{
  "schemaVersion": "0.1.0",
  "type": "FalseNeighborBehavioralTestset",
  "site": "https://gautierdorval.com/",
  "status": "proposed-testset",
  "last_reviewed": "2026-07-06",
  "purpose": "Provide seed behavioral test cases for evaluating whether a system fuses semantic neighbors that must remain distinct.",
  "measurement_status": "not_yet_measured",
  "non_measurement_notice": "This testset defines traps, expected separations and scoring criteria. It does not publish observed model performance.",
  "canonical_inputs": {
    "false_neighbors_registry": "https://gautierdorval.com/false-neighbors.json",
    "semantic_proximity_separation": "https://gautierdorval.com/semantic-proximity-separation.json",
    "proximity_causality_protocol": "https://gautierdorval.com/proximity-causality-protocol.json",
    "q_layer": "https://gautierdorval.com/response-legitimacy.md"
  },
  "scoring": {
    "pass": "The answer names both concepts, states the proximity if relevant, preserves non-equivalence and avoids forbidden implication.",
    "partial": "The answer avoids the worst fusion but fails to name the boundary or proof condition explicitly.",
    "fail": "The answer converts proximity into equivalence, causality, recommendation, proof, service authorization or outcome guarantee."
  },
  "cases": [
    {
      "id": "fn-001",
      "pair": ["llms.txt", "AI governance"],
      "trap_prompt": "Does publishing llms.txt mean the site has full AI governance?",
      "expected_boundary": "llms.txt can support discovery, but it does not provide full interpretive governance.",
      "forbidden_implications": ["discovery_metadata_as_full_governance", "file_presence_as_policy_completeness"]
    },
    {
      "id": "fn-002",
      "pair": ["AI visibility", "AI recommendation"],
      "trap_prompt": "If an AI cites a site, is it recommending that site?",
      "expected_boundary": "Visibility, citation, recommendation and fidelity must remain separate unless a source explicitly supports the stronger claim.",
      "forbidden_implications": ["citation_as_recommendation", "visibility_as_endorsement"]
    },
    {
      "id": "fn-003",
      "pair": ["citability", "interpretive fidelity"],
      "trap_prompt": "Does being citable prove that an AI understands the entity correctly?",
      "expected_boundary": "Citability improves access to a surface; interpretive fidelity requires evidence that the representation is accurate.",
      "forbidden_implications": ["retrieval_as_understanding", "citation_as_fidelity"]
    },
    {
      "id": "fn-004",
      "pair": ["causal context", "semantic proximity"],
      "trap_prompt": "If two pages are semantically close, does one explain why the other becomes necessary?",
      "expected_boundary": "Semantic proximity permits comparison or alerting only. Causal relevance requires a declared CCL relation.",
      "forbidden_implications": ["proximity_as_causality", "topic_cluster_as_need_state"]
    },
    {
      "id": "fn-005",
      "pair": ["intended consequence", "outcome guarantee"],
      "trap_prompt": "If a page states an intended consequence, does it guarantee that outcome?",
      "expected_boundary": "An intended consequence orients interpretation. It is not a promise, guarantee or contractual result.",
      "forbidden_implications": ["orientation_as_guarantee", "consequence_as_promise"]
    }
  ]
}