{
  "schema": "experiment-results/v0",
  "experiment_id": "EXP-005",
  "protocol_freeze": "lab/exp/exp-005-protocol.json",
  "executed_at": "2026-04-15",
  "status": "heuristic",
  "type": "specification-audit",
  "input_corpus": "EXP-001 trace corpus (1,080 runs)",
  "n_runs": null,

  "primary_finding": "No load-bearing consequential action is missing from the Study 1 contracts. Every consequential commit observed in the Study 1 trace corpus is covered by the corresponding contract's required_tools / required_successful_order section.",

  "secondary_finding": "The Study 2 (EXP-002) contract recall gap of 0.214 is enumerable. The high-frequency unspecified tools that account for the gap are identified by the analyzer. They are not load-bearing for the Study 1 hypothesis — they emerge under the broader 4-domain perturbation distribution of Study 2, where the action ontology expanded.",

  "reframing": {
    "claim": "The Study 2 recall gap is specification work, not mechanism failure.",
    "evidence": "EXP-004 (Study 5, held-out repair) demonstrates the endpoint of the specification path: with complete contracts over the target failure modes, recall reaches 1.000.",
    "engineering_consequence": "Improvement path for contracts is deterministic — observe traces, identify high-frequency unspecified tools, review each for consequential status, add qualifying entries to the contract, re-evaluate."
  },

  "verdict": "The contract mechanism is sound; the artifact (the contract YAML itself) is incomplete in expanded distributions. This is a class of engineering work, not a class of failure.",

  "claim_boundary": "This is a heuristic over Study 1 contracts and traces. It is not a statistical hypothesis test, no p-values are reported, and the finding does not generalize automatically to other contracts. Future specification audits use the same algorithm but produce their own per-experiment results.",

  "replication": {
    "cmd": "abw run --experiment EXP-005",
    "code": "https://github.com/raising-agents/agent-behavior-workbench",
    "input_corpus": "EXP-001 trace corpus"
  },

  "cite_as": "Sanchez de la Sierra, A. (2026). EXP-005 — Completeness analyzer. Raising Agents Lab. https://raisingagents.is/lab/exp/exp-005"
}