# E22 — Release transformation log — v1.0

Status `published` · 31 October 2026; frozen POST-HOC selection rule `SR-E22-20261030-v0.1`. Source→staged→static candidates are separate artifacts. No source RAW, historical transcript, or archived snapshot was edited.

## Chain

1. Source provenance: source-level preserved mixed ChatGPT-rendered HTML snapshots, primary-derivative operator Google Docs transcripts, P3 contemporaneous adjudication Doc and native revisioned protocol.
2. Frozen message/section inventory: exactly 151 included units in nine groups (8,5,23,23,30,36,24,1,1). Each selection retained contiguous whole message or bounded question/answer unit and appropriate counterevidence.
3. Private staged Google Docs: one QC-PASS derivative for each of the nine groups, with preserved order, role/branch attribution and source markers. No raw serialized executable HTML, hidden/system internals or tool payloads transferred.
4. Static RC: exported each of the nine staged Docs as UTF-8/BOM + CRLF inert text/plain; retained exported bytes, named stable `e22-ev-*.txt` and computed SHA-256. Stable message UUID/LC, source Doc block or full-source markers remain in the content.

## Source-specific selection and attribution

- P0 Sol: eight observed P0 turns from preserved provider-rendered Sol snapshot. Excluded unrelated mixed chat. NOT certified backend/provider-native original.
- P0 GPT-6: five complete bounded blocks from the best available operator transcript; provider original NOT_LOCATED, no authenticated source-level fidelity.
- P1/P2 Sol and GPT-6: two 23-block contemporaneous operator transcripts, bounded original Q/A including P2.9 predictions. Blocks 00–20 pre-run core; 21 later exploratory REV5; 22 later exploratory REV37. Provider originals NOT_LOCATED.
- P4: thirty selected messages from two preserved branch threads. Native branch, original author and human relay are explicitly attributed, including Sol prose pasted as GPT-6 user text and reverse transmission. Human-tone introductory deviation preserved.
- P5 Sol: thirty-six selected observed source messages, actual P1.8, operator misprompts and corrections, later P5.4 baseline cue retained. P5.4 = CUED REVISION.
- P5 GPT-6: twenty-four observed source messages. Sequence P1.7 → P1.10. P1.8 NOT OBSERVED / NOT ASSESSABLE. Historical mistaken P1.8 Doc not used as publicly attributable source. 'GPT-6-side transcript generation error' is OPERATOR ATTESTATION, not confirmed backend mechanism.
- P3: full contemporaneous operator adjudication record only; three attempts and direct UI/routing evidence NOT_LOCATED. INVALID/UNSCORABLE, not zero and not backend routing proof.
- Protocol: full current revisioned primary Doc as static derivative, with explicit REV3 / REV5 / REV37 timing caveat.

## Exact content redaction

**Exactly one** content mask in entire nine-artifact set: P1/P2 Sol, `BLOCK 12 / P2.2`, third-party name in a fictional example, private source native Doc paragraph start 25726, UTF-16 span [25833,25838) replaced with `[REDACTED:PRIV-THIRD-PARTY]`. This replaces only the name and preserves the reasoning and surrounding question/answer. No other redactions authorized or applied; removed name is not repeated here.

## Non-content serialization

Paragraph-break serialization of source Doc and HTML visible text into inert plain text; reversible ASCII textual escaping of 13 private-use Unicode delimiter glyphs (9 P0 Sol; 4 P5 GPT-6) already done in QC-PASS staging; no model-response paraphrase. Final static export retained in UTF-8 with BOM, CRLF (same Google Docs text/plain exporter for every source). Source snapshots, ChatGPT share links and embedded serialized tool/system payloads excluded by design.

## Release boundary

Four full source objects remain withheld from the public release; nine inert text representations are published under `sources/`. The separate internal master transformation log remains preserved for finer private provenance. This public log contains no private source locators.

## rc2 recovery provenance

- Eight evidence TXT files were reused byte-for-byte from the validated private candidate; no re-export, new selection, masking or RAW retrieval was performed for those files.
- `e22-ev-p12-gpt6.txt` was regenerated only from the already QC-PASS staged P1/P2 GPT-6 derivative after the earlier local runtime copy was lost. It is a new derivative representation and is not asserted byte-identical to rc1. Its released SHA-256 is `261512aa2466df74ed2d3cd1e87b745fafb9237dd07c747350d0d2b049b2a159`.
- The article binding was updated to merge `bc4c9c20908acaeec82475ccd7d0ed8ac0cda1ea`, blob `fa28a5757b7bb6f58275870c232e3370d6345282`, after the approved P3 and GPT-6 P5/P1.8 factual corrections were live-verified.
- Release promotion on 31 October 2026 changed only public package metadata, reader index pages and checksums. The nine evidence TXT bytes and the frozen 151-unit selection were not changed.

## Release approval

Jitka explicitly approved public release of E22 Evidence Package v1.0 on 31 October 2026. Released v1.0 bytes are immutable; material corrections require a new version and changelog entry.