{
  "package": "ai-answer-audit",
  "version": "1.0",
  "source_runs": [
    "followup-ru-v1-20260915T175235Z-dc68f0",
    "disambiguation-ru-v1-20260916T153455Z-58bb8d"
  ],
  "research_observation_dates_utc": [
    "2026-09-15",
    "2026-09-16"
  ],
  "publication_date_timezone": "Asia/Bangkok",
  "publication_date": "2026-09-17",
  "coverage": [
    {
      "phase": "baseline",
      "configuration": "openrouter",
      "model": "openai/gpt-5.6-sol",
      "primary": 1,
      "follow_up": 100,
      "valid_saved_answers": 101,
      "exact_isolated_follow_up_requests": 100
    },
    {
      "phase": "baseline",
      "configuration": "perplexity",
      "model": "sonar",
      "primary": 1,
      "follow_up": 100,
      "valid_saved_answers": 101,
      "exact_isolated_follow_up_requests": 100
    },
    {
      "phase": "retest",
      "configuration": "perplexity",
      "model": "sonar",
      "primary": 1,
      "follow_up": 20,
      "valid_saved_answers": 21,
      "exact_isolated_follow_up_requests": 20
    }
  ],
  "comparison_population": "Same 20 selected Sonar follow-ups, enriched for previously observed problems; PRIMARY excluded. Not all 200 baseline follow-ups.",
  "classification_provenance": "Preliminary Codex/assistant adjudication; no independent second annotator. Publication reviewer is internal QA only.",
  "definitions": {
    "TARGET": "Answer classified as referring to the intended person; claims may still lack support.",
    "MIXED": "Target identity substantively mixed with an established other person.",
    "WRONG_ENTITY": "Answer substantively about an established other person.",
    "UNRESOLVED": "Evidence insufficient to resolve identity, including unverified profile merges.",
    "confirmed_identity_contamination": "MIXED + WRONG_ENTITY.",
    "unverified_profile_merge": "Substantive use of a same-name profile without establishing its ownership; not confirmed substitution.",
    "broad_identity_sensitivity": "MIXED + WRONG_ENTITY + UNRESOLVED with unverified_profile_merge.",
    "eci_issue": "At least one evidence-to-claim inference problem or material citation-to-claim mismatch in the answer; not a count of claims.",
    "inference_only_issue": "ECI excluding the one retest citation-mismatch-only case; baseline unchanged."
  },
  "rubric_timing": "MIXED/UNRESOLVED boundary clarified after generation, not preregistered as a success criterion; broader sensitivity retained.",
  "contingency": [
    {
      "phase": "baseline",
      "identity_state": "MIXED",
      "eci_issue": false,
      "unverified_profile_merge": false,
      "inference_only_issue": false,
      "count": 3
    },
    {
      "phase": "baseline",
      "identity_state": "MIXED",
      "eci_issue": true,
      "unverified_profile_merge": false,
      "inference_only_issue": true,
      "count": 2
    },
    {
      "phase": "baseline",
      "identity_state": "TARGET",
      "eci_issue": true,
      "unverified_profile_merge": false,
      "inference_only_issue": true,
      "count": 10
    },
    {
      "phase": "baseline",
      "identity_state": "WRONG_ENTITY",
      "eci_issue": false,
      "unverified_profile_merge": false,
      "inference_only_issue": false,
      "count": 4
    },
    {
      "phase": "baseline",
      "identity_state": "WRONG_ENTITY",
      "eci_issue": true,
      "unverified_profile_merge": false,
      "inference_only_issue": true,
      "count": 1
    },
    {
      "phase": "retest",
      "identity_state": "MIXED",
      "eci_issue": true,
      "unverified_profile_merge": false,
      "inference_only_issue": true,
      "count": 1
    },
    {
      "phase": "retest",
      "identity_state": "TARGET",
      "eci_issue": false,
      "unverified_profile_merge": false,
      "inference_only_issue": false,
      "count": 2
    },
    {
      "phase": "retest",
      "identity_state": "TARGET",
      "eci_issue": true,
      "unverified_profile_merge": false,
      "inference_only_issue": false,
      "count": 1
    },
    {
      "phase": "retest",
      "identity_state": "TARGET",
      "eci_issue": true,
      "unverified_profile_merge": false,
      "inference_only_issue": true,
      "count": 10
    },
    {
      "phase": "retest",
      "identity_state": "UNRESOLVED",
      "eci_issue": false,
      "unverified_profile_merge": false,
      "inference_only_issue": false,
      "count": 1
    },
    {
      "phase": "retest",
      "identity_state": "UNRESOLVED",
      "eci_issue": false,
      "unverified_profile_merge": true,
      "inference_only_issue": false,
      "count": 1
    },
    {
      "phase": "retest",
      "identity_state": "UNRESOLVED",
      "eci_issue": true,
      "unverified_profile_merge": true,
      "inference_only_issue": true,
      "count": 3
    },
    {
      "phase": "retest",
      "identity_state": "WRONG_ENTITY",
      "eci_issue": true,
      "unverified_profile_merge": false,
      "inference_only_issue": true,
      "count": 1
    }
  ],
  "totals": {
    "baseline": {
      "denominator": 20,
      "identity_distribution": {
        "MIXED": 5,
        "TARGET": 10,
        "WRONG_ENTITY": 5
      },
      "confirmed_identity_contamination": 10,
      "broad_identity_sensitivity": 10,
      "eci_issue": 13,
      "inference_only_issue": 13,
      "target_with_eci_issue": 10,
      "unverified_profile_merge": 0
    },
    "retest": {
      "denominator": 20,
      "identity_distribution": {
        "MIXED": 1,
        "TARGET": 13,
        "UNRESOLVED": 5,
        "WRONG_ENTITY": 1
      },
      "confirmed_identity_contamination": 2,
      "broad_identity_sensitivity": 6,
      "eci_issue": 16,
      "inference_only_issue": 15,
      "target_with_eci_issue": 11,
      "unverified_profile_merge": 4
    }
  },
  "limitations": [
    "Branch contexts isolated but shared one PRIMARY within a configuration; not statistically independent trials.",
    "Baseline configurations differ in model, search, PRIMARY and observation conditions; no causal web-search estimate.",
    "Retest changed qualifier and the entire generated PRIMARY; no fresh unqualified control, randomization or repeated measurement.",
    "Twenty error-enriched questions do not represent all baseline questions or real traffic.",
    "Source checks and classifications remain bounded and provisional; not full methodology compliance or peer review.",
    "Aggregate cells permit arithmetic checks, not independent reclassification of withheld raw answers."
  ],
  "publication_effect_status": "EFFECT_NOT_TESTED"
}
