Strict all-eight ArtifactRelease V2 candidate and frame baseline — 2026-09-09

Verdict

The exact committed source fd7e5fec68cb948a326b41dc69be76f27156da10 completed the checked genesis-candidate runner on hbox with exit 0. All eight program links were freshly compiled under swarm-build, their shipped build diagnostic count was zero, and the freshness gate admitted all eight. The source-pinned Product handoff, eight role manifests, five-role execution set, genesis infrastructure profile pair, checked Upgrade gate, reproducible gate, run record and successor campaign pack all sealed.

The reproducible gate and campaign pack were then independently reopened with their source-pinned verifiers. A separate canonical tools/gate frames capture and an independent capture assembled from fresh exact-source objects agreed byte-for-byte. The accepted baseline in tools/gates/frames-baseline.json therefore covers this commit and these eight links. It does not cover later General, Structured, or Claims native changes; those changes remain frame debt until the next exact-source completion build and two-capture admission.

This is local reproducible-build evidence. It is not validator execution, a deployment, or chain evidence.

Source and builder

The run began at 2026-09-09 00:37:49 EDT and wrote exit status 0 at 00:53:45 EDT. Its persistent driver log contains no Unit run-u scheduler no-op marker and no runner refusal.

Shipped links

sbf_shipped_feature_suffix is unconditional and empty at the source revision: every ELF below is the ordinary production build with no Cargo feature. The role provenance descriptors record the exact locked invocation.

The separate Trading CU diagnostic is profiled/trading.so, SHA-256 6397b6fc1a7ed30d3c79a8241b3d068607bdcb33235019a2a65d21a51a3ca6dd. It alone carries --features hot-cu-profile; it is measurement-only and is absent from every checked role manifest and release set. The candidate's own Trading frame report uses that diagnostic feature and was consequently not used to admit the production frame baseline.

Sealed candidate

The candidate records infrastructure_lineage=genesis and profile version 1+2. These independent reopen checks passed after the runner exited:

artifact_provenance.py verify-reproducible-gate
  gate_sha256=a2423b95a11f0917a16b0d5199c761eb81c7db20c8e6a57a92de07d890a32e69
  link_count=8
  source_revision=fd7e5fec68cb948a326b41dc69be76f27156da10

successor_campaign_pack.py verify
  successor campaign release pack verified
  sha256=c8f14757c784744be98ed4726f608ef5838e4c480882c18193db13b32e55140e

Independent frame admission

Capture B was a full fresh tools/gate frames --at fd7e5fec... --capture run in a detached checkout, with all source, targets, logs and output under /tank/dregg-build/dclutch-frames-fd7e5fec6-v2-20260909.

Capture A reused the checked candidate's seven fresh, exact-semantics production frame objects. Trading was rebuilt alone, without hot-cu-profile, in a new target under swarm-build; its log contains the fresh top-package compile marker and zero over-bound diagnostics. The source-pinned frame parser produced all eight complete per-function reports, and the source-pinned frames assemble command constructed Capture A.

The two captures are byte-identical:

capture A SHA-256 = f661439753dc4d10ea9fecff9385da3b59d42cb15ca7ab5601a45f9e3f6f4a3d
capture B SHA-256 = f661439753dc4d10ea9fecff9385da3b59d42cb15ca7ab5601a45f9e3f6f4a3d

tools/gate frames accept emitted the baseline now installed in the tree:

schema=dclutch-sbf-frame-baseline-v1
commit=fd7e5fec68cb948a326b41dc69be76f27156da10
link_count=8
bound_bytes=4096
sha256=c6bf2d9704362434a9230b3519ebd0d57ea9ca7510de97c3ecfd0beaa7ba6088

frames check passed against the accepted baseline. frames owed from that baseline through the captured commit reported zero intervening commits and all eight dependency closures resolved over twenty first-party crates. The capture-composition record SHA-256 is f61f3e16a816319b119f4b704932fb6076f3a69abbd31c030a6695e0470b8b5a; its eight object/report/build-log rows hash to e17b360c5b1c37bd1178d31e97c10fef3c71a97044f090b6f78f8a6ce3146a24.

The 4,032-byte Trading maximum is 64 bytes below the 4,096-byte SBPF v0 wall. It is admitted evidence, not spare budget: later changes must be remeasured rather than inferred safe from this margin.

At integration HEAD c76ab9c64fa1119bbe441c7d27de409e964b7a25, the closure-aware frames owed census correctly refused on seven post-fd7 commits: four General changes, two Structured changes, and one Claims change. That red is the intended handoff to the next all-eight build, not evidence against the accepted fd7 capture.