| key | value | band | kind | how it is derived |
|---|---|---|---|---|
| Deployment frequency | 50 over 1d (185/wk) | ELITE | MEASURED | count of DONE frames in ws_sync.jrnl (each = a delivered capability) |
| Lead time for changes | median 1h (30 matched pairs) | ELITE | MEASURED | DONE_epoch - most-recent prior KICKOFF of the same workstream |
| Change failure rate | 42% (21 incident-class open debts) | LOW | PROXY | deploy-failure open debts (regression/rollback/revert/outage/brick) / deploys (declared coarse-upper-bound heuristic; broad coordination/stale debts EXCLUDED) |
| Time to restore (MTTR) | — | — | PENDING | filed→eaten span needs the debthist- join (F740b); NOT asserted rather than faked |
| axis | Nishi | Jellyfish / LinearB / Swarmia / DX |
|---|---|---|
| instrumentation | ZERO manual + ZERO git-poll: derived from the native delivery event stream | poll git/CI/Jira webhooks; batch ETL |
| honesty | each key tagged MEASURED / PROXY / PENDING; a key it cannot ground is shown PENDING, never faked | single confident numbers |
| agents as workers | autonomous zero-Claude DONE frames counted as first-class deliveries | humans-only |
| target | why |
|---|---|
| dora.dev 2024 State of DevOps | the canonical thresholds + the AI-impact findings to re-band against |
| DX Core 4 (getdx.com) | unify DORA+SPACE+DevEx; the 4-dim model to extend these keys with effectiveness/quality |
| debthist- join (F740b) | filed→eaten timestamps to turn MTTR from PENDING into MEASURED |