Contracts Reference

Enums

Every closed value set, in one place.

These are the fixed enumerations the contracts use. They are closed sets — a value outside them fails parsing rather than being silently coerced. Other pages link here instead of redefining values.

Measurement

`ScoreStatus`

Value	Meaning
`scored`	A meaningful value was produced.
`blocked`	A required metric could not honestly run.
`non_evaluable`	The example lacks evidence for this metric.
`skipped`	The metric was intentionally not run.
`error`	The evaluator or evidence parser failed unexpectedly.

`Validity`

Value	Meaning
`valid`	The value is a sound measurement.
`invalid`	A value was produced but cannot be trusted.
`not_measured`	No measurement was attempted.
`not_applicable`	The metric does not apply to this example.

`Severity`

Value	Meaning
`info` · `warning` · `error`	Diagnostic severity — describes the finding, not the run outcome.

Metrics

Enum	Values
`Lens`	`reference`, `non_reference`
`ScoreType`	`binary`, `continuous`
`Direction`	`higher_is_better`, `lower_is_better`
`Aggregation`	`mean`, `median`, `min`, `max`, `rate`, `none`

Units & evidence

Enum	Values
`UnitKind`	`call`, `step`, `trajectory`, `session`
`DataPolicy`	`permitted`, `redacted`, `forbidden`, `missing`, `unknown`
`JudgeEvidenceStatus`	`ok`, `timeout`, `provider_error`, `malformed`, `missing` (only `ok` may carry a value)
`TraceFormat`	`local`, `opentelemetry`, `openinference`

Authority

Enum	Values
`DatasetStatus`	`proposed`, `validated`, `retired`
`MetricStatus`	`draft`, `informational`, `calibrating`, `gating`, `retired`
`ThresholdApproval`	`proposed`, `approved`
`JudgeCalibration`	`uncalibrated`, `calibrating`, `calibrated`, `drifted`, `retired`
`AuthorityLevel`	`none`, `informational`, `gating`

Outcome

Enum	Values
`Verdict`	`informational`, `pass`, `fail`, `blocked`
`BaselineState`	`comparable`, `not_comparable`, `missing_baseline`, `comparison_not_requested`
`ExitClass`	`zero` (0), `nonzero_fail` (1), `nonzero_blocked` (1), `infrastructure_error` (2)

Extension lanes

Enum	Values
`LaneStatus`	`ran`, `skipped`, `blocked` (never a score or verdict)
`LanePort`	`trace_source`, `score_sink`, `judge_model`, `task_runner`

Confidence & decisions

Enum	Values
`IntervalMethod`	`wilson` (proportions), `student_t` (means); `none` is a rejected sentinel. See Estimate.
`DecisionStatistic`	`point`, `lower_confidence_bound`, `upper_confidence_bound`; `null` resolves to the conservative bound for the direction. See Decision Policy.
`GrantStatus`	`matched`, `mismatched`, `expired`, `missing`. See Judge Integrity.

Next steps

Trust Model

Measurement State

How ScoreStatus × Validity behave.

Reference

Vocabulary

The terms behind the enums.