{
  "artifacts": {
    "raw_results": "https://darkoza.com/science/reports/procedural-graph-pilot/results.json",
    "report_html": "https://darkoza.com/science/reports/procedural-graph-pilot/report.html",
    "report_markdown": "https://darkoza.com/science/reports/procedural-graph-pilot/report.md",
    "self_test": "https://darkoza.com/science/reports/procedural-graph-pilot/self-test.json",
    "source": "https://darkoza.com/science/reports/procedural-graph-pilot/procedural_graph_pilot.py",
    "source_sha256": "cb66421c7184b256287781b3320b19bbcd95b0be76dc381616f1b1429bd93962"
  },
  "controls": {
    "candidate_accepted_by_pilot_gate": true,
    "completion_gate_tests": {
      "empty_gates_rejected": true,
      "failed_gate_rejected": true,
      "missing_evidence_reference_rejected": true,
      "passed_evidence_linked_gate_accepted": true
    },
    "model_weights_updated": false,
    "production_graph_updated": false,
    "unseen_context_abstention": {
      "passed": true,
      "prediction": null,
      "reason": "no_context_specific_evidence"
    }
  },
  "dataset": {
    "actions": 13,
    "contexts": [
      "deployment_failure",
      "disk_pressure",
      "network_timeout"
    ],
    "external_calls": 0,
    "held_out_records": 12,
    "model_generated_code_executed": false,
    "train_records": 30
  },
  "evidence_status": "reproducible synthetic engineering pilot",
  "id": "procedural-graph-pilot-2026-10-29",
  "limitations": [
    "Synthetic deterministic workflow fixture only; not real-world agent validation.",
    "Held-out cases use distinct identifiers but replicate known procedure structures; unseen workflow generalization is not established.",
    "The evaluator target set is predeclared and has not been independently reviewed by blind human evaluators.",
    "No model weights or production graph were updated; no production A/B test was run.",
    "No claim of scientific novelty, general intelligence, self-awareness, or production readiness."
  ],
  "method": "Deterministic synthetic workflows; context-agnostic baseline versus context-aware transition graph trained only on training traces; failed transitions penalized; held-out decisions checked against a fixed evaluator specification.",
  "metrics": {
    "absolute_gain_percentage_points": 16.667,
    "baseline_accuracy": 0.833333,
    "baseline_correct": 40,
    "baseline_decisions": 48,
    "candidate_accuracy": 1.0,
    "candidate_correct": 48,
    "candidate_decisions": 48,
    "no_per_context_regression": true,
    "per_context_baseline": {
      "deployment_failure": {
        "accuracy": 0.75,
        "correct": 12,
        "decisions": 16
      },
      "disk_pressure": {
        "accuracy": 0.75,
        "correct": 12,
        "decisions": 16
      },
      "network_timeout": {
        "accuracy": 1.0,
        "correct": 16,
        "decisions": 16
      }
    },
    "per_context_candidate": {
      "deployment_failure": {
        "accuracy": 1.0,
        "correct": 16,
        "decisions": 16
      },
      "disk_pressure": {
        "accuracy": 1.0,
        "correct": 16,
        "decisions": 16
      },
      "network_timeout": {
        "accuracy": 1.0,
        "correct": 16,
        "decisions": 16
      }
    }
  },
  "published_at": "2026-10-29",
  "reproduction": [
    "python3 procedural_graph_pilot.py --self-test",
    "python3 procedural_graph_pilot.py --cycle 99"
  ],
  "schema_version": "1.0",
  "title": "Context-aware procedural graph and evidence-gated completion"
}