This issue records a proposal. Implementation requires a selected work brief and separate execution approval.
Background and problem
The retained pilot describes reviewer contribution by change size. They do not establish an optimal reviewer count.
The completed library checkpoint found useful defects and substantial coordination cost. It did not enrol that project in this pilot.
Proposal
Define a protocol for up to three explicitly enrolled product deliveries. Describe each reviewer's:
- Reproducible contribution.
- Disagreement.
- Terminal outcome.
- Cost.
At the first enrolled delivery, propose a calendar expiry from expected cadence for human approval.
Scope and exclusions
Each delivery requires its own complete approved execution budget and a cycle ceiling if a loop is authorised. Twelve bounds initial review and triage sampling only, not a separate runtime allowance.
Do not restart the library checkpoint or make it a prerequisite.
Acceptance criteria
- These rules are explicit:
- Enrolment.
- Change-size classification.
- Observation.
- The pilot stops at three terminal deliveries or the approved expiry, whichever occurs first.
- Early expiry produces a bounded partial aggregate report.
- Failed or inconclusive cases cause no automatic expansion.
- Conclusions claim neither statistical power, defect absence nor calibrated budget constants.
Alternatives and open questions
Retain current single-reviewer delivery until the human approves a pilot protocol. Do not turn research references into universal reviewer-count rules.
References
This issue records a proposal. Implementation requires a selected work brief and separate execution approval.
Background and problem
The retained pilot describes reviewer contribution by change size. They do not establish an optimal reviewer count.
The completed library checkpoint found useful defects and substantial coordination cost. It did not enrol that project in this pilot.
Proposal
Define a protocol for up to three explicitly enrolled product deliveries. Describe each reviewer's:
At the first enrolled delivery, propose a calendar expiry from expected cadence for human approval.
Scope and exclusions
Each delivery requires its own complete approved execution budget and a cycle ceiling if a loop is authorised. Twelve bounds initial review and triage sampling only, not a separate runtime allowance.
Do not restart the library checkpoint or make it a prerequisite.
Acceptance criteria
Alternatives and open questions
Retain current single-reviewer delivery until the human approves a pilot protocol. Do not turn research references into universal reviewer-count rules.
References