ArtifactReleaseV2 bounded code verification — 2026-09-09
Registry admission now verifies the complete Loader code payload in consecutive 1 MiB chunks. The code commitment binds the total length, chunk profile, ordered offsets, exact chunk lengths and native SHA-256 of each actual chunk. Flat ELF SHA-256 remains offchain build provenance. ArtifactReleaseV2 replaces V1; it does not preserve a parallel accepted codec.
The existing temporary staging account owns verification progress (344 bytes for ArtifactReleaseV2, 296 bytes for ordinary records). Every Finalize rechecks the native Loader identity, ProgramData link, deployment slot, authority and payload length. Only complete coverage and a matching commitment close staging and refund it. Readers must continue requiring the canonical staging account absent before trusting an artifact as finalized.
Actual SBF runtime control
programs/dclutch-registry-sbf/tests/artifact_commitment_v2.rs ran the actual Registry ELF in Agave ProgramTest, with a real Loader deployment padded to 4,194,304 bytes. This is SBF runtime evidence, not local-validator or devnet execution and not a complete cohort build.
| Transaction | Actual CU | Result |
|---|---|---|
| Finalize chunk 1 | 542,117 | Progress 1 MiB; raw and cursor retained |
| Finalize chunk 2 | 542,020 | Progress 2 MiB; raw and cursor retained |
| Finalize chunk 3 | 542,020 | Progress 3 MiB; raw and cursor retained |
| Finalize chunk 4 | 542,604 | Complete; raw retained; cursor closed and refunded |
| Wrong final commitment | 545,015 | Exact RegistryError::ArtifactReleaseCodeCommitmentMismatch; cursor rollback |
| Native Loader Extend | 2,370 | Actual extension of 10,240 bytes |
| Continue after Extend | 12,424 | Exact RegistryError::ReleaseSuperseded; cursor rollback |
Both hostile partial publications were subsequently aborted through the real Registry instruction. Each raw account and cursor disappeared; the sponsor received their combined lamports minus the actual transaction fee. No code payload or staging state was replaced through a harness setter between chunks. The native Loader deployment feature profile explicitly permits the fixture ELF's SBF version; the test sets the runtime transaction limit to 1,400,000 CU.
The accepted control is four transactions for this 4 MiB payload. The current 2,735,824-byte Trading build needs three, and the Loader maximum 10,485,715-byte payload needs ten, derived from the same chunk profile. Those latter counts are geometry, not claims of executed campaigns. The 1 MiB profile is provisional: changing it requires a successor commitment schema and a measured full Finalize transaction, because chunk geometry is part of the code identity.
Provenance and remaining execution
Recovered completed job: /tank/dregg-build/dclutch-structured-root-v2-work-20260908. Its source directory is a recorded working overlay, not a Git checkout or committed release. Recovery compared every Registry and Registry-SBF Rust source and Registry manifest with live tree HEAD 4bee7cc3af1313901253ea6860809d1a088d6c1a plus its dirty V2 migration. All implementation files matched; the only initial difference was an additional cursor unit test. Subsequent changes were fixture corrections, formatting and comments. Thus the runtime supports this implementation, not a claim that all current source was built together.
registry-v2-recovered-implementation-sha256.txt: exact recovered hashes for 26 implementation source files, SHA-256b2aaed64e0db3afbd9e180366df6023ce037eacc2e72a11565fd7e0a1edf856b.registry-v2-runtime3.log: SHA-256866816ea729017895dea8a34a8bf126508f7859381586daefaef935192a87620; one test passed, zero failed, completed 2026-09-09T02:04:15Z.- Registry ELF
target/deploy/dclutch_registry_sbf.so: SHA-25648ae0e74a7146dedf4af46c6b97a66a93425ad3d87447f6eae8d0e016fc1608f. registry-v2-check.log: native package check completed successfully.registry-v2-sbf.log: Registry SBF link completed successfully.registry-v2-cursor-final.log: the exact qualified cursor control passed 1/1, refusing wrong schema width, incomplete upload with verification progress, noncanonical initial/completed progress and an artifact tail on an ordinary record. An earlier unqualified exact filter ran zero tests and is discarded.
registry-v2-resolution-fixtures.log: six explicitly selected SVM-harness constructor controls each passed 1/1, binding the actual ELF tail to the native commitment, distinguishing flat SHA provenance and detecting changed bytes. SHA-256d53ceef6b89bc59b0959d1d5d87ead69bf8835b4bf42746a5ea656305743cd6a.registry-v2-activation-fixture.log: the migrated accepted mutable-slot activation and subsequent supersession control passed 1/1. SHA-2567885e65e61cbe4ae7a4f1a268d0ec51249fb30c57354e67787a352b02c51dda5.
The coherent V2 migration still owes an exact committed all-eight-program build, frame baseline capture, complete economic lifecycle execution and a fresh devnet cohort. This measurement does not discharge those obligations.