batch B audit: highest trust Gold surfaces

[Riverbraid]

Status

BATCH B AUDIT INVENTORY ITEM / PARTIALLY REMEDIATED

Scope

Batch B covers:

• Riverbraid-Safety-Gold
• Riverbraid-Refusal-Gold
• Riverbraid-Crypto-Gold
• Riverbraid-GPG-Gold
• Riverbraid-Manifest-Gold
• Riverbraid-Governance-Gold
• Riverbraid-Harness-Gold

Confirmed public entry strengths

• Safety, Refusal, Crypto, GPG, Manifest, and Harness have README role and evidence boundary language.
• Refusal and Harness link back to Evaluation Kit or Documentation.
• Manifest explicitly states GitHub Pages is disabled and verification is repository based, not deployment based.
• Most package scripts point to verify.mjs or feature-flow verification.

Current finding state

1. Riverbraid-Governance-Gold README and workflow target were remediated. The README was normalized and a bounded audit_final.js scaffold check was added. Execution evidence remains required in Riverbraid-Governance-Gold#1.
2. Riverbraid-Harness-Gold/runtime-binding.js and src/runtime-binding.js were patched to remove CI GPG skip behavior and shell-string GPG execution. Execution evidence remains required in Riverbraid-Harness-Gold#5.
3. Riverbraid-GPG-Gold still has a GPG-secret workflow surface. This is not inherently wrong, but secret validity and workflow evidence remain manual or evidence-gated.
4. Several workflows still use ubuntu-latest and tag-pinned actions rather than pinned runner/image/action digests or SHA-pinned actions.
5. Safety and Harness workflow-versus-package verifier path differences still need package-script-level classification.
6. Safety has a legacy script that prints a verified-style message. That should not be treated as verification evidence unless separately bounded.
7. Central community health coverage now exists in .github for SECURITY.md, CONTRIBUTING.md, and SUPPORT.md. Per-repo root coverage and root LICENSE coverage remain policy decisions and audit items.

Search-limited non-findings

Search did not return visible hits in Batch B for common private key markers, GitHub token markers, AWS secret marker, npm token marker, pull_request_target, permissions: write-all, secrets.GITHUB_TOKEN, pipe-to-shell patterns, or common nondeterminism markers.

This is search-limited and does not prove absence across history, settings, artifacts, dependencies, release assets, or unindexed surfaces.

Required follow-up

• Run execution evidence for Governance Gold and Harness Gold patched paths.
• Confirm whether GPG workflows still need fresh CI secrets.
• Classify workflow verifier path versus package verifier path for Safety and Harness.
• Decide whether to add per-repo SECURITY, CONTRIBUTING, SUPPORT, and LICENSE surfaces beyond central .github coverage.
• Preserve registry freshness lock until explicit registry succession gate exists.

Boundary

This issue records audit inventory only.
It does not claim Batch B is secure, complete, production ready, externally audited, or free of defects.
It does not mutate registry, verifier behavior, protocol, hash, seal, manifest, tag, or release state.

[Riverbraid]

Phase 4 proof-of-concept scope is now closed under the bounded claim PROOF_OF_CONCEPT_SCOPE_GREEN_SYNCED_LOCAL_AND_REMOTE_EXTRA_ARTIFACT_AUDITED_NOT_CERTIFICATION_NOT_PRODUCTION_NOT_EXTERNAL_AUDIT.

Closeout evidence sequence: 90 final active scope green; 91 synced local/remote; 93 local artifacts cleaned; 94 remote tracked extras audited green.

Remaining future work should continue under Post-Phase-4 roadmap issue #18.

Boundary: no registry, release, tag, protocol, hash, seal, manifest, or workflow mutation is made by this comment.