E006 PUBLIC REPRODUCIBILITY / AUDIT PACKAGE - v1.0 - 2026-10-06

This package is a publication derivative of completed E006. It is not the
private master freeze and does not reopen or extend E006.

The experiment/ tree contains D0 and its historical pilot, D1 development
and regression, Generator B and its seeds/certificates, the rejected bridge
analysis, author clarification, final preregistration/evaluator qualification,
all 60 raw E006 diagnoses, saved scores and original execution audits.
Historical NOT EXECUTED documents describe earlier phases, not the final result.
Historical failed scoring drafts remain identified as drafts.

PUBLICATION-MANIFEST.json is authoritative for the files in THIS package.
Its source_sha256 binds the historical original bytes; sha256 binds the public
copy. source_byte_identical is false where machine locations were redacted.
Original MANIFEST/SHA256SUMS files within experiment/ are historical evidence:
their bindings concern original bytes and are NOT replacement manifests for
redacted copies. Redacted locations are not runnable routing configuration.

Raw E006 inputs, stdout/stderr, parsed diagnostics and scoring records are
preserved byte-for-byte. No case, label, score, witness or semantic fact is
redacted. Source code pins for D0, D1 and evaluator are preserved.
All synthetic hidden worlds are released after the test; they contain no
personal or operational secrets. A future blind test requires a new corpus.

Read-only verification (Python 3.9+, standard library only):
  python3 -B verify-publication.py

Optional independent publication check on COPIES of stored outputs:
  python3 -B verify-publication.py --replay-saved
This recomputes formal scores offline without invoking D0/D1, SEARCH, network,
memory writes or the historical harness. It is a new publication verification,
not a second historical E006 execution or historical evaluator invocation.

Python in the historical run: 3.14.2 on macOS arm64; no extra packages.
The frozen original input/output limits and timing are recorded in the copied
configuration and runtime records. Matching a new execution on another runtime
would be a separately labelled replication, not a repair of E006.

Integrity checks establish byte identity, not trusted timestamps, absence of
all possible reads, formal third-party assurance, or general competence.
Generator separation was behavioural with shared filesystem permissions;
access logs were author-generated. The post-audit used the same instance.
Neither OS syscall tracing nor an independent human audit is claimed.

Research claim: independently generated, previously unseen finite Boolean
cores were diagnosed correctly across tree, postfix and relation inputs in
the registered E006 space. 60 representations = 20 cores in 10 minimal pairs;
do not treat these as 60 independent statistical observations.

Not established: general epistemic competence, arbitrary-domain reasoning,
consciousness/self-knowledge, autonomous SEARCH, autonomous PASS control,
memory integration or production Colony autonomy.

No new license grant is asserted for historical code or data. Existing notices
remain. Public availability permits inspection; obtain permission for uses
requiring a license. © 2026 Joachim Müller-Peter. All rights reserved.
