Developers · API reference

Evaluation API documentation

A guide to reading the dispatch.evals response. Use field-level source labels and definitions to understand what a response can support. This page does not publish a performance report.

Public read-only request

curl 'https://steadywrk.app/api/dispatch/analytics/evals?period=rolling_30d'

MCP tools/call parameters

{ "name": "dispatch.evals", "arguments": { "period": "rolling_30d" } }

Evaluation API JSON MCP integration guide Developer hub

Read the response by field

schema_version · period: Identify the response schema and requested reporting period. These labels describe the response format and reporting horizon.
estimated · provenance.by_field: The estimated summary flag does not describe every field. Read provenance.by_field for the source class attached to each reported figure.
goal_fulfillment: Contains completion_rate, nte_accuracy, redispatch_rate and human_override_rate. Interpret each field using its own provenance and methodology definition.
operational.latency_source · operational.pricing_latency_source: Keep dispatch timing separate from pricing API response timing. Each has its own source label; neither label establishes the source of the other.
methodology: Read definitions, exclusions and window bounds together. The window label does not establish that every field is a measurement from that period.
trust_signals.build_hash: Identifies the serving build. A build identifier describes software revision, not the quality or freshness of a measurement.

Source labels and limits

canonical_estimate: A canonical estimate from the registry. It is not a measurement of the requested reporting period.
live_telemetry: The response labels this field as derived from observations. Check the field definition and any associated freshness timestamp before interpreting it.
unavailable: No qualifying reading is supplied for the field. A null value means unavailable, not zero.

The response is self-reported, not an independent audit or a signed snapshot. A successful API response is not a service guarantee. Read methodology and provenance.by_field together; a summary flag cannot replace them.

Requests are rate limited. Handle retry responses and the returned rate-limit headers. A cached response may not reflect the latest available observations.

Read the response by field

schema_version · period

Identify the response schema and requested reporting period. These labels describe the response format and reporting horizon.

estimated · provenance.by_field

The estimated summary flag does not describe every field. Read provenance.by_field for the source class attached to each reported figure.

goal_fulfillment

Contains completion_rate, nte_accuracy, redispatch_rate and human_override_rate. Interpret each field using its own provenance and methodology definition.

operational.latency_source · operational.pricing_latency_source

Keep dispatch timing separate from pricing API response timing. Each has its own source label; neither label establishes the source of the other.

methodology

Read definitions, exclusions and window bounds together. The window label does not establish that every field is a measurement from that period.

trust_signals.build_hash

Identifies the serving build. A build identifier describes software revision, not the quality or freshness of a measurement.

Source labels and limits

canonical_estimate

A canonical estimate from the registry. It is not a measurement of the requested reporting period.

live_telemetry

The response labels this field as derived from observations. Check the field definition and any associated freshness timestamp before interpreting it.

unavailable

No qualifying reading is supplied for the field. A null value means unavailable, not zero.

Requests are rate limited. Handle retry responses and the returned rate-limit headers. A cached response may not reflect the latest available observations.