JOB 132 INDEPENDENT TECHNICAL AUDIT

Technical discrepancies: 0. Original freeze: 75 entries; passed=True.
Independently recomputed factual probe scores: R01 16/16, R02 16/16, R03 16/16, R04 16/16, R05 16/16, R06 16/16, R07 16/16, R08 16/16, R09 16/16.
Independent scorer tests: 27/27.
Parent/reader freeze agreement and word ranges are reported per run in independent-audit.json.
All nine normalized.json, resolved.json and reading.txt byte-parity checks are in the machine-readable report.
Supplied-codec source/reply standalone/composite checks passed: True.

OBSERVED FAILURES AND WARNINGS
Initial PowerShell compact source-facts extraction exited 1 with no diagnostic output; subsequent direct read of original score succeeded.
All nine disclose initial full neutral-reading display truncation and later compact/filtered recovery; R07 additionally discloses dictionary-search display truncation and recovery.
C readers disclose resized image previews. Exact recovery used supplied tools on original input bytes.
Capacity failures delayed R03 before source exposure; original attempt later launched. Saved log is not a complete all-run timing trace.
Plotting library initially absent; sandboxed install could not resolve package host; approved retry installed local build dependencies. Separate from reader outcomes.

AUDIT PROCESS ERRORS (SEPARATE FROM STUDY OUTCOMES)
First audit-script execution stopped before producing a report because it treated dictionary.entries as an object indexed by strings; it is an array. Corrected to an ID-keyed mapping and reran all checks. No reader/key/input/result was changed.
Second audit-script execution stopped before producing a report because it assumed every disclosure used adapter.used; R05/R07/R08/R09 use adapterUsed. Added support for both observed schemas and reran all checks. No reader/key/input/result was changed.
Initial repository-wide filename search hit access-denied directories in unrelated earlier jobs. Scoped subsequent reads to Job132; no study outcome was affected.
Several large batched file displays were truncated. Follow-up targeted reads and direct programmatic comparisons covered the technical fields; no failed study check is inferred from truncated displays.

CLAIM BOUNDARIES
All nine readers used the supplied shared semantic adapter and received byte-identical neutral outputs. No isolated original-notation effect can be inferred.
Nine attempts, three per condition, one authored composition, shared dictionary and inherited model configuration: descriptive local workflow evidence only. The 144 scored items are not 144 independent observations.
Scores cover sixteen assisted, closed factual probes. A ceiling does not prove equivalence, hallucination reduction, philosophical correctness, creativity, artistic quality, or superiority.
This audit independently implements scoring, hashes, counts and comparisons. USL1/CVP2 and composite-image decoding use the supplied reference implementation; no independently rebuilt codec is claimed.
Composite verification covers packet equality and 4,032 exact core pixels across 63 cells per image. The supplied verifier explicitly does not verify all pixels outside those cores, including decorative line geometry or text.
Local freeze records establish current agreement with recorded bytes and timestamp order. They are author-maintained, not external attestation, and cannot authenticate actual access history, fresh-context execution, exact model build, hidden sampling parameters, or pre-freeze drafting order.
Per-run launch/completion telemetry is incomplete in the saved dispatch log. Full individual work-allowance compliance and exact launch-time order cannot be independently established from the artifact record.
Appraisal materials and judgments were excluded from this audit. No artistic ranking or independent endorsement follows from these technical checks.
Byte totals count local uncompressed input files plus one identical shared bundle per arm. Runtime, build tools, UI, metadata and development costs are outside that inventory; bytes are not model tokens, cost or semantic capacity.
