Redline current publication closure, 2026-11-09
redline-current-module.cjs now accepts the explicit version
redline-current-publication-daa44c07c-v1. The default path retains the old
manifest and refuses today's changed providers. Preparation is offline only:
node tools/redline-current-module.cjs --prepare scratch/runs/OWNED/fresh \
--publication-version redline-current-publication-daa44c07c-v1
The Redline-specific adapter authenticates the retained 1235 manifest, exact published 5c8780318 CPU mirror/build inputs, f977ebf7c collector, d0a9965a8 launcher, d54aa57e3 helper, released 06a42152c outer-caller ancestry and d7459f3f6/daa44c07c snapshot repair. It adds the actual trace's wire helper, startup boundary, caller, and snapshot validator transitively. Every exception names published Git bytes or the two owned preparer files; unrelated drift still refuses. No provider implementation was changed or copied.
The resulting plan points at the actual single-comma fixture controller. Its existing child-launch site invokes the published outer adapter, which starts the generated owning caller. The controller's joined completion invokes the published bounded retrieval implementation. The generated small imports are relative and work under another staging root. Prelaunch reauthenticates the complete narrow closure before any custody child spawn. The child verifies its generated closure digest and actual pinned bytes without running Git under ptrace. Default-OFF adapter argv and spawn options are unchanged.
Evidence is in scratch/runs/20261109-redline-current-closure-preparation-1595/.
git-base.txt records the exact base; owned.patch contains only these four
paths. There are no source archives, media copies, or WASM copies. Small evidence
records and completed raw journals are losslessly compressed with SHA/size and
reconstruction recipes. Shared disk briefly crossed the 1.5 GB floor despite
ongoing own compression; subsequent compression restored headroom. No foreign
run cleanup was performed.
Validation:
- Memory candidate passed before changing the current preparer. Its first full-caller run correctly caught the absent transitive wire-helper pin.
- Final production CLI preparation, full generated-caller preflight, real outer custody launch of that preflight, bounded retrieval, and relocated imports pass. These never start Chrome, a server, or a game. Both virtual staged historical closure and each missing Python provider refuse; the complete staged snapshot validator executes successfully against synthetic evidence.
- Twenty-three missing/SHA/drift/unknown-version controls refuse before spawn. Default-OFF literal argv/spawn parity passes. Prepared inputs remain small; all provider/module/media bytes are referenced in place.
- Existing nine actual caller cases pass: direct/rename, errno 20/9/2, child-exec, cancellation and OFF normal/cancel. Assertions are unchanged; an evidence-only in-memory test hook compresses completed own journals.
- Snapshot replacement/growth/FIFO/symlink/expansion and fd growth/truncation, compressed/expanded caps, damaged/truncated/empty and concatenated gzip controls pass. Four launcher, four observer, 46 kernel, eight reaper, 14 readiness-boundary, trace-prefix and fixture-shutdown controls pass. Readiness retains 100 waits / 101 stats / zero reads in the missing case.
flock /tmp/wine-build.lock bash tools/build.shpasses. Test groups have finite independent watchdog receipts, actual waits and separate outer pipe EOF; custody-owned identities are checked reaped. Collector command output remains merged; outer EOF does not manufacture separate command/browser EOF.
Failures retained: an early fixed 40-second offline outer preflight expired while redundantly authenticating Git under ptrace. A fresh run at the same bound passes after moving that work before spawn; no running deadline was extended. Another early test looked for the collector's merged command output in the outer pipe and failed; the final test checks the actual command log. The legacy full trace suite refuses the current CPU because its immutable startup-admission provider still selects historical a320d4c8 bytes. This work does not retarget that admission or claim native readiness.
Two initial Python test invocations omitted their run-directory environment
overrides: launcher setup rewrote the default 1505 compiler metadata before
refusing existing cases, and observer tests wrote their default 1485 evidence.
Those invocations and errors/results are retained, with no restoration claim.
Fresh owned-directory invocations pass. One shutdown invocation also used its
default base, creating a fresh suite-* directory; the owned-directory repeat
passes. These are evidence-location mistakes, not provider edits.
Classification remains strict: CRC-valid compressed-empty streams are
EMPTY_INVALID; damaged gzip stays invalid. Inherited 1515's four invalid
stdout/stderr files were two zero-byte stdout files and two 10-byte truncated
stderr files, not four zero-byte files. Parent review's 17 empty/invalid cases
include valid compressed-empty streams. None is promoted to native proof.
The original deadline plus 30-second cleanup, sticky first failure, sole ptrace wait owner and all readiness/input/browser-child semantics remain in the published providers. Cancellation browser wait is UNKNOWN, historical startup and PID 8107 are UNKNOWN, per-stop fsync cost is UNMEASURED, nativeReady is FALSE. No Box was created or contacted: ID/state NOT_CREATED, expiry/stop receipt N/A. Any native attempt, including resolving the historical CPU admission refusal, requires separate authorization. No native retry is authorized here.