Measured

Which rules teams silence

Every scan that opts into telemetry sends the rule ids that fired and the ones the team had silenced in .gatetestignore — integers per rule id, never code, paths or repository names. A silenced rule is a rule someone judged not worth acting on in their codebase. That is the closest thing to a false-positive rate a scanner can publish without reading customers’ code, so it is published, rule by rule, worst first.

No data yet

No scan in the last 90 days carried per-rule counts. The table fills as CLI and MCP scans on engine v1.61+ report in; a rule needs 5 scans before it is ranked.

Above 20% the rule is on the retirement list: fixed against the corpus or withdrawn. Below it, the rule stays and the silencing is the team’s call. Your own machine keeps the same ledger locally — gatetest --noise ranks the rules in your repository — and opts out with GATETEST_NO_TELEMETRY=1. Precision on the corpus →