coverage-gate

Run every lane bound to the trigger, collect reports from lanes with coverage set to include, and merge them into one total. Thresholds apply to the merged result, using an exact per-line union when all reports carry line detail.

Absolute mode requires all metrics to meet configured thresholds. No-decrease mode records a per-branch baseline on first run and only blocks when coverage drops below that baseline, while critical-path thresholds stay fixed in both modes.

A report with zero measurable lines scores 100% under the 0/0 convention, so the gate treats it as a failure. Also check that each lane writes its own coverageSummaryPath, since shared paths cause silent undercounting.

Each test's assertions are classified by level: behavioral checks like return values and state rank Level 1-5, while mock call and argument checks rank Level 6-7. A test whose assertions are all Level 6-7 fails the quality gate.

Unmeasured metrics are treated as null, producing a warning rather than a failure. For example, coverage.py reports no functions and go-cover reports no branches, so those thresholds should be set to 0.

The v8 ignore next directive silently fails on nullish coalescing, ternaries, catch bodies, and short-circuit operators. The gate flags these and requires v8 ignore start and stop range pairs instead.