CVP2-E01 — verified receiver results
Prepared 8 September 2026

Both returned USL1 objects are byte-identical to the frozen sender:
100,551 bytes
SHA-256 47f19645d26ec974b89dab7ecb920b981477655e01bc104ecd98aea15da91ae5

Both exact expanded sources:
98,470 bytes
SHA-256 1f2b83affee5fe09aaf774d620d3cf4e396f093f1eee8372e7d0c9b6e3649a9c

Input PNG: 125,718 bytes, 896×896, 4×4 physical pixels per data cell
SHA-256 9a9d2abd38eaf45e184dccfaeabe2c0154ed48886b8f9e2e9c60d1e45a8233d5
24,358 distinct data colours; full RGB24 alphabet permitted.

Carried context: 106 dictionary records, 11 concept tokens, 12 literal tokens,
zero lexical tokens. Twelve top-level clauses and fifteen clause nodes.
The reply is a separate field. Dictionary head:
1cb4608b9ae8455b7506cee71590f16b7384f4b99284fbc4975c329b0592a8cb

METHODS
ChatGPT disclosed an earlier English interpretation visible in its conversation.
It is not a blinded semantic replication. Its new reported implementation
reruns successfully with standard-library Python: all 74 delivered tests and
four additional lower-stage rejection probes pass. Fourteen of fifteen final
result files reproduce exactly on Windows; the checksum manifest alone uses
CRLF instead of LF, with identical hash/name entries.

The organizer explicitly confirmed Claude's completely fresh conversation,
given only the PNG and instructions, with no prior USL context. Its returned
data are correct. Its shorter prose preserves the central invitation and most
key distinctions; complete qualifications remain in the exact recovered source.

CLAUDE CODE AUDIT
The unmodified exported script stops at a missing reconstructed.txt file.
A documented write of its own newly expanded bytes permits source-stage
reproduction: 108 true logged booleans and 15,040 zero RS syndromes on the intact
image. The delivered script stops before its reported final concept-token,
alias and selected-pin joins and often logs failed checks without stopping.
Our separate audit verifies the actual returned bindings and context.

Thirteen component probes: 6 invalid inputs unflagged, 4 diagnosed but execution
continues, 3 integer controls raise. Examples include unchecked trailing zlib
data, an unjoined shifted concept-token span, and continued payload export
after an RS parity failure. These are scoped stage probes, not altered images
passing the published original-file hash. Counts use different units and are
not comparative model scores.

CONCLUSION
Exact RGB24 recovery succeeded for the frozen message. Claude adds one
organizer-reported fresh-context interpretation. ChatGPT provides a complete
reproducible receiver for this intact E01 recipe. Claude's correct output is
stronger evidence than the completeness of its particular exported validator.

This is one code-assisted expression, not a population success rate, optical
test, English-independent grammar, or demonstration of identical judgment.
The historical dictionary and grammar travel inside the packet. Integrity
hashes do not authenticate authorship or certify semantic truth.

The research archive includes original returns, frozen inputs and sender key,
both readings, local audit additions, probes and reproduction instructions.
For a fresh receiving exercise use only the two-file receiver kit.
