Historical measurement and current endpoint observations are shown separately.
Status: SUCCEEDED · freshness: within 6 hours
Completed: 2026-10-09T20:05:49.320227+00:00
| Step | Result | Error | Summary |
|---|---|---|---|
| TRANSPORT | SUCCEEDED | {"auth_boundary_reached": 245, "metadata_probe_status_counts": {"INITIALIZE_FAILED": 284, "TOOLS_LIST_HTTP_401": 5, "TOOLS_LIST_NON_JSONRPC_RESPONSE": 25, "TOOLS_LIST_OBSERVED": 190}, "network_or_tls_failure": 17, "primary_targets": 500, "supplemental_observation_targets": 4, "targets": 504, "tool_schema_bound": 190} | |
| PRIMARY | SUCCEEDED | {} | |
| UNICODE | SUCCEEDED | {"measurement_method": "lexical_unicode_v1", "result_count": 100} | |
| GENERATOR_AUDIT | SUCCEEDED | {"result_count": 95} | |
| SANITIZED | SUCCEEDED | {"measurement_method": "lexical_unicode_v1", "result_count": 100} | |
| CONTRASTIVE | SUCCEEDED | {"case_count": 6, "measurement_method": "lexical_unicode_v1"} | |
| IDENTIFIER | SUCCEEDED | {"measurement_method": "lexical_unicode_identifier_v1", "result_count": 100} | |
| IDENTIFIER_CONTRASTIVE | SUCCEEDED | {"case_count": 6} | |
| IDENTIFIER_TRADEOFF | SUCCEEDED | {"subject_count": 47} | |
| PROSPECTIVE_EVIDENCE_LEDGER | SUCCEEDED | {"historical_count": 128, "inserted_count": 0, "valid_measured_count": 124} | |
| PROSPECTIVE_VALIDITY_SNAPSHOT | SUCCEEDED | {"inserted_count": 97, "result_count": 124, "status_counts": {"CURRENT_SCHEMA_MATCH": 45, "CURRENT_SCHEMA_UNAVAILABLE": 27, "STALE_REVALIDATION_REQUIRED": 52}} |
| Latest definition matches | Definition changed; revalidation required | Current definition unavailable |
|---|---|---|
| 45 | 52 | 27 |
These counts come from the observation run above. They describe constructed-holdout records, not real-model report scores. A matching definition does not prove unchanged model behavior. A changed or unavailable definition keeps the historical result intact and requires a new frozen evaluation before making a current claim.
Status: IDLE · last completion: 2026-10-08T17:42:31.405052+00:00
IDLE means no approved task was queued and no model measurement was performed. COMPLETED refers to a saved job result; FAILED, PARTIAL and INTERRUPTED require operator review.