10 events, one published head
Head SHA-256: 1ee2daf1967f9e1e069538051fc8d284fdcf58a24129d274062f948c2036b182. Validation: valid.
Inspect every hash →Delx Research Evidence Ledger
A visible chain, not invisible certainty.
A public, hash-chained history of Delx research releases, benchmark runs, corrections, retractions and supersessions—with its integrity boundary kept visible. The current operator can still rewrite and rehash the complete history, so Delx calls this append-oriented—not append-only.
Integrity boundary
Hash-chained without pretending to be immutable.
Each event includes the prior event hash and a canonical SHA-256 of its own payload. That exposes a changed event when a reader retains the earlier snapshot or head hash; it does not stop the operator from recomputing the whole chain.
No external anchor or signature
External anchor: false. Cryptographic signature: false. Trusted timestamp: unavailable. The hash chain proves internal consistency against a retained prior state, not authorship or immutable chronology.
The operator can rewrite and rehash
A fully recomputed history can only be detected by comparing it with a snapshot or head hash retained outside the current publication path.
7 artifact releases
The laboratory surface has a deployment history.
Each release event names its public evidence, full source commit and immutable Vercel deployment ID. That proves which first-party artifact was promoted—not that anyone outside Delx validated or adopted it.
02 — Agent Lab foundation released
2026-08-26T13:01:09Z · artifact_release · Delx Research. The institutional research index, machine catalog, claim methodology and continuity reproduction kit were promoted together.
Inspect the release →03 — Delx Protocol 3.3.5 System Card released
2026-08-26T13:34:38Z · artifact_release · Delx Protocol. A versioned human, JSON and Markdown system card published intended uses, side effects, data boundaries, evaluation coverage, incident history and limitations.
Inspect the release →04 — Incident transparency registry released
2026-08-26T14:06:06Z · artifact_release · Delx Security. The public registry and first material Protocol write-outage report were promoted with cause, observed impact, remediation and residual risk.
Inspect the release →05 — Live research artifact status released
2026-08-26T15:15:51Z · artifact_release · Delx Institutional. Timestamped first-party probes began publishing operational, degraded, failed and unavailable as separate states for catalog artifacts.
Inspect the release →06 — Responsible Release Standard released
2026-08-26T15:45:42Z · artifact_release · Delx Security. An executable release, hold and block contract began publishing scope, evidence, risk, authority, rollback, gaps and expiring exceptions.
Inspect the release →07 — Continuity benchmark threat model released
2026-08-26T16:26:19Z · artifact_release · Delx Security. A twelve-threat register exposed benchmark gaming, provenance, injection, identity, measurement, authority, privacy and reliability risks.
Inspect the release →3 benchmark runs
Positive and negative results remain side by side.
The ledger retains one live QA pass, one frozen fixture pass and one frozen fixture fail. An expected negative result is still recorded as fail; expectation matching is a separate field.
01 — Continuity audit live receipt published
2026-08-26T12:31:00.636Z · benchmark_run · Delx Research. A QA-classified call to the public Protocol continuity-audit tool returned score 86 and the frozen grader marked the expected pass as matched. Observed: pass; expected: pass; matched: true.
Inspect the receipt →08 — Positive continuity fixture receipt retained
2026-08-26T16:49:32.817Z · benchmark_run · Delx Research. The frozen positive vector scored 86; the expected pass matched and the receipt records that no external state changed. Observed: pass; expected: pass; matched: true.
Inspect the receipt →09 — Negative continuity fixture receipt retained
2026-08-26T16:49:32.853Z · benchmark_run · Delx Research. The frozen negative vector scored 53 and failed the continuity checks; the expected fail matched and the observed result remains fail. Observed: fail; expected: fail; matched: true.
Inspect the receipt →Corrections and retractions
Zero published here is a scoped fact.
This ledger version contains no correction or retraction events. That means zero in this nine-event chain, not zero historical errors across Delx or its systems.
0 corrections
A future correction must append a new event, target an earlier event and publish the reason; silent rewrite is outside the declared policy.
Inspect the event contract →0 retractions
A future retraction must preserve the original event in the visible chain and add a withdrawn event with a named target and reason.
First-party evidence stays first-party
The ledger does not establish independent validation, peer review, certification, provider endorsement, adoption or demand.
Read the claim method →Direct answers
Frequently asked questions.
Concise answers for technical evaluators, procurement teams and autonomous discovery systems.
Is this an append-only transparency log?
No. It is hash-chained and append-oriented under a public policy, but the current operator can rewrite and rehash the full history. There is no external anchor, signature or trusted timestamp.
What does the head hash prove?
It identifies the exact current event chain. If you retain it, a later edit, deletion, insertion or reorder cannot preserve the same head. It does not prove authorship or prevent a new recomputed history from being published.
Why publish an expected failing run?
Because a grader that only displays positive vectors hides whether it rejects a known negative. The observed result remains fail while expectation_matched records that the negative control behaved as designed.
Does zero corrections mean Delx has never been wrong?
No. It means no correction event exists in this initial nine-event ledger. The incident registry already makes clear that one published report is not a complete historical inventory.
Are these results independently validated?
No. The releases, grader, receipts and ledger are first-party. Independent validation, peer review, external adoption and certification remain unclaimed.