Review changes to the quality bar itself

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

A coverage gate requires 70% and the candidate has 68%. A patch lowers it to 60% and skips a failing test, then appears green. Review both implementation and changes to the success bar.

Mechanism

Fix scope from branch baseline through the full tree, including staged, unstaged and new files. Compare thresholds, skipped/deleted tests, removed assertions, suppressions, stubs and exceptions; inspect meaning to distinguish legitimate replacement from weakening. Meet the original contract or record exception reason, scope, owner and expiry for authority-based ruling. An unavailable guard means unknown, not clean.

Bad example

Lower 70% to 60%, skip the failure and report the fix complete.

Good example

Compare the full patch with the original 70% bar. Report lowering and skip explicitly, repair uncovered behavior first, or propose a scoped, explained, expiring exception for ruling. Green cannot hide a lower bar, and text matches cannot alone reject equivalent test rewrites.

Why the change matters

When authors change checks too, green can come from weaker conditions. Reviewing the bar preserves the original meaning of success instead of substituting a pass label for behavior.

Observable expectation

Teaching 68% fails 70% and passes 60%, but the lowering is a finding. Removing redundant tests with preserved equivalent coverage needs semantic review. Missing baseline or a guard error makes checking incomplete.

Limits

Shallow diff guards can misflag refactors and miss semantic weakening. External tools also have limits/configuration. Standards can be revised with evidence and authority, not silently bypassed. Nondeterministic metrics require comparable conditions.