Refactor non-RAG assurance scripts and speed focused checks

data/repo-awareness-snapshot.json

@@ -2416,6 +2416,11 @@
         "section": "decisions",
         "catalogued": false
       },
+      {
+        "path": "docs/audit/codescene-non-rag-remediation-2026-10-09.md",
+        "section": "audit",
+        "catalogued": false
+      },
       {
         "path": "docs/superpowers/rag-upgrade/local/specs/2026-08-21-trusted-admin-document-ingestion-design.md",
         "section": "superpowers",
@@ -19587,6 +19592,14 @@
         "outcome": "APPROVE. The remaining bottom band clipped live result content above the hidden dock because both phone application owners used viewport-sized fixed roots, a physical-iOS paint path that can disagree with correct DOM geometry. Both owners now share a bounded in-flow dynamic-viewport shell; hidden reserve remains zero, the last viewport pixel remains content-owned, and viewport resize preserves the reading offset. No P0-P3 finding remains. Highest residual risk is physical-device iOS compositing beyond desktop WebKit emulation.",
         "checks": "`verify:cheap` PASS (393 files; 3519 passed / 2 skipped); focused Therapy and dashboard production WebKit PASS; `verify:ui` PASS (314/314); `verify:pr-local` PASS including production build/client-secret scan and 36-case offline RAG fixtures; no live provider-backed checks."
       },
+      {
+        "date": "2026-10-10",
+        "ref": "codescene-non-rag-current-20261010",
+        "head": "a9189921bcc182950e04ee41cb8ceca42f2adb2a",
+        "scope": "codescene-non-rag-current-20261010 independent governance integration dirty diff",
+        "outcome": "No governance blocker found; independent source-only dirty-diff review; current staffing declaration preserved",
+        "checks": "Input identities and limitations: external integration-20261010/independent-governance-review.md under psychsift-codescene-remediation-20261009; twelve reviewed source hashes in report; no provider checks or behavioral-equivalence claim"
+      },
       {
         "date": "2026-07-30",
         "ref": "codex/review-pr1429",

docs/audit/codescene-non-rag-remediation-2026-10-09.md

@@ -0,0 +1,183 @@
+# CodeScene maintainability remediation: non-RAG phase
+
+This work follows the full-file CodeScene audit of `BigSimmo/PsychSift` at
+`7f3303c009df84a1e4e2b3e54ea6e4b32f3bf147`. Audit warnings describe maintainability;
+they do not establish functional defects. Individual file scores are not
+CodeScene's official repository score.
+
+The full finding backlog, plan, source hashes and verification evidence are kept
+in the local task workspace outside this repository. Task identity:
+`psychsift-codescene-remediation-20261009`. This document explains the maintained
+source changes; the task checkpoint records their current acceptance state.
+
+## Scope and boundaries
+
+RAG remediation is paused at Josh's request. This includes retrieval and ranking,
+answer generation and verification, extraction/chunking/index/embedding producers,
+related evaluation and protected tests. Source-rendering components are treated
+conservatively as adjacent to that deferred work.
+
+Local maintainability edits do not authorize provider traffic, production access,
+publication or deployment. Existing unrelated changes in the original checkout
+must be preserved. The remediation uses an isolated checkout with one writer per
+owned file and one controller for shared verification.
+
+## Implemented work packages
+
+- `scripts/generate-design-system-adoption.mjs`: separates visual-provenance and
+  adoption-manifest checks into ordered validation phases. Each phase keeps the
+  existing error text, rejection conditions, ordering and early-return behavior.
+  PNG header, structure and decoded-data checks, candidate source-binding phases
+  and class-expression resolution have distinct helpers. Invalid-image refusal,
+  exact suite/path checks, lexical shadowing, unknown values and cycles remain
+  part of their original contracts.
+- `scripts/design-system-contract-utils.mjs`: separates class-expression AST
+  resolution from class-root and token-contract reporting. Unknown bindings,
+  shadowing, cycles, alternative order and the edge-conflict early stop retain
+  their existing meaning. Internal raw-scale checks receive a cohesive options
+  object.
+- `scripts/ledger-inbox.mjs`: separates request validation, payload construction,
+  request application, cancellation, reconciliation and CLI dispatch. Receipt
+  identity, fingerprints, dry-run behavior, locking and existing remote-check
+  semantics must remain unchanged.
+- `scripts/lib/tenancy-scan.mjs`: the separate privacy-sensitive work package
+  simplifies AST evidence collection without changing tenancy rules. Acceptance
+  requires focused offline proof and independent privacy/schema review.
+- `scripts/ci-change-scope.mjs`: separates owner-manifest path checks, direct and
+  derived change classification, Lighthouse budget parsing and self-test groups.
+  Path containment, symlink refusal, policy patterns, output order and the full
+  set of routing assertions are preserved.
+- `scripts/check-codex-cloud-setup.mjs`: separates TOML/MCP parsing, inventory
+  checks, setup phases and runtime validation phases. Credential restrictions,
+  defaults, command requirements, diagnostic order and dependency checks remain
+  intact. This source refactor does not change or provision any environment.
+
+## Verification and completion
+
+Verification uses existing focused behavioral tests and original-versus-candidate
+output comparisons where appropriate. Required local domain checks and independent
+reviews apply to stable source hashes. CodeScene re-reviews record remaining
+warnings; helper extraction alone does not establish whole-file completion.
+
+Do not regenerate adoption manifests, visual baselines, gate figures, debt
+baselines or snapshots to make a check pass. Existing failures must be diagnosed
+and distinguished from regressions on the audited base.
+
+The programme remains incomplete while non-deferred findings lack implementation
+and acceptance evidence or a reviewed, finding-specific reason to retain their
+structure. Deferred, blocked, pending, failed and unscored work must remain visible
+in the same task checkpoint. A local receipt export does not imply ledger sync,
+CI success, merge, deployment or release acceptance.
+
+## Restored checkpoint — 9 October 2026
+
+The six saved script refactors were restored byte for byte at their original base commit. Their source hashes are unchanged, so the saved CodeScene analysis and independent source reviews remain historical evidence for exactly those versions.
+
+Fresh local verification completed on the restored checkout:
+
+- Adoption: 58 tests passed; the check-only repository contract passed for 55 components and 183 roots.
+- Adoption parity: 23 synthetic PNG cases, 28 AST cases and 1,592 production files compared without differences.
+- Cloud: all 35 tests passed, preserving the existing assertions and timeout limits.
+- CI: all 129 tests across three suites passed, including the complete 384-case Bash matrix.
+- Installed dependencies: 709 package locations and 48,240 files matched the saved locked tree.
+
+The earlier interrupted and failed runs remain in the external evidence history. No additional source refactor or test relaxation was needed to close these verification gaps. These results accept the selected behavior-preserving batch locally; they do not establish that every CodeScene warning is fixed, nor hosted CI, publication, deployment or clinical readiness.
+
+The remaining non-RAG inventory contains 1,272 warning-bearing files: 1,253 queued, 13 requiring ownership review and six partially refactored files with remaining findings. There are 4,904 original unresolved finding instances. The broad RAG deferral was widened to include 15 adjacent source/evaluation test files. Every future queued package still needs current-source and caller inspection.
+
+The original checkout has advanced independently since the audit. The restored batch is isolated at its audited base and has not been integrated into that newer checkout. The canonical checkpoint, complete source copies, patch, recovery report, local receipt and hashes are kept in the existing external audit workspace.
+
+## Local testing speed improvements — 9 October 2026
+
+The same restored checkout now contains three focused performance improvements:
+
+- Adoption manifest builds index imports and reuse source analysis within each operation. Later builds observe changed source and configuration.
+- Design-contract helpers privately reuse at most four ASTs with identical filenames, source text and parser settings. The text-role scanner skips sources and tokens that cannot match its unchanged rule, retaining CSS errors and exact finding locations.
+- The existing policy-only Cloud test harness skips runtime bootstrap and runtime diagnostics. Ordinary setup retains its runtime and credential checks; policy refusals, atomic replacement and failure status remain intact.
+
+Six complete local suites passed: 60 adoption/visual-baseline tests, 73 design-helper/status tests, 35 unchanged Cloud tests and four new policy-harness tests. The complete design-system contract gate passed, including figures, token divergence, adoption and design-sync checks. Baseline/current comparison covered 2,351 tracked source files and 25,861 helper comparisons without differences. Separate fixtures cover cache keys, eviction, changed inputs and scanner errors.
+
+Controlled timings and their limitations are recorded in the external `testing-speed-20261009` evidence directory under the existing remediation workspace. Benchmarks use fresh processes and disable result reuse. The existing content-pinned test receipts were also checked separately; a reused result is labelled as reused proof.
+
+The adoption and contract utility hashes changed in this phase, so their earlier CodeScene reviews are historical. Fresh reviews of these versions are saved separately. The remaining maintainability programme and broader RAG deferral are unchanged. These local performance checks do not establish full-suite, hosted CI or release acceptance, and the prepared changes remain isolated from the newer primary checkout.
+
+## PR #3512 integration repair - 13 October 2026
+
+The published head `29fa506084d57ad72d12427728fc4d3aa8939bca` was merged with main
+`08d8fb3eb10c039c507c3a13ebd43b7761b6ed87` in an isolated repair checkout. The only
+content conflict was the CI scope self-test: the instruction-continuation assertion
+now remains in the workflow/policy group, before the extracted dependency/container
+group. Main's instruction routing, starter-sharing tenancy inventory and 224-route
+expectation are retained. The original refactors and immutable review record are
+preserved.
+
+Fresh provider-free evidence: the CI scope self-test, static Cloud contracts and
+owner-scope scanner passed. Ten focused suites passed all 230 tests, covering CI
+scope, tenancy, adoption, design utilities, Cloud setup/policy and ledger handling.
+
+Those integration runs were fresh executions on the source later committed as
+`66d21b5f482007cb47092afdea8e6ff25f85f7d4`; they were not reused receipts. The
+decisive console output was:
+
+```text
+CI change scope self-test passed.
+[Codex Cloud Check] PASS: static Cloud contracts match.
+✓ owner-scope phase 1: 125 src/app/api files clean against 56 owner-scoped tables.
+✓ owner-scope phase 2: 169 direct, 32 user-keyed, 29 derived-tier, 2 untiered-table and 1 dynamic-table-dispatch queries scoped on their chain or declared; 8 versioned-RPC call sites, all literal.
+Test Files  10 passed (10)
+Tests  230 passed (230)
+```
+
+The Vitest run finished at `2026-10-13T06:16:18.233Z`, with result reuse disabled
+(`GATE_RECEIPTS=refresh`). Its ten explicit owners were CI scope, API tenancy,
+design-system adoption, design-contract utilities, Cloud setup, Cloud policy
+harness, ledger inbox, remote ledger interlock, ledger cancellation and visual
+baseline adoption. The later snapshot-only commit `d77d16c2e168d4360b3fd4d597eb4252c695d895`
+did not rerun those suites; its fresh snapshot check reported
+`[repo-awareness] in step with data/repo-awareness-snapshot.json (307 pages, 576 documents, 2779 reviews)`.
+
+Dependency provisioning initially waited behind another worktree's exclusive lease;
+normal coordinator admission later succeeded and an exact-lock local donor was
+reused. Historical results above remain historical. Hosted CI and release readiness
+still require their own evidence. No merge, deployment, live canary or provider call
+is authorized by this repair.
+
+## Unresolved review findings - 13 October 2026
+
+The shell-comment parser, add-request domain mismatch and stale-lock removal race
+were inherited from the integration base; they were still present, so green CI
+did not dispose of those comments. The follow-up preserves the existing refusal
+paths while checking quoted/escaped hashes, rejecting add priorities/types the
+writer cannot accept, and atomically serializing lock acquisition and stale
+recovery. Tests use isolated lock fixtures, never the real reconciliation lock.
+An abandoned admission claim fails closed and requires reviewed recovery; this
+change does not authorize deleting a live or unclear claim.
+
+The policy-only shortcut now requires both an explicit `--policy-only-test-harness`
+invocation and `NODE_ENV=test`; an inherited stop flag alone refuses before
+runtime/configuration work. Ordinary setup without the stop flag is unchanged.
+
+Fresh direct checks of the follow-up source reported:
+
+```text
+[Codex Cloud Check] PASS: static Cloud contracts match.
+ledger inbox self-test passed.
+```
+
+The initial follow-up Vitest attempt did not run: normal admission returned
+`DATABASE_HEAVY_RUN_ADMISSION_BUSY` while another dependency task held capacity.
+After normal release, the final review-fix source passed five focused owners with
+result reuse disabled (`GATE_RECEIPTS=refresh`): ledger inbox, remote ledger
+interlock, ledger cancellation, Cloud setup and the policy-only harness.
+
+```text
+Test Files  5 passed (5)
+Tests  109 passed (109)
+```
+
+These tests include quoted/escaped hashes and nested shell substitutions, invalid
+add domains and accepted writer domains, the deterministic stale-recoverer race,
+an abandoned admission claim, and inherited/ordinary-runtime harness refusal.
+No test or lock bypass was used. The earlier 230 passing tests remain historical
+evidence for the integration source. Local passing checks and unresolved review
+threads do not establish merge or release clearance.

docs/branch-review-records/ab93745778dff4eab8369ee651e62e0b1d488603e6f1b84a31a20751433515d6.record.md

@@ -0,0 +1 @@
+| 2026-10-10 | codescene-non-rag-current-20261010 | a9189921bcc182950e04ee41cb8ceca42f2adb2a | codescene-non-rag-current-20261010 independent governance integration dirty diff | No governance blocker found; independent source-only dirty-diff review; current staffing declaration preserved | Input identities and limitations: external integration-20261010/independent-governance-review.md under psychsift-codescene-remediation-20261009; twelve reviewed source hashes in report; no provider checks or behavioral-equivalence claim |