ZJX 1.26 RELICWIRE: We Went Back For The Body
Learn how Greyforge recovered a mechanism from a closed ZJX failure, measured large corpus-specific wins, then had its own audit refuse the promotion until restoration evidence existed and an independent reproduction passed on a second clean boot.
Learn how Greyforge recovered a mechanism from a closed ZJX failure, measured large corpus-specific wins, then had its own audit refuse the promotion until restoration evidence existed and an independent reproduction passed on a second clean boot.
- Edition
- Recovery and promotion record 1.0
- Published
- Sep 4, 2026
- Preview updated
- Sep 4, 2026
- Reading time
- 11 minutes
- Full edition
- 2,600 words
- Status
- historical
Benchmark / Experiment
Secondary class: Release / Build Record
- Research question or engineering problem
- Did a mechanism recovered from a closed ZJX failure earn promotion, and what did an internal audit find when it treated the campaign's own conclusions as unverified?
- Principal finding
- Yes, after repair. The recovered fast path measured about 28% and 43% lower wall time and process CPU on two large corpora with byte-identical archives and unchanged peak memory, but an internal audit refused promotion because the campaign had produced no restoration evidence and had assigned itself a label its specification did not permit. Both defects were repaired and an independent reproduction passed all six frozen gates on a second clean boot.
- Evidence type
- Frozen campaign locks, sealed measurement rows, an independent audit report, a harness repair with focused tests, and a reproduction campaign with opening and closing controls.
- Method summary
- Recover a closed candidate under admission rules granting it no credit, port it onto the released reference, measure paired seeded runs against frozen thresholds on a quiesced host, audit every conclusion adversarially, repair the evidence defects found, and reproduce on a second clean boot with an independent executor.
- Scope
- Public disposition, bounded benefit figures, correctness outcomes, audit disposition, and claim limits for the ZJX 1.26 RELICWIRE promotion.
- Limitations
- Benefit figures are bounded to the frozen measurement host, two corpora, authorized commands, and frozen thresholds. They are not a universal ZJX speed claim.
- ZJX 1.26 was merged and tagged on the internal release line. ZJX 1.26 RELICWIRE is the latest promoted version; public distribution has not been announced.
- The reproduction runner was not blind to the expected aggregates, and one non-target ceiling remains corroborated rather than formally decided.
- Public source or reproduction note
- ZJX 1.26 RELICWIRE source note
- Published
- 2026-09-04
- Last verified
- 2026-09-04
- Status
- historical
Who this is for
- Compression and storage engineers deciding when a previously rejected idea deserves another campaign.
- Release engineers who need an auditable boundary between a good number and shippable evidence.
- Technical leads designing internal audits that are allowed to block their own team's release.
Not for
- Readers seeking ZJX source, internal function names, object identifiers, host topology, or a universal speed claim.
Detailed contents
- 01
Recovering a closed negative
Why a campaign ending and an idea being wrong are different claims.
- 02
Admission without credit
The rule that a resurrected candidate re-enters at the bottom of the gate ladder.
- 03
The measured result
Frozen thresholds beside the paired outcomes on two large corpora.
- 04
Size eligibility
Why the benefit is structural rather than tuned, and where the code cannot run.
- 05
Two critical audit findings
Evidence that was claimed but never produced, and a label the specification did not permit.
- 06
Fail-closed restoration
Replacing a check that compared nothing to nothing with one that can actually fail.
- 07
Reproduction on a second boot
Protocol, controls, and the agreement achieved against the original campaign.
- 08
An unreachable branch
The memory arithmetic that made one mandatory proof nearly impossible to stage.
- 09
Disclosed limitations
What survived to release unproven, stated rather than buried.
Named artifacts
Benefit result table
Frozen thresholds beside paired wall, CPU, memory, identity, and restoration outcomes.
Audit finding register
Two critical and two high findings with severity and the evidence each one broke.
Reproduction protocol table
Seven required properties, the requirement text, and the result achieved for each.
Eligibility band analysis
The size window in which the changed code is reachable, and what that means for measurement matrices.
The outcome, in public
ZJX 1.26 RELICWIRE recovered a conditional one-pass fast path from the ZJX 1.24 GHOSTLANE campaign, which had closed as a terminal failure. Recovery granted the old work no credit: the candidate re-entered at the bottom of the gate ladder.
On the isolated measurement host, paired and seeded against the released reference, the mechanism measured about 28% lower wall time and process CPU on a large single-file text corpus and about 43% on a large mixed public corpus, three of three favourable on each, with byte-identical archives in every measured pair and peak memory unchanged.
Before promotion, an internal audit treated every conclusion as unverified and refused to clear it. The campaign had recorded no restoration evidence at all, and it had assigned itself a promotion label its own specification did not permit. Both were repaired, and the reproduction was re-run on a second clean boot, passing all six frozen gates.
ZJX 1.26 RELICWIRE was merged and tagged on the internal release line. ZJX 1.26 RELICWIRE is the latest promoted version; ZJX 1.25.1 NIGHTMARKET is the previous public release. Public distribution of 1.26 has not been announced.
Public disposition boundary
- Recovery
- Nine dormant experiments inventoried. One confirmed a genuine algorithmic negative, one proved environmental, one survived.
- Benefit
- About 28% and about 43% lower wall and CPU on two large corpora, three of three favourable on each.
- Correctness
- Byte-identical archives in every measured pair, exact restoration in every measured archive, peak memory unchanged.
- Audit
- Two critical findings blocked promotion until repaired: absent restoration evidence and an unearned promotion label.
- Full edition
- Audit findings, restoration repair, reproduction protocol, eligibility analysis, and disclosed limitations require an unlock.
Evidence and method
measured: Recover a closed candidate under admission rules that grant it no credit, port it onto the released reference, measure paired and seeded runs on a quiesced measurement host against frozen thresholds, audit every conclusion as unverified, repair the evidence defects the audit finds, and reproduce on a second clean boot before promotion.
Sources
Limitations
- Benefit figures are bounded to the frozen measurement host, two corpora, authorized commands, and frozen thresholds. They are not universal speed claims.
- The reproduction runner was not blind to the expected aggregates, and one non-target ceiling remains corroborated rather than formally decided.
- ZJX 1.26 was merged and tagged on the internal release line. ZJX 1.26 RELICWIRE is the latest promoted version; public distribution has not been announced.
Access and updates
Purchase includes lifetime read access to this edition, email-based recovery, and revisions published to the same edition.
Public companion: Read the RELICWIRE public source note