” that never existed. Producer checks passed; consumer paths remained broken.

Action

Before marking the breakdown ready, add this small coverage table to the existing parent record: 

Parent invariant | Owning ticket ID | Consumer check | Evidence
Complete serving model | T-07 | Existing read API and UI parity | Pending

The row is illustrative: replace T-07 with a real ticket. Review with this prompt: “For every accepted parent invariant, identify its owning ticket and observable end-to-end check. Resolve every ‘another ticket’ exclusion. Report uncovered requirements; do not invent owners or silently drop scope.” Block readiness on missing ownership. Require actual consumer evidence before parent completion; a filled table is not that evidence.

Acceptance check

Use an owned disposable fixture with producer success, rows present, and HTTP liveness all green, but omit the serving-model hydration ticket and its behavior. The coverage review must reject readiness, and unchanged read-API/UI checks must fail. Restore the real owner and complete serving data; then require those same checks to pass. Retain the parent revision, coverage rows, consumer outputs, and before/after results. Do not accept an empty successful response where the fixture expects records.

Evidence

Issue 924 specifies the omitted path, misleading checks, model, and fresh-session topology. The inspectable to-tickets source already requires complete vertical slices; the gap is proving parent coverage, not adding that slogan. Hamel Husain and Shreya Shankar’s workflow evaluation guide independently prioritizes user-goal success over step-level diagnostics.

Caveat

This is one unconfirmed field report, without public application code, raw sessions, or maintainer-confirmed repair. The coverage procedure and fixture are our untested adaptation, not a shipped skill change. A matrix can repeat an incomplete PRD: retain domain review and real consumer acceptance. Allow equivalent implementations; judge behavior, not identical diffs or mandatory tool sequences.

Compact source notes

  1. exwer: /to-tickets can leave PRD requirements orphaned — August 21, 2026; open report and same-day model/session clarification. Incident remains confidence_unconfirmed; not independent reproduction.
  2. Matt Pocock: to-tickets source — moving main, retrieved September 15, 2026; checked against raw main. Requires complete slices and permits expand–contract/integration-branch exceptions. No explicit parent-invariant coverage table in the inspected source. Retrieved history ends July 8 and does not establish live latest state.
  3. Husain and Shankar: How do I evaluate agentic workflows? — June 29, 2025. Method-level corroboration, not validation of this incident. Anthropic’s evaluation guide — January 9, 2026; distinguishes environment outcomes from transcript claims and warns about rejecting valid alternatives.
  4. Anchor lens: inspected omission report and source requirements. Unity lens: preserve accepted parent outcomes across decomposition; proposed invariant, not measured cross-harness equivalence. Procedure utility: confidence_likely. Accepting the deliberately uncovered requirement or broken consumer falsifies the proposed gate.