# ZJX 1.23 BLACKWIRE public source note

Published: 2026-08-31
Verified: 2026-08-31
Status: current release evidence

## Claim boundary

ZJX 1.23 BLACKWIRE is the current released ZJX production line for Linux x86-64. Its canonical `zjx` and `zjx-archive` programs are native ELF executables. The qualified archive lifecycle completed without Python or Node, while the permitted native Brotli codec remained available where required.

This source note publishes the release identity, artifact digests, bounded qualification result, and limitations needed to evaluate the public claims. The full Chronicle contains the retained invalid-campaign analysis and harness-repair lessons and requires an unlock.

Canonical Chronicle: <https://greyforge.tech/chronicles/zjx-1-23-blackwire>
Product page: <https://greyforge.tech/products/zjx>
Public release record: <https://greyforge.tech/research/zjx-1-23-blackwire-release-record-2026-08-31.md>
Public Campaign 3 evidence: <https://greyforge.tech/research/zjx-1-23-blackwire-campaign3-evidence-2026-08-31.json>

## Release identity

- Release: ZJX 1.23 BLACKWIRE
- Tag: `v1.23`
- Target: Linux x86-64
- Compiler: Zig 0.16.0
- Compiler SHA-256: `2317bbb91798556d9d0f38aabdac23db83f0979b25f767259ae474546724087c`
- Build mode: `ReleaseSafe`
- Incremental compilation: disabled
- Release asset inventory: six, comprising three artifacts and three digest sidecars

The released executable files are the sealed qualified artifacts. Release packaging did not rebuild, relink, or substitute them.

## Qualified artifacts

| Artifact | Bytes | SHA-256 |
|---|---:|---|
| `zjx-1.23-blackwire-linux-x86_64` | 7,164,360 | `ebeeeecab227dde4ff9bb0116d672b253289e4b42f70893839b66d9657bb9390` |
| `zjx-archive-1.23-blackwire-linux-x86_64` | 6,696,840 | `1aca307183719c2823efadbc0e0b9d38c1d4320c1aaf21994e7a0fdb1551a262` |
| `zjx-1.23-source.tar.gz` | 474,872 | `3d003c7ae7f6c1fef81ffd78ee59f0c2bcbadfe0ca863a26fcc003af52542d32` |

All six release assets were downloaded from the operator-controlled release and revalidated. The three digest sidecars passed. Both executable artifacts remained native Linux x86-64 ELF files and reported `ReleaseSafe` build mode.

## Production boundary

| Dimension | Before BLACKWIRE | ZJX 1.23 BLACKWIRE |
|---|---|---|
| Canonical `zjx` | Wrapper-controlled product boundary | Native Zig ELF |
| Canonical `zjx-archive` | Mixed runtime and wrapper assumptions | Native Zig ELF |
| Python in production | Possible through canonical dispatch | Not required or automatically selected |
| Node in production | Not clean-host proven | Not required; clean-host lifecycle passed |
| Experimental dispatch | Could share product-facing entry points | Explicit laboratory boundary only |
| Benchmark default | Could include wrapper or interpreter cost | Native product identity |
| Reference code | Operationally adjacent | Verification and laboratory boundary |

Reference and laboratory code remains in the source tree. The release claim is narrower: interpreted ZJX code no longer owns or automatically enters the canonical production path.

## Campaign 3 qualification result

Campaign 3 completed VALID and PASS under the frozen seven-corpus contract. Candidate and control archive bytes matched for every qualified corpus and repetition. All four operations remained within the declared regression thresholds, with zero threshold violations.

The machine-readable [public Campaign 3 evidence record](https://greyforge.tech/research/zjx-1-23-blackwire-campaign3-evidence-2026-08-31.json) preserves the measurements, counts, integrity digests, and claim limits behind this table.

| Operation | Median wall-time delta | Median CPU delta |
|---|---:|---:|
| `pack` | -0.23% | -1.36% |
| `test` | -11.81% | -3.43% |
| `inspect` | -94.37% | -74.24% |
| `unpack` | -17.33% | -5.45% |

The negative values indicate lower measured time for the candidate within this campaign. They do not establish a universal speedup.

Additional qualified observations:

- Startup median: candidate `0.0016957125` seconds; control `0.068739469` seconds; measured delta `-97.53%`.
- Candidate pack peak RSS was equal to or lower than control for every qualified corpus.
- Candidate process classification recorded five ZJX-native classifications and one native-codec classification.
- Candidate subprocess count: one permitted native Brotli codec.
- Forbidden-interpreter attempts: zero.
- Forbidden-interpreter executions: zero.
- Unclassified child processes: zero.
- Decision: `NO_MATERIAL_BLACKWIRE_RUNTIME_REGRESSION`.

## Archive and release validation

The bounded postrelease smoke passed `zjx --help`, pack, inspect, test, and unpack. The restored tree matched the source byte-for-byte. The companion `zjx-archive` create and verify lifecycle passed, including the archive digest check.

The release does not introduce a new archive format or a new compression-ratio claim. [ZJX 1.21 STREAMWALL](https://greyforge.tech/chronicles/zjx-1-21-streamwall-bounded-memory) remains the relevant bounded-memory lineage; BLACKWIRE changes the canonical runtime boundary and qualifies that boundary against retained compatibility evidence.

## Invalid-campaign handling

Two earlier BLACKWIRE campaigns were retained as INVALID because their harness contracts were not adequate for a product conclusion. Their timings are excluded from the release result. Campaign 3 ran only after the harness was separately qualified.

The exact failure mechanisms and the reusable qualification method are part of the paid Chronicle. This public note preserves enough information to audit the released claim without giving away the full internal failure analysis.

## Limitations

- Performance measurements apply to the frozen host, seven corpora, commands, repetitions, and thresholds used by Campaign 3.
- The figures are not guarantees for other machines, files, codecs, or workloads.
- Exact candidate-control archive identity applies to the declared evidence set, not every possible input.
- The process audit permits declared native codec subprocesses; it does not claim a single-process implementation.
- Release validation used the sealed candidate and published assets. It did not rebuild the artifacts or begin ZJX 1.24 work.

## Evidence interpretation

Use the release record and this note for public product facts. Treat Campaign 1 and Campaign 2 only as retained evidence that the harness rejected invalid runs. Treat Campaign 3 as the sole performance qualification result for BLACKWIRE.

For the broader rules behind bounded compression claims, see [ZJX and the Discipline of Compression Evidence](https://greyforge.tech/chronicles/zjx-compression-evidence-discipline).
