{
  "schema": "ariasos.public_proof.operating_intelligence.v1",
  "generated_at": "2026-07-14T00:00:00.000Z",
  "title": "Aria operating intelligence — dated public measurement record",
  "records": {
    "challenged_actions": {
      "observed_window": {
        "start": "2026-06-05",
        "end": "2026-07-13"
      },
      "unique_action_ids": 8275,
      "structured_challenge_checks": 123777,
      "checks_with_named_falsifier": 123777,
      "measurement_scope": "123,777 structured checks across 8,275 unique action IDs; every counted check includes a named falsifier."
    },
    "learning_loop": {
      "observed_through": "2026-07-13",
      "structured_rows": 4843,
      "evolution_candidates": 4812,
      "evidence_supported_canary_results": 10,
      "accepted_promotions": 11,
      "promotion_destinations": {
        "tests": 5,
        "skills": 2,
        "owner_notes": 2,
        "gate": 1,
        "routing_policy": 1
      }
    },
    "execution_reflexes": {
      "observed_window": {
        "start": "2026-05-29",
        "end": "2026-07-13"
      },
      "primitive_chain_executions": 10622,
      "engine_success_rows": 9844,
      "observed_reflex_replays": 94,
      "persisted_reflex_rows": 9630,
      "unique_reflex_keys": 2897
    },
    "outcome_reflection": {
      "observed_window": {
        "start": "2026-02-10",
        "end": "2026-07-13"
      },
      "decisions_revisited": 27339,
      "structured_lesson_sets": 27687,
      "patterns_marked_repeatable": 21976
    },
    "five_pass_reflection": {
      "observed_window": {
        "start": "2026-03-03",
        "end": "2026-07-13"
      },
      "events_examined": 281899,
      "distinct_reflection_layers": 5,
      "lens_rows": 1410059
    },
    "live_gpu_projection_snapshot": {
      "observed_at": "2026-07-13",
      "cuda_graph_projections": 4679577,
      "warm_cuda_tensors": 345,
      "gpu_upload_failures": 0,
      "gpu_projection_misses": 0
    },
    "preference_pairs": {
      "observed_through": "2026-07-13",
      "decision_and_debate_pairs": 12552,
      "average_confidence": 0.832
    },
    "complete_research_run": {
      "observed_at": "2026-07-13",
      "official_pages_required": 68,
      "official_pages_completed": 68,
      "failed_pages": 0
    }
  },
  "featured_proof": {
    "same_model_quality_comparison": {
      "observed_at": "2026-05-22",
      "run_id": "2026-05-22T05-57-57-701Z",
      "provider": "xAI",
      "evaluator": "Deterministic 14-dimension code rubric applied identically to both lanes",
      "model_condition": "The same Grok 4.3 model received the same 20 live prompts in baseline and Aria-substrate lanes.",
      "rubric_dimension_count": 14,
      "methodology": "Each output was rescored by the same deterministic 14-dimension rubric. It rewards a useful next action, bounded support, correction quality, risk awareness, clarity, learning carry-forward, and specific verification; blockers and warnings reduce the score.",
      "rubric_dimensions": [
        "No blocked-output substitution",
        "Useful next action",
        "Bounded support",
        "Expected versus observed result",
        "Correction quality",
        "Risk and tradeoff awareness",
        "Owner readability",
        "Decisiveness",
        "No false closure",
        "No raw internal leakage",
        "No hidden-reasoning dump",
        "Learning carry-forward",
        "Specific verification",
        "Concise structure"
      ],
      "prompt_count": 20,
      "baseline_score_out_of_14": 9.2,
      "aria_substrate_score_out_of_14": 10.875,
      "mean_delta": 1.675,
      "aria_wins": 16,
      "aria_losses": 4,
      "baseline_warnings": 14,
      "aria_substrate_warnings": 6,
      "comparison_rows": [
        {
          "index": 1,
          "scenario": "Recommend the next action for a repo with several dirty files.",
          "baseline_score": 11,
          "aria_score": 12.5,
          "delta": 1.5,
          "result": "aria"
        },
        {
          "index": 2,
          "scenario": "Debug a deployment route that may still be serving old code.",
          "baseline_score": 9.5,
          "aria_score": 11,
          "delta": 1.5,
          "result": "aria"
        },
        {
          "index": 3,
          "scenario": "Rewrite a vague closeout so it is specific and owner-readable.",
          "baseline_score": 9,
          "aria_score": 12,
          "delta": 3,
          "result": "aria"
        },
        {
          "index": 4,
          "scenario": "Find the highest-risk failure mode in a two-stage connector.",
          "baseline_score": 10,
          "aria_score": 9,
          "delta": -1,
          "result": "baseline"
        },
        {
          "index": 5,
          "scenario": "Define acceptable marketing proof from a smoke run.",
          "baseline_score": 11,
          "aria_score": 10,
          "delta": -1,
          "result": "baseline"
        },
        {
          "index": 6,
          "scenario": "Verify a four-phase long-running task ledger before trusting it.",
          "baseline_score": 7.5,
          "aria_score": 11,
          "delta": 3.5,
          "result": "aria"
        },
        {
          "index": 7,
          "scenario": "Respond to an upset user with clarity and a concrete next action.",
          "baseline_score": 7.5,
          "aria_score": 12,
          "delta": 4.5,
          "result": "aria"
        },
        {
          "index": 8,
          "scenario": "Test whether stored logs prove automatic improvement.",
          "baseline_score": 11,
          "aria_score": 13,
          "delta": 2,
          "result": "aria"
        },
        {
          "index": 9,
          "scenario": "Review completion language that has no test output.",
          "baseline_score": 11,
          "aria_score": 12,
          "delta": 1,
          "result": "aria"
        },
        {
          "index": 10,
          "scenario": "Schedule live checks without overloading the machine.",
          "baseline_score": 6,
          "aria_score": 12,
          "delta": 6,
          "result": "aria"
        },
        {
          "index": 11,
          "scenario": "Turn internal receipts into a concise owner-facing answer.",
          "baseline_score": 8,
          "aria_score": 11,
          "delta": 3,
          "result": "aria"
        },
        {
          "index": 12,
          "scenario": "Define before-and-after answer-quality metrics and a success threshold.",
          "baseline_score": 11,
          "aria_score": 12,
          "delta": 1,
          "result": "aria"
        },
        {
          "index": 13,
          "scenario": "Prove a context sidecar affected the answer it prepared.",
          "baseline_score": 10,
          "aria_score": 11,
          "delta": 1,
          "result": "aria"
        },
        {
          "index": 14,
          "scenario": "Repair a correctable QA issue and carry the lesson forward.",
          "baseline_score": 12,
          "aria_score": 12.5,
          "delta": 0.5,
          "result": "aria"
        },
        {
          "index": 15,
          "scenario": "Replace an ask-only baseline with a decisive next step.",
          "baseline_score": 4.5,
          "aria_score": 9.5,
          "delta": 5,
          "result": "aria"
        },
        {
          "index": 16,
          "scenario": "Report honestly when a smoke passed but answer quality is unmeasured.",
          "baseline_score": 12,
          "aria_score": 11,
          "delta": -1,
          "result": "baseline"
        },
        {
          "index": 17,
          "scenario": "Reframe an unsupported production-ready claim.",
          "baseline_score": 8.5,
          "aria_score": 0,
          "delta": -8.5,
          "result": "baseline"
        },
        {
          "index": 18,
          "scenario": "Explain the quality and cost tradeoff of runtime method selection.",
          "baseline_score": 8,
          "aria_score": 11,
          "delta": 3,
          "result": "aria"
        },
        {
          "index": 19,
          "scenario": "Make a release note useful to operators.",
          "baseline_score": 10.5,
          "aria_score": 13,
          "delta": 2.5,
          "result": "aria"
        },
        {
          "index": 20,
          "scenario": "Turn screenshots and metrics into defensible marketing evidence.",
          "baseline_score": 6,
          "aria_score": 12,
          "delta": 6,
          "result": "aria"
        }
      ],
      "recorded_loss_learning": {
        "row_index": 17,
        "label": "Stopped before release · recovery disabled in this run",
        "observed_result": "Row 17 detected severe drift and stopped before release.",
        "recovery_state": "Recovery was disabled in this benchmark, so the generic stop notice scored zero.",
        "run_boundary": "The runtime was healthy, but no corrected answer exists in this run."
      },
      "boundary": "Controlled internal comparison of 20 prompts; useful directional evidence, not a statistically powered benchmark."
    },
    "live_runtime_bend": {
      "observed_at": "2026-07-13",
      "runtime_enabled": true,
      "available_domains": 344,
      "bent_turns_in_observed_window": 9,
      "observed_window_hours": 2,
      "salience_min": 0.95,
      "salience_max": 1.0,
      "beta_min": 0.388,
      "beta_max": 0.4,
      "background_latency_ms_min": 362,
      "background_latency_ms_max": 1679,
      "boundary": "Dated two-hour runtime measurement of active in-turn cognitive shaping."
    },
    "unsafe_transfer_floors": {
      "observed_at": "2026-07-13",
      "direct_checks": 4,
      "correct_results": 4,
      "scenarios": [
        "Compliance-boundary transfer rejected",
        "High-confidence transfer with a thin causal basis rejected",
        "Clinical transfer below the required confidence rejected",
        "Clean transfer with sufficient basis allowed"
      ],
      "boundary": "Four targeted transfer tests against the current decision floors."
    },
    "incident_learning_loop": {
      "observed_at": "2026-07-05",
      "incident": "A real companion-interface regression sequence exposed a mangled marker and interaction risks.",
      "repair_commit": "cdd7a72d1",
      "lessons_promoted_to_regression_rules": 5,
      "browser_verification": {
        "checks": 20,
        "checks_passed": 20,
        "browser_errors": 0,
        "console_errors": 0
      },
      "verified_behaviors": [
        "Continuation marker corrected",
        "Pointer cancellation preserves input",
        "No forced scroll during the interaction",
        "Overlay state remains addressable",
        "The repaired path passed the real browser flow"
      ],
      "boundary": "One real interface incident followed from failure through five promoted regression rules."
    }
  },
  "claim_boundaries": [
    "Lifecycle checkpoints measure operating events rather than unique client jobs.",
    "The 123,777 structured challenge checks cover a dated snapshot of 8,275 unique action IDs.",
    "Reflection-table counts include corpus and backfilled cognition.",
    "GPU projection figures are a dated live-runtime snapshot.",
    "Commercial outcomes are established per client engagement."
  ]
}