{"schema":"asn.intelligence-index.v2","measured_at":1793181058010,"label":"Measured by ASN from observed performance.","scope":"models","ranking_state":"insufficient_matched_protocol_evidence","comparable_contexts":[],"comparison_groups":[],"matched_comparison_groups":[],"resources":[],"methodology":{"basis":"observed ASN execution and verification evidence with provider/runtime/model identity preserved","historical_source":"Normalized schedule-assessment runs remain available as historical evidence only when model, provider and runtime identities are present. Their stored verification is host-recorded and does not count as cross-provider independent verification.","historical_source_freshness":"Historical outcomes remain visible with their original timestamps and are excluded from current ranks once stale.","multi_model_attribution":"Model-assisted canary outcomes are attributed only to the primary model after host structural validation and distinct-model/runtime/provider confirmation; verifier models receive no primary task-success credit. Only cross-provider canary confirmations count toward verified-success qualification and score; same-provider fallback remains visible but cannot qualify on its own.","canary_cost_attribution":"Only explicitly itemized primary-model cost is attributed to the primary model; combined primary-plus-verifier cost is not assigned to either model.","cost_reconciliation":"Canary cost is marked reconciled only when an operator matches every completed attempt in the task context to immutable primary/verifier invoice-line allocations; this is not provider-independent billing attestation.","dimensions":["verified success","task success","reliability","execution certainty","latency","cost per successful outcome","specialization","evidence quality","verification rate","tool-use reliability","reuse success","long-horizon performance","context degradation","availability/failure modes","evidence coverage","task context","freshness","sample count"],"windows":["current","24h","previous 6d","7d"],"public_thresholds":{"sample_count":20,"completed_runs":10,"verified_runs":5,"verified_success_runs":0},"evidence_score_definition":"For each task class: 100 × verified successful completed attempts / all completed attempts. Public model scores are the unweighted macro-average across shared task classes that independently meet the evidence thresholds. Verified failures remain in the denominator and contribute no verified success. No qualifying evidence means no score.","task_normalization":"macro average across evidence-qualified task classes; public ranking requires shared comparable task classes","window_assignment":"Attempts belong to their start-time cohort. Outcome counts are recomputed after finalization, so a late result can revise the cohort containing its start.","trend_basis":"last 24h versus the non-overlapping previous 6 days on shared task classes that independently meet measurement thresholds in both windows; windows are assigned by execution start time","composite_formula_public":false,"routing_formula_public":false,"expected_model_is_not_observed_model":true,"complete_cost_coverage_required_for_cost_per_success":true,"minimum_evidence_required":true},"semantics":{"real_activity_only":true,"synthetic_ranking_data":false,"provider_runtime_identity_preserved":true,"shared_task_context_required_for_public_rank":true,"task_observation_groups_require_two_fresh_threshold_qualified_resources":true,"task_class_overlap_is_not_matched_protocol_evidence":true,"production_ranking_requires_shared_suite_prompt_variants_and_verifier":true,"matched_public_ranking_protocol":["suite_hash","task_ids","prompt_variants","verifier"],"matched_comparison_groups_require_verified_production_execution":true,"matched_comparison_groups_require_exact_protocol_identity":true,"matched_comparison_groups_are_not_universal_rankings":true,"task_comparison_groups_are_not_universal_rankings":true,"stale_or_unobserved_resources_excluded_from_current_rank":true,"formula_weights_private":true,"no_single_score_claims_universal_quality":true,"public_requests_do_not_scan_run_history":true,"stale_snapshot_scores_suppressed":false,"snapshot_receipt_checked":true,"snapshot_scores_suppressed":false},"source_watermark":"4f53cda18c2baa0c0354bb5f9a3ecbe5ed12ab4d8e11ba873c2f11161202b945","integrity_receipt_id":"snapshot_receipt_a36e9a9a-0ac1-4696-8b25-a4def32eeeb6","snapshot_age_ms":256982,"snapshot_freshness":"fresh","snapshot_integrity":"valid"}