Controls Need Observations and Producers
What: A control's verdict needs both an observable subject and a practice that produces its required evidence. Repeated failures, accumulating exceptions and narrow passes deserve explicit interpretation; none alone establishes that a control is defective.
Why: A gate with no producing practice either blocks indefinitely or invites records written only to satisfy it. A constant verdict can hide a changing subject. An exception that never gets reviewed can outlive the condition that justified it. A narrow pass can be mistaken for a wider assessment when its exclusions are invisible.
Evidence: The review-ledger command's --open-round creates an incomplete scaffold: the review,
sweep and disposition still have to happen. The session producer emits boundaries, not hook execution
outcomes or usage. These are inspectable limits of the shipped mechanisms, not evidence that every
adopter observes the same failure or that these rules improve delivery outcomes.
How:
- R176 — Inspect constant verdicts against changing subjects. Print the comparison beside the verdict and state the observations it covers. A deliberate hold is a decision with a review date, recorded where the project records decisions. A local reminder does not establish that record or excuse the underlying failed check.
- R177 — Count widenings beside their subjects. A register that only grows is a design signal, not proof that its entries are wrong. Record its size and the subject's size; inspect them together.
- R178 — State the scope of a pass. Name what the run did not measure beside what it did. A pass over selected files is not a tree audit, and a session boundary is not a check result.
- R179 — Establish the producer before installing its gate. A control may not gate on a record for which the method has not established a producer. Name the producing practice and mechanism, the artifact it writes, the consuming control, and what happens when production fails. Install the mechanism with the gate. Automatic writers and human disposition practices both qualify; blank scaffolds do not. Where the practice is absent, state the obligation without selecting its gate.
The gate and producer inventory maps these obligations to the installable kit. Fresh installs select review-ledger enforcement only when the manual review practice is explicitly adopted; upgrades retain existing selection. This does not make platform review comments readable by the ledger checker. Delivery emission is advisory and cannot substitute for a required control's own inputs.
The constant-verdict history row, by-path health probe, expiring local health hold and cumulative widening counts remain proposed tooling. R176–R178 are normative obligations here, not claims that those mechanisms ship. The third shape—a narrower pass read as a wider one—is named without a new detector. Observation cadence limits every comparison; a human decision remains necessary when a diagnostic itself stays constant.