# ZJX 1.24 GHOSTLANE public source note Published: 2026-09-01 Verified: 2026-09-01 Status: closed negative experiment, not released ## Claim boundary ZJX 1.24 GHOSTLANE was a bounded fast-path experiment on the released ZJX 1.23 BLACKWIRE parent. Its corrected materiality campaign observed the fast path, passed exactness and resource gates, and measured a material speed improvement on the authorized paired comparisons. The candidate then failed the frozen repository-wide validation gate and was not tagged, released, or deployed. The 1.24 version number is consumed and will not be reused. ZJX 1.23 BLACKWIRE remains the current released ZJX production line. This source note publishes the release disposition, the bounded materiality result, the full-validation counts, and the limitations needed to evaluate the public claims. The full Chronicle contains the corrected-treatment analysis, per-corpus figures, the failure-class breakdown, and the transferable release doctrine, and requires an unlock. Canonical Chronicle: Machine-readable evidence: Current release line: Product page: ## Experiment identity - Experiment: ZJX 1.24 GHOSTLANE - Parent: released ZJX 1.23 BLACKWIRE (`v1.23`) - Mechanism class: conditional one-pass fast path for one codec stage of solid storage, gated by a pure admission check against the existing resource governor - Not changed: archive format, decoder contract, candidate universe, candidate order, compression levels, memory ceiling, spool ceiling, pruning, concurrency - Toolchain: Zig 0.16.0, `ReleaseSafe` - Corpora: enwik8 and Silesia (public benchmark corpora) - Lifetime pack-start ceiling: 20; consumed: 16 - Final disposition: `FAIL` at gate R12; gate R13 not run; release not authorized; version not reusable ## Campaign dispositions | Campaign | Starts | Disposition | Reason | |---|---:|---|---| | Attempt 1 | 1 to 8 | INVALID | Every qualifying command selected chunked storage, so the solid-storage fast path was never executed. The timings describe the parent-like path and are decision-ineligible. | | Attempt 2 | 9 to 16 | MATERIALITY PASS | Fast path observed, four exactness pairs passed, resource gate passed, and the frozen timing thresholds were met. | | Release closeout | none consumed | FAIL | Repository-authoritative full validation returned nonzero. Clean reproduction (R13) was not run. | INVALID and FAIL are distinct states. INVALID means the experiment did not answer the question. FAIL means a valid experiment reached a frozen gate and missed it. ## Attempt 2 materiality result | Check | Threshold | Result | |---|---|---:| | Paired wins | At least 3 of 4 | 4 of 4 | | Median corpus improvement | At least 3% | 22.35% | | Each corpus median improvement | Greater than 0% | Met | | Fast path observed | Required | Yes | | Peak memory non-regression | Within frozen allowance | Pass | | Exactness pairs | 4 of 4 | 4 of 4 | | Focused native proof | All | 40 of 40 | | Thresholds changed during campaign | Must be no | No | Exactness covered archive bytes, archive size, restored tree identity, and normalized trace identity for every pair. Candidate median peak memory was at or below the parent on both corpora. The 22.35% figure is the median of the two corpus median-of-runs improvements under the frozen campaign. It is bounded to the retained laboratory host, the two corpora, the authorized commands, and the frozen thresholds. It is not a universal ZJX speed claim, and it does not describe a released product. ## Release closeout result Native validation: - Build steps: 8 of 8 succeeded - Tests: 39 of 40 passed, 1 skipped - Formatting: clean Laboratory validation: - Tests run: 232 - Failures: 2 - Errors: 7 - Skips: 20 - Exit: nonzero The nonzero exit made gate R12 a terminal FAIL. All nine terminal cases occurred in code outside the GHOSTLANE fast path. Their root causes are recorded as SUSPECTED, not PROVEN: the retained logs prove the symptoms but do not adjudicate whether each expectation or the laboratory image was at fault. ## Public claim map Safe to state: - ZJX 1.24 GHOSTLANE was evaluated, failed release qualification, and was not released. - The corrected campaign observed the fast path and measured 4 of 4 paired wins with a 22.35% median improvement against the frozen thresholds. - The first campaign was INVALID because it did not execute the treatment. - Full validation failed with 2 failures and 7 errors in 232 laboratory tests, none in the changed code path. - ZJX 1.23 BLACKWIRE remains the current release. Not supported: - GHOSTLANE broke the ZJX test suite. - GHOSTLANE was release-ready and only an environment problem stopped it. - ZJX became 22.35% faster. ## Limitations - Performance measurements apply to the frozen laboratory host, two public corpora, authorized commands, and thresholds of the corrected campaign. - The result belongs to an unreleased candidate. It does not describe the current ZJX 1.23 BLACKWIRE product or later releases. - Root causes of the nine terminal cases were not fully adjudicated inside the campaign and remain SUSPECTED. - One expected standalone failure bundle named by an archival registry could not be located; the export records it as NOT_FOUND rather than substituting a replacement. - Repository identity, object identifiers, host paths, binary digests, internal function and fixture names, and raw traces are excluded from the public record. ## Evidence interpretation Treat attempt 1 only as retained evidence that the campaign protocol rejected a non-treatment measurement. Treat attempt 2 as the sole materiality result for the fast path. Treat the release closeout as the release decision: FAIL, not released. For the released line, see the [ZJX 1.23 BLACKWIRE source note](https://greyforge.tech/research/zjx-1-23-blackwire-source-note-2026-08-31.md). For the rules behind bounded compression claims, see [ZJX and the Discipline of Compression Evidence](https://greyforge.tech/chronicles/zjx-compression-evidence-discipline). For the earlier negative result in this lineage, see [ZJX 1.22 NULLSPOOL](https://greyforge.tech/chronicles/zjx-1-22-nullspool).