Three-layer review keeps decisions visible

A useful review model separates agent findings, owner verification, and the team decision. This prevents an automated review from becoming an implicit approval.

Each layer has a different question and evidence threshold.

LayerQuestion
L1 agent reviewWhat risks or gaps are visible in the checked scope?
L2 owner reviewAre findings relevant and reproducible?
L3 human decisionIs the remaining risk acceptable for merge or release?

Review checkpoints

  1. Define the scope and review focus.
  2. Collect findings with file, line, and evidence.
  3. Reproduce important findings or mark them as hypotheses.
  4. Check the actual diff and required tests.
  5. Record GO, HOLD, or the next required action.

Decision note

Scope: <paths>
L1 result: <findings>
L2 result: <confirmed gaps>
Required approvals: <roles>
Residual risk: <list>
Decision: GO / HOLD
Status: documented example; approval result: Unknown.

Canonical lesson source · JavaRush

Open JavaRush material