tomevault-io Code Review
Review establishes whether a proposed change is safe to advance. It is not a style pass and not a forge operation. The reviewer evaluates the change against the work unit, behavior contract, implementation plan, and verification evidence, then classifies findings as blocking or non-blocking.
Review Focus
Scope honesty. Confirm the change stays inside the claimed work. A change outside the work unit is either removed or filed as follow-up work before approval.
Correctness. Look for behavior regressions, broken interfaces, invalid schemas, unreachable graph paths, stale fixtures, and tests that no longer prove the behavior they name.
Semantic-shift detection. Treat "cleanup" skeptically when it changes meaning. Renames, schema vocabulary changes, and routing changes are semantic unless proven otherwise.
Evidence quality. Verification must be fresh and relevant to the claimed behavior. Passing commands are not enough when the tests do not cover the changed behavior.
Documentation impact. User-visible or authoring-surface changes require documentation that explains the current behavior. Historical decision records may name superseded designs, but current guides must not teach obsolete interfaces.
Fix-or-file discipline. Blocking findings must be fixed in the change before approval. Non-blocking findings may be filed as follow-up only when the current change remains correct without them.
Finding Classification
blocking: approval would ship incorrect behavior, misleading current docs, missing required evidence, scope creep, or a broken contract/interface.non-blocking: useful improvement that does not affect correctness, traceability, or the ability to continue the workflow.
Corruption Modes
rubber-stamp-review: approving because tests pass without checking whether the tests prove the claimed behavior.style-substitution: spending review attention on formatting preferences while missing behavior or contract drift.scope-amnesia: accepting unrelated changes because they are small.historical-doc-rewrite: editing decision records to hide superseded states instead of marking current guidance accurately.
Source: tesserine/groundwork — distributed by TomeVault.