Runtime & Integrations · Evaluation Core
Scorecard
The primary summary artifact.
The
Scorecard is the machine-and-human summary every report renders from:
verdict, metrics, authority, baseline_state,
and diagnostics. Report wording must be backed by these typed fields — never invented.
The Scorecard is the contract between the core and everything that displays a result. Because a report can only render what is here, it can never claim more authority than the run holds. The Read a Scorecard guide walks the fields for end users; this page is the contract.
Fields
| Field | Holds |
|---|---|
verdict | The VerdictPayload: verdict, ci_should_fail, gate lists, reasons. |
metrics | Per-metric AggregatedMetric: value, included_count, status_counts, aggregation. |
authority | Per-metric ResolvedAuthority: can_gate, blocked, level, typed reasons. |
baseline_state | The comparability state for this run. |
diagnostics | Run-level structured diagnostics. |
JSON is primary
scorecard.json is the source of truth; report.md, the terminal
summary, and CI annotations are renderings of it. None of them recompute the verdict.
No per-call identity here
The Scorecard is the compact aggregate — it carries no per-score subject identity. That lives on
the per-score records in RunRecord, which is what
/evalglass view --by-call groups by.