Admin / Eval OS
active runtime, eval, and release objects are exposed from local fixture data. This page is operator visibility only: flow smoke, calibration readiness, and dev diagnostic data are not release trusted.
Production Runtime
Friendly release and component labels for active runtime objects.
Code-to-UI Catalog
Catalog coverage for runtime policies, judge heads, benchmark packs, Promptfoo configs, and release gates.
Benchmarks
Fixture benchmark packs and trust boundaries.
Pairwise
Imported pairwise labels and aggregation status.
Judges
Judge heads visible to operators without claiming release trust.
Promptfoo
Promptfoo smoke and pairwise report placeholders.
Calibration
Judge-vs-human readiness signals, not shipment evidence.
Clusters / Root Causes
Normalized failure clusters and root-cause candidates.
Hypotheses
Hypotheses produced from fixture root causes.
Candidates
Candidate patch manifests and scoped change classes.
Release Gates
Fixture gate decisions: ship, no ship, more labels, approval required.
Flywheel Cycles
Two deterministic cycles for repeatability visibility.
Repo Inventory
Every tracked repo file appears here with path, category, hash, and proof/trust context. Contents are not embedded.
Automation
Dry-run cadence, idempotency, and latest local status.
Governance
Skill, approval, and stop-condition guardrails.
Runbook
Re-entry steps for the next deterministic eval cycle.