ty-everett · Sep 23, 2026 · Sep 23, 2026 · Sep 23, 2026 · Sep 23, 2026 · Sep 23, 2026
@@ -66,6 +66,14 @@ updates:
       - '*'
     multi-ecosystem-group: monthly-stack-maintenance
     rebase-strategy: auto
+    # The npm updater currently fetches unrelated ancestor pnpm support files
+    # and mistakes these standalone package-lock projects for sub-workspaces.
+    # Exclude only those two ancestor files here; the root entry owns them.
+    exclude-paths:
+      - '../../pnpm-lock.yaml'
+      - '../../pnpm-workspace.yaml'
+      - '../../../pnpm-lock.yaml'
+      - '../../../pnpm-workspace.yaml'
     cooldown:
       default-days: 7
     ignore:

@@ -52,7 +52,8 @@ jobs:
 
       - uses: astral-sh/setup-uv@c771a70e6277c0a99b617c7a806ffedaca235ff9 # v9.0.0
         with:
-          version-file: tools/codegen/pyproject.toml
+          # Keep CI reproducible while pyproject permits Dependabot's uv resolver.
+          version: '0.11.32'
           python-version: '3.12'
           enable-cache: true
           cache-dependency-glob: tools/codegen/uv.lock

@@ -20,7 +20,12 @@ pnpm --filter @bsv/conformance-runner-ts test
 
 The structural runner's `build` command verifies its JavaScript syntax.
 The TypeScript runner is linted and checked in strict mode before executing its
-tests. `test` executes the required vector suite. `validate` checks vector and
+tests. Its workspace dependencies include wallet-toolbox because the wallet
+dispatchers typecheck against its built declarations. For a clean targeted
+build, run `pnpm --filter '@bsv/conformance-runner-ts...' --if-present build`
+before the TypeScript runner's typecheck; the declared dependency graph builds
+those prerequisites even when the changed package is air-gap or CHIRP.
+`test` executes the required vector suite. `validate` checks vector and
 implementation metadata without executing the cases. Generated vectors live
 under `conformance/generated`; edit their source specifications and run the
 owned generator rather than editing generated output.

@@ -13,6 +13,7 @@
     "@bsv/air-gap": "workspace:^",
     "@bsv/chirp": "workspace:^",
     "@bsv/sdk": "workspace:^",
+    "@bsv/wallet-toolbox": "workspace:^",
     "@jest/globals": "^30.4.1",
     "@types/node": "^26.1.2",
     "@typescript/native": "npm:typescript@7.0.2",

@@ -4,8 +4,8 @@ title: Wallet UTXO Lifecycle
 kind: meta
 domain: wallet
 version: 'n/a'
-last_updated: '2026-08-13'
-last_verified: '2026-08-13'
+last_updated: '2026-09-23'
+last_verified: '2026-09-23'
 review_cadence_days: 30
 status: stable
 tags: ['architecture', 'BRC-100', 'wallet', 'utxo', 'storage']
@@ -24,6 +24,12 @@ It is written against two implementations, [`@bsv/wallet-toolbox`](../packages/w
 against it, and the places where either one deviates are collected in
 [Implementation differences](#implementation-differences).
 
+The TypeScript paths and lifecycle were rechecked against TS Stack main
+`57d72e24a565d090bd605cf9767785c51ea4f179` on 2026-09-23. The Go diagrams and
+comparisons below preserve the **2026-08-13 review snapshot**; they are historical
+findings, not claims about the latest Go release. Recheck the linked Go source
+before using a difference as a current cross-implementation defect.
+
 ## How to read the diagrams
 
 Each lane is a layer, and time runs downward. The lanes below appear in every diagram on
@@ -105,14 +111,14 @@ sequenceDiagram
     end
 ```
 
-The write transaction opens at `storage/methods/createAction.ts:196` and every call from
-`insertTransaction` onward is inside it. The transaction row is born `unsigned`
-(`createAction.ts:690`). If anything downstream throws, the cleanup path drives it to
-`failed` (`:309`) and records a forensic row (`:312`) rather than deleting evidence.
+The write transaction in `storage/methods/createAction.ts` contains
+`insertTransaction` and its subsequent writes. The transaction row is born
+`unsigned`. If construction fails after recording the plan, the cleanup path
+drives it to `failed` and records a forensic row rather than deleting evidence.
 
-`markChangeInputsSpent` (`:1398`) is the moment funding becomes exclusive: it flips the
+`markChangeInputsSpent` is the moment funding becomes exclusive: it flips the
 selected change outputs to `{spendable: false, spentBy: transactionId}` under the row
-locks taken by `findFundingOutputsForUpdate` (`:1376`).
+locks taken by `findFundingOutputsForUpdate`.
 
 ### Storage call ledger
 
@@ -166,7 +172,7 @@ sequenceDiagram
 
 The `pendingSignActions` cache is process memory on the `Wallet` instance. A `reference`
 issued by one process cannot be signed by another, and cannot survive a restart —
-`Wallet.ts:1056` throws `WERR_NOT_IMPLEMENTED` rather than attempting recovery. Go stores
+`signer/methods/signAction.ts` throws `WERR_NOT_IMPLEMENTED` rather than attempting recovery. Go stores
 these in a pluggable repository instead; see [difference 6](#implementation-differences).
 
 ### processAction — commit and broadcast
@@ -429,6 +435,14 @@ custom instructions and tags.
 
 ### abortAction
 
+The diagram below describes ordinary actions. BRC-177 expiring `noSend` actions
+have a separate durable lifecycle: aborting a released signed action requests
+revocation instead of immediately freeing its inputs. `TaskNoSendExpiry` uses
+positive chain and UTXO evidence to arbitrate target-versus-reclaim races;
+reclaiming, reclaimed, broadcast, target-won and conflicted states retain their
+specific guards. See `StorageProvider.abortAction` and `TaskNoSendExpiry` before
+operating on expiring actions.
+
 ```mermaid
 sequenceDiagram
     autonumber
@@ -504,11 +518,14 @@ sequenceDiagram
     end
 ```
 
-Nineteen tasks ship in `src/monitor/tasks/`. Beyond those above, `TaskReorg` and
+The task classes in `src/monitor/tasks/` are selected by the Monitor profile;
+not every class is registered by default. Beyond those above, `TaskReorg` and
 `TaskNewHeader` handle chain reorganisation, `TaskCheckNoSends` settles `nosend`
 transactions, `TaskUnFail` retries operator-flagged failures, `TaskArcSSE` consumes
 broadcaster push events, `TaskPurge` and `TaskCleanupActionBatches` reclaim storage, and
-`TaskSyncWhenIdle` replicates to backup stores.
+`TaskSyncWhenIdle` can be registered to replicate to backup stores. The default
+and multi-user profiles include `TaskNoSendExpiry`, `TaskReviewProvenTxs` and
+`TaskReconcilePendingTransactions`; `TaskMineBlock` is mock-chain-only.
 
 ## The Go implementation
 
@@ -771,8 +788,10 @@ list-outputs special operations.
 
 ### 11. Background convergence uses different mechanisms
 
-TypeScript ships nineteen registered Monitor tasks; Go registers four. That comparison is
-misleading on its own, because Go moves much of the same work off the scheduler:
+The TypeScript default and multi-user profiles each schedule sixteen tasks plus
+two housekeeping tasks, with a mock-only miner added for mock chains. The archived
+Go review found four scheduled tasks. Counts alone do not describe convergence:
+that Go implementation moved much of the work off the scheduler:
 
 - **Event consumers.** `pkg/monitor` runs an SSE broadcast-event pipeline with a persisted
   replay cursor (`arcade_sse_last_event_id`) plus reorg and new-tip consumers. Reorg

@@ -3,8 +3,8 @@ id: infra-chaintracks-server
 title: 'Chaintracks Server'
 kind: infra
 version: '1.2.0'
-last_updated: '2026-08-12'
-last_verified: '2026-08-12'
+last_updated: '2026-09-23'
+last_verified: '2026-09-23'
 review_cadence_days: 30
 status: stable
 tags: [chaintracks, block-headers, spv, merkle, infrastructure]
@@ -175,6 +175,16 @@ them as a fallback until the first new generation is complete. Roll back the
 service image without deleting this root. Older releases continue to see their
 flat files; the new content-addressed and generation directories are additive.
 
+## Verification scope
+
+Rechecked on 2026-09-23 against TS Stack main
+`57d72e24a565d090bd605cf9767785c51ea4f179`: v1/v2 route handlers,
+`server.ts` readiness and upstream/worker limits, the edge policies, and
+`BulkHeaderSnapshotPublisher` generation retention and atomic current pointer.
+These are source contracts. Operators must separately verify the deployed image,
+ready endpoints, durable storage and public routes during each rollout; a source
+version or passing unit test does not prove a live service has been upgraded.
+
 ## When to deploy this
 
 - Running `@bsv/wallet-toolbox`-based wallets in production (the toolbox calls Chaintracks for SPV)

@@ -74,6 +74,28 @@ Review the 40 exact run links, classification threshold, sample summaries,
 workflow or runner changes, and the stated median/p95 budget. Never loosen a
 budget solely to make a red trend green.
 
+## Sampling and investigation
+
+The collector searches up to ten pages of successful PR runs, deduplicates source
+heads, and paginates each run's complete job list. It retains 20 full-scope and
+20 targeted runs, including job and step timings. When history cannot supply the
+required sample, it writes the partial report before failing; incomplete samples
+cannot establish a new baseline. API requests and pagination are bounded.
+
+The September review found that the July baseline predates the optical-codec
+mutation target and later QA additions. In the recovered 40-run sample, full-scope
+median/p95 was 2,007/2,254 seconds and targeted median/p95 was 566/833 seconds.
+The optical-codec mutation job dominated full runs (median 1,378 seconds), while
+wallet coverage dominated the longer targeted runs. Those observations do not by
+themselves justify raising the budget. Keep the historical baseline until a
+reviewed workload comparison and measurements justify a replacement.
+
+The duplicate-tracking codec regression now bounds its fixture's search and
+avoids constructing a successful Jest matcher for every redundant frame. It
+still sends more than the actual tracking limit, checks every acceptance result,
+repeats the final frame and verifies exact recovered bytes. Mutation and property
+coverage, payload sizes, test counts and timeout budgets remain governed.
+
 ## Complete release acceptance
 
 Dispatch `CI` manually on the reviewed main commit to select the entire governed

@@ -2,9 +2,9 @@
 id: dependency-release-policy
 title: 'Dependency and Release Policy'
 kind: reference
-version: '1.3.1'
-last_updated: '2026-09-15'
-last_verified: '2026-09-15'
+version: '1.3.2'
+last_updated: '2026-09-23'
+last_verified: '2026-09-23'
 review_cadence_days: 30
 status: stable
 tags: [reference, dependencies, security, releases]
@@ -61,6 +61,19 @@ Security updates are grouped only within a package-manager ecosystem and are
 never delayed into the monthly cross-ecosystem version update. First-party
 `@bsv/*` versions remain owned by the release graph.
 
+Standalone infrastructure uses npm `package-lock.json` files outside the pnpm
+workspace. Its Dependabot entry excludes only the unrelated ancestor
+`pnpm-lock.yaml` and `pnpm-workspace.yaml` support files, avoiding the updater's
+sub-workspace misclassification; the root entry continues to own both files.
+Every infrastructure manifest and npm lockfile remains monitored. Remove this
+workaround when Dependabot respects standalone npm boundaries beneath a pnpm root.
+
+The Python code generator accepts uv >=0.11.32 so Dependabot can resolve its
+locked dependency graph with the uv version supplied by GitHub. The codegen
+workflow retains an explicit uv 0.11.32 pin, Python 3.12, `uv run --locked`, and
+byte-for-byte generated-output verification. A resolver proposal does not
+silently change the CI toolchain or authorize different generated types.
+
 Major changes that alter a runtime, compiler, or persisted-data contract are
 held from the routine monthly PR until their focused migration is ready:
 

@@ -410,11 +410,23 @@ describe('AirGapDecoder', () => {
     const dec = new AirGapDecoder()
     expect(dec.accept(e.partAt(0)).have).toBe(1)
     const redundant: number[] = []
-    for (let seq = 2; redundant.length < MAX_TRACKED_SEQS + 1; seq++) {
+    // Bound fixture generation too: a broken mapping must fail this test,
+    // not make the fixture search forever during mutation testing.
+    for (
+      let seq = 2;
+      seq < MAX_TRACKED_SEQS * 16 && redundant.length < MAX_TRACKED_SEQS + 1;
+      seq++
+    ) {
       const blocks = blocksForPart(seq, 2)
       if (blocks.length === 1 && blocks[0] === 0) redundant.push(seq)
     }
-    for (const seq of redundant) expect(dec.accept(e.partAt(seq)).ok).toBe(true)
+    expect(redundant).toHaveLength(MAX_TRACKED_SEQS + 1)
+    for (const seq of redundant) {
+      const accepted = dec.accept(e.partAt(seq)).ok
+      // Avoid allocating thousands of successful Jest matchers in every mutant.
+      // Keep the same literal-true assertion, with full diagnostics on failure.
+      if (accepted !== true) expect({ seq, accepted }).toEqual({ seq, accepted: true })
+    }
     // The tracker is full; new and repeated sequence numbers are simply
     // re-processed as redundancy instead of being remembered, and the honest
     // part still completes the message.