Strix required workflow resolves evidence binder from target trusted workspace instead of protected central owner

[seonghobae]

Symptom

BandScope PR #865 exact head 1f4877413e2eed30b224eaf1b095af3b0b905cb0 has a terminal Strix failure in job 105972205103 even though the protected central .github owner at e6334e229581a918e2f22de18733b76fa65d7e71 contains scripts/ci/strix_evidence_binding.py.

The job first passes required-workflow smoke and materializes the protected BandScope base as the trusted workspace. The model provider preflight reaches contextual-orchestrator / orchestrator/free; a later Strix model call times out, but the terminal policy failure is independent and deterministic:

ERROR: Strix evidence binder is missing: .../trusted-workspace/scripts/ci/strix_evidence_binding.py

The protected central quick gate itself documents the binder as scripts/ci/strix_evidence_binding.py. Current scripts/ci/strix_quick_gate.sh derives binder="$REPO_ROOT/scripts/ci/strix_evidence_binding.py"; under the reusable target-repository job, REPO_ROOT is the target trusted workspace rather than the protected central workflow checkout. BandScope correctly does not carry this central CI implementation file, so the required gate exits 2 before it can produce/classify review evidence.

Why this is an owner defect

This binder is part of the central Strix evidence-binding/runtime contract, not application source. Copying it into every consumer repository would duplicate the canonical owner and make required-workflow behavior depend on an undeclared target-repository implementation detail. Weakening or skipping the binder is also invalid because it is the mechanism that classifies pr_delta / repository_baseline / context_dependency and sanitizes unsupported remediation claims.

The BandScope #865 lane has therefore recorded this as a central owner prerequisite and is not adding a local copy or blind-rerunning Strix.

Expected repair

Resolve the binder (and any companion central runtime dependency) from the exact protected .github checkout/materialized central runtime used by the required workflow, while continuing to evaluate target-repository changed-file/base evidence against the trusted target workspace. Add a contract test proving a consumer repo with no scripts/ci/strix_evidence_binding.py can still run the required Strix gate from the protected central owner.

Do not change failure semantics for a genuinely missing binder in the central owner and do not fall back to mutable target-branch source.

Evidence / scope

• Consumer: ContextualWisdomLab/bandscope#865
• Consumer exact head: 1f4877413e2eed30b224eaf1b095af3b0b905cb0
• Failed job: 105972205103
• Protected central owner inspected: .github@e6334e229581a918e2f22de18733b76fa65d7e71
• Protected central owner does contain scripts/ci/strix_evidence_binding.py and its doctoring/tests.
• No open .github issue/PR matching the exact missing-binder/trusted-workspace failure was found at filing time.

This issue is only the central required-workflow path defect. The provider timeout in the same job is a separate transient/service outcome and should not be conflated with the deterministic missing-binder exit.

[seonghobae]

LineageWeave provides a second exact consumer specimen for this same owner defect.

Consumer: ContextualWisdomLab/LineageWeave#899
Exact head: c943060c7c16f74faf48d1ee40eaa5301c830065
Strix workflow: 35558853933
Strix job: 106321781463
Artifact: strix-reports (10656007820, SHA-256 169a527d7d13852d4ffcaf99c83551893d70449c7dd89608ecc695918d211a7b)

The runner, central Strix source checkout, target workspace materialization, CO sidecar provisioning, orchestrator/free model path, and Strix installation all succeeded. The Strix scan itself reached status=completed, scan_results.success=true, emitted a penetration report, and produced SARIF with 0 results. The wrapper still made the job terminal FAILURE because the post-scan evidence-binding stage reported:

ERROR: Strix evidence binder is missing: /home/runner/work/_temp/trusted-workspace/scripts/ci/strix_evidence_binding.py

This cleanly separates the defect from provider/model execution: unlike the earlier BandScope specimen, LineageWeave completed the actual scan successfully. The failure is therefore deterministic central-runtime path ownership after scan completion, not a LineageWeave source/security finding and not a reason to copy strix_evidence_binding.py into the consumer repo.

Please keep the repair owner-side: resolve the binder from the protected .github runtime checkout, preserve target-workspace evidence as data only, add a consumer-without-local-binder regression, and retain fail-closed behavior if the protected central binder is genuinely absent. No leaf rerun/no-op wake commit is being used.

[seonghobae]

Additional LineageWeave canary: translation-ledger PR #929 exact d4f42f579663e88a0c9af0cc492aa6ff7cae96ee, Strix workflow 35236142575, job 105329535481, artifact strix-reports 10524752712 (SHA-256 77af2e4db1983311a37a8d0727008bf6aae1eaee5e68563f530ee857ce6c27ff).

Again, the Strix engine itself completed and emitted a report plus SARIF with 0 results; the report explicitly assessed the new translation-ledger endpoint/cache/migrations and recorded scope limitations. The wrapper then terminated on the same deterministic post-scan path error:

ERROR: Strix evidence binder is missing: /home/runner/work/_temp/trusted-workspace/scripts/ci/strix_evidence_binding.py

This specimen is particularly useful because the raw Strix report contains scope-qualified and some broader security claims. The missing central binder means the workflow never reaches the canonical evidence-classification/sanitization step that should bind those claims to pr_delta / repository baseline / context dependency. Therefore neither the 0-result SARIF nor broad prose claims should be promoted as authoritative required-gate acceptance until #2291/#2292 lands and the unchanged consumer head is revalidated normally.

LineageWeave will keep #929 Draft and will not vendor the binder or convert the raw report into a synthetic pass.

[seonghobae]

Additional exact-head LineageWeave canary: LineageWeave#1039@9da817da8306c8ac8c631ff1c19f6c74dd6e35a1, Strix run 34671908661 attempt 2, job 106391096873.

The job admitted/materialized the exact target, self-tested the required workflow, provisioned contextual-orchestrator, installed Strix, and reached Run Strix (quick). Attempt 2 then failed there after the model path timed out, and artifact strix-reports 10661318604 (sha256:62411f81faafbee2e39b3ddc9d511e89caafd6eecf5f061d83abbc3872ad9d45) was uploaded. Its gate console independently ends with the same deterministic owner-path failure:

ERROR: Strix evidence binder is missing: /home/runner/work/_temp/trusted-workspace/scripts/ci/strix_evidence_binding.py

This exact-head rerun must be kept separate from run attempt 1 / job 103494782412, whose now-expired artifact was the historical exact-head security report that had already yielded three Medium product findings. Attempt 2 did not produce a replacement vulnerability verdict, so it neither clears nor supersedes those historical findings; it adds another consumer reproduction of the central binder defect plus a separate transient/provider timeout outcome.

Do not copy the binder into LineageWeave or reinterpret the attempt-2 failure as an application finding. Canonical repair remains #2291 exact-head acceptance/protected integration, then unchanged consumer revalidation.

[seonghobae]

Additional exact consumer reproduction from Wardnet confirms this is the same deterministic central-owner defect, independent of the scanner result.

Consumer: ContextualWisdomLab/wardnet#435
Exact consumer head: 11e7fabfca76e0051fa883905228a0dbe472f4df
Required-workflow run/job: 35751756065 / 106864846606
Protected central workflow SHA: .github@e6334e229581a918e2f22de18733b76fa65d7e71

The trusted Strix run completed its PR-scope scan and reported:

Vulnerabilities 0 (No exploitable vulnerabilities detected)

It then exited 2 solely because the central quick gate resolved the binder from the target trusted workspace:

ERROR: Strix evidence binder is missing: /home/runner/work/_temp/trusted-workspace/scripts/ci/strix_evidence_binding.py

Wardnet intentionally does not carry the central binder and will not copy it locally or weaken/bypass the required check.

RED→GREEN acceptance for this reproduction:

1. the reusable Strix workflow resolves strix_evidence_binding.py (and any companion central runtime dependency) from the exact protected .github workflow/runtime checkout, not from consumer source;
2. changed-file/base/head evidence remains evaluated against the trusted consumer workspace and exact PR head/base;
3. the binder still fails closed if missing from the protected central owner—no mutable target-branch fallback;
4. on unchanged wardnet#435@11e7fabfca76e0051fa883905228a0dbe472f4df, the required Strix run reaches terminal success after binding the zero-finding report to the exact consumer evidence.

This is a central required-workflow/runtime repair; no Wardnet product-security change is authorized by this failure.

[seonghobae]

OriginWeave now reproduces this owner defect with a stronger exact-head specimen.

Consumer: ContextualWisdomLab/OriginWeave#37
Consumer exact head: 219b43bfa87ab4fd90a77bb962df2797925bf661
Required Strix run/job: 35816714371 / 107075508403
Protected central workflow source: .github@e6334e229581a918e2f22de18733b76fa65d7e71

The scan itself completed and reported Vulnerabilities 0 (No exploitable vulnerabilities detected). The required job nevertheless terminated immediately afterward with exit 2:

ERROR: Strix evidence binder is missing: /home/runner/work/_temp/trusted-workspace/scripts/ci/strix_evidence_binding.py

This isolates the deterministic failure from provider/Caido availability for this specimen: contextual-orchestrator preflight reached gateway.status=ready, Strix completed its PR-scope analysis, and the failure occurred only when post-scan evidence binding resolved the binder from the target repository trusted workspace. Organization code search confirms the binder exists in the protected central owner at scripts/ci/strix_evidence_binding.py, not in OriginWeave.

OriginWeave will not copy the central binder locally or blind-rerun the failed required job. Please keep acceptance on the owner repair: resolve the binder from the exact protected central checkout, preserve target-workspace evidence inputs, and prove the reusable workflow works for a consumer that intentionally has no local binder. No clean-scan result from this failed job is being promoted to required-check GREEN.