Checkwright

Checkwright v0.9.0

2026-07-19

Checkwright is the verification layer under agent orchestration, and this release makes two of its own delivery-tooling signals tell the truth. A cost report that carried a confounded number beside a clean one, and a recovery contract that read completed work as interrupted, were both reporting something other than what they measured. Neither fix adds a mechanism; both remove a claim the tooling could not support.

Tightened gates

None.

Renamed knobs

None.

Behavior changes

  • drift-kit/templates/economics.md — the /economics narrative no longer chains delegation-kit’s usage-trend as a third cost surface. The chain narrows to overhead-meter → stage-economics, and stage-economics becomes the sole cost-attribution surface. The budget-% rate-window footprint is account-wide — confounded by overlapping sessions and by a second operator on the same account — so it was the wrong instrument for per-iteration attribution, and carrying it beside the per-transcript token SSOT put a confounded advisory number next to a clean one a reader could over-trust as the iteration’s cost. This removes a role, not a tool: bin/usage-trend.sh is untouched, and usage-verdict and the agent-budget-guard all stand. Reconcile only if you copied the template out and expect three cost surfaces in your report.

  • delegation-kit/SPEC.md — the resume-journal recovery contract’s DONE-absence clause is rescoped onto whether the supervisor consumed the agent’s return. Agents routinely complete without appending DONE, so the former unconditional reading (“a journal without DONE means that unit was interrupted”) false-read finished work as interrupted. Now: on the ordinary path the consumed return plus the supervisor’s post-commit verification is the recovery contract, so the marker is redundant and its absence implies nothing; only in a cold read — a journal found with no return ever consumed, the agent’s session having died before returning — is the marker the sole signal, and there the original reading holds. The inline-findings rule is unchanged and is now stated as exactly that cold arm’s insurance. Mirrored into templates/agent-execution.md. Reconcile if your supervisor tooling treats a DONE-less journal as an interruption signal on the ordinary path.

  • lifecycle-kit/templates/skills/scope.md — the template’s opening and closing lines claimed queue promotion unconditionally while its middle narrowed promotion to debt only where the roster splits out a dedicated authoring stage. Both ends now carry that conditional, and the reason is stated with its owner: writing the amendment is promoting the entry (canon-kit/SPEC.md §The amendment lifecycle), so the stage that authors is necessarily the stage that promotes. No ruling changed — this states what was already ruled, at the two places a scope session actually reads.

Upgrading

Sync the vendored kit directories wholesale at v0.9.0, regenerate the generated artifacts (the pre-commit hook and the graph projection), then run the full battery.

No allowed reds. Tightened gates is empty, so a clean upgrade turns nothing red; the behavior changes above are reconciled by reading, not by a gate.

The behavior changes above are declared for reading, not a mechanical scan. If a gate reds that this note does not name, the upgrade smoke was supposed to catch it first — open an issue, because that is a defect in the release rather than work for you.