Performance

CI-enforced performance thresholds from the quality gate (D24).

Deterministic quality gate

39 metrics, all code-computed. No LLM-as-judge.

View history →
MetricThresholdSeverityWeight
Line coverage≥ 65WARN4
Branch coverage≥ 52WARN3
Real-data E2E coverage≥ 25WARN3
Unit test failures≤ 0WARN6
E2E test failures≤ 0WARN6
Unapproved visual diffs≤ 0WARN3
tsc errors≤ 0WARN6
Code duplication %≤ 5WARN2
Unused public exports≤ 25WARN1
npm audit critical≤ 0BLOCK5
npm audit high≤ 0BLOCK4
npm audit medium≤ 5WARN1
Gitleaks findings≤ 19BLOCK5
Dependency majors behind≤ 3WARN1
Tab nav p95 (ms)≤ 500WARN2
Copilot first token p95 (ms)≤ 1500WARN2
Copilot full response p95 (ms)≤ 5000WARN2
API read p95 (ms)≤ 200WARN2
Max bundle size (KB)≤ 200WARN2
CLS p95≤ 0.1WARN1
LCP p95 (ms)≤ 2500WARN1
axe critical violations≤ 0WARN4
axe serious violations≤ 0WARN3
axe moderate violations≤ 3WARN1
ADR coverage (commit trailer)≥ 1WARN2
SLO: dashboard render p95 (ms)≤ 500WARN2
SLO: copilot first token p95 (ms)≤ 1000WARN2
SLO: audio backfill latency p95 (min)≤ 30WARN1
SLO: ARX overview availability % (7d)≥ 99.9WARN2
SLO: R2 ingest e2e p95 (min)≤ 120WARN1
Brain eval recall@10 (overall)≥ 0.75WARN4
Brain eval recall@10 (audio)≥ 0.65WARN1
Brain eval recall@10 (email)≥ 0.65WARN1
Brain eval recall@10 (slack)≥ 0.65WARN1
Brain eval recall@10 (SEC filings)≥ 0.65WARN1
Brain eval recall@10 (manual-upload)≥ 0.65WARN1
Brain eval recall@10 (multi-source)≥ 0.65WARN1
Adversarial review cost per session ($)≤ 10WARN1
Adversarial framework heartbeat missing count (7d)≤ 0WARN3

BLOCK metrics fail the gate and prevent a push. WARN metrics are recorded in LOGS/quality/*.jsonl but do not block. Overrides require an ADR plus a Quality-Override: ADR-XXXX trailer; auto-expires after 14 days.