Codex-9e8r6t /root/release_audit. Root authorized this preservation-only directory while running separate native checks. No new tests, compilation, GUI actions, source edits, Git staging or publication were performed to create this archive.

Scope and byte preservation

The 37 original records (201,835 bytes) were copied byte-for-byte from the two private evidence directories. inventory.json records their original repository-relative paths, sizes and SHA-256 values, plus the generated README and source readback record. It excludes itself to avoid a recursive hash. The directory contains 40 files total. No executable, DLL, rlib, model output, credential or dependency tree is included.

desktop/ preserves the original relative layout of .local/contextual-source-20260911/. oracle-preparation/ preserves concise records from .local/staged-source-oracle-20260911/. Embedded paths, timestamps, line endings, compiler diagnostics and manifest wording are deliberately unmodified. They describe the original host and run; they do not assert that private executables or dependency paths exist in another checkout.

Root should preserve these bytes through Git using the same narrow attribute convention as neighboring experiments:

harness/experiments/contextual-source-20260911/** -text whitespace=cr-at-eol

This worker did not modify the root attributes or index. The existing global text normalization would otherwise change some CRLF records. Verify every inventory hash again after staging or a checkout with that rule.

Results and distinct source snapshots

Evidence	Meaning
`desktop/focused1*`	Missing private topology Draft Clone; compiler error, Cargo101, no tests. The already-failed compiler was stopped; raw diagnostic retains its process exit code.
`desktop/focused2*`	Extra guide `explain` argument; compiler error, Cargo101, no tests, same stopped-failed-compiler policy.
`desktop/focused3*`	Test called private App method; compiler error, Cargo101, no tests. Corrected to the actual Message route.
`desktop/focused4*`	Successful contextual-source22/22, 2.04s; frozen input manifest and terminal status.
`desktop/tests-verified/`	Root's full675/675, 448.33s test duration, plus exact executable identity `6E677F58...BE38CC0`. This is the earlier snapshot.
`desktop/clippy1*`	Completed strict-lint failure on three constant-event `then` expressions; later corrected to `then_some`.
`desktop/focused5*`	Final contextual-source23/23, 0.98s. The seven-file delta adds the captured document/exact owner recovery regression and narrow guide/lint corrections. Both dated input manifests are retained (about44KB each) because their metadata and hashes bind different executions.
`desktop/guide-final*`	Final guide3/3, 0.20s.
`desktop/clippy2*`	Final all-target strict Clippy with warnings denied, PASS, 4m29s.
`desktop/gui-final*`	Matching final Windows GUI build PASS, 6m20s.
`desktop/fmt-final-status.json`	Workspace formatting PASS. The original runner produced no `fmt-final.log` file because there was no output; no substitute log was fabricated.
`desktop/tests-final/manifest.json`	Final test executable `405577A3...626F38`, inventory676; affected23 and guide3 were executed.
`desktop/gui-verified/manifest.json`	Final GUI `52DE82E4...246ACCE`, with runtime identity. Native acceptance is a separate root task.

There is no full final676 pass in this evidence. The earlier full675 result must not be relabeled as execution of the later binary. Input manifests identify source-file hashes and contemporaneous Git context, not a claim that uncommitted desktop changes were contained in the listed commit or that every external build input was archived.

The original run-final-gates.ps1 and stage-final.ps1 are preserved as raw records. They document serialized commands, failure stops, unchanged-input checks and refusal to overwrite frozen artifacts. They assume repository cwd, the task-local Windows toolchain and private output paths; they were not run from this archive. Any future rerun needs an explicitly coordinated compiler slot and fresh output paths, and produces new evidence rather than replacing these records.

Independent oracle preparation

oracle-preparation/manifest.json binds the independently authored evaluator, fixed two-file baseline, designated core rlib and executable. build-initial.log is an actual zero-byte successful diagnostics log, accompanied by its exit/time JSON. Self-check and baseline logs/status passed. The small CLI baseline check records admission with expected warnings, without source writes. These records establish evaluator preparation and falsifier execution, not acceptance of GUI-produced staged-source objects. The already-owned oracle source, fixture and report remain at harness/staged-source-authoring-contract.rs, harness/fixtures/staged-source-authoring/ and harness/release-staged-source-authoring-oracle-20260911.md; they are not duplicated here.

source-freeze-readback.json is a new preservation-time check: all13 desktop source hashes matched the final focused5 snapshot. The implementation and independent-review reports remain outside this directory and were not edited by this task.

To verify this archive without running any application, from this directory:

$inventory = Get-Content -LiteralPath inventory.json -Raw | ConvertFrom-Json
foreach ($entry in $inventory.files) {
    $file = Get-Item -LiteralPath $entry.path
    if ($file.Length -ne $entry.bytes -or
        (Get-FileHash -LiteralPath $file.FullName -Algorithm SHA256).Hash -ne $entry.sha256) {
        throw "Evidence mismatch: $($entry.path)"
    }
}