Developers · Public proof · metrics v2.5.6
Public evals feed — the JSON agents already read.
This page renders the public dispatch.evals feed. Same endpoint MCP clients hit. No API key. Figures are self-reported operational estimates on a rolling 30-day window — not third-party audited.
curl https://steadywrk.app/api/dispatch/analytics/evals?period=rolling_30d
curl -s https://steadywrk.app/api/mcp -H "content-type: application/json" -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"dispatch.evals","arguments":{"period":"rolling_30d"}}}'Goal fulfillment (public estimated feed)
Values below prefer the public JSON when reachable; otherwise they fall back to the drift-gated canonical metrics registry. Dispatch latency is a canonical estimate — never a measured production sample while estimated:true.
Completion rate
estimated90%
Jobs closed successfully on the rolling window.
NTE accuracy
estimated91%
Settled invoice vs quoted not-to-exceed (NTE variance ±9%).
Human override
estimated3%
Agent decisions escalated or reversed by a person.
Redispatch rate
estimated6%
Work orders that needed a second routing pass.
Dispatch latency p50
estimated340 ms (estimated)
Accept → contractor notification (median). Canonical estimate — not a measured sample.
Dispatch latency p95
estimated890 ms (estimated)
Tail latency for routing + outreach. Canonical estimate — not a measured sample.
Trust signals & posture
- Production posture (30d)
- Not published
- Network posture
- Contractor network in onboarding
- Coverage posture
- Multi-state (on request)
- Security posture
- see /security
Operational volume fields are measured integers or unpublished. STEADYWRK does not invent a posture string when the feed has no count.
This public evals snapshot is registry record #evals-rolling-30d.