Sample artifact · Evaluate
Release evaluation scorecard
Illustrative structure for a go / no-go release decision. Numbers are examples for format only — not client results.
| Dimension | Gate | Example readout |
|---|---|---|
| Groundedness / task quality | Pass threshold on golden set | 92% pass · target ≥ 90% |
| Safety / policy | Zero critical policy failures | 0 critical · 2 review notes |
| Regression | No material drop vs last release | -1.2% · within budget |
| Latency p95 | Within agreed SLO | 2.4s · target ≤ 3.0s |
| Cost per successful task | Within budget band | $0.18 · band $0.10–$0.25 |
| Access / leakage checks | No unauthorized retrieval | Pass |
Decision block (example)
Recommendation: Ship to limited production with monitoring on latency and cost. Block full scale until policy review notes are closed.
Owners: Product release owner · Security reviewer · Platform engineer