Data
Tickets in, reasons out.
The reviewer needs what a careful human reviewer would look at, and it leaves a written reason behind every call.
- Ticket
- Title, description, acceptance criteria
- Change
- Files and systems it touches
- Tests
- Planned or existing checks
- History
- Past incidents linked to the same area
- Out
- Verdict, critique, one-line reason
The batch is frozen into a snapshot when the review starts, so the reasons can always be matched to exactly what the critics saw. See the shared data pipeline.
The approver's decisions are the most useful data. When a person overrules the judge, that case is saved. Over time those overrides show where the critics are too strict or too loose, and the thresholds are adjusted by a person, not silently by the system.