# Changelog

> **This is the PUBLIC form of this artifact. Findings are listed in full, but the details of SECURITY findings — which rule fired, in which file, on which line, and how to fix it — are deliberately withheld, and any secret-scanner results are excluded entirely. Where detail is absent here it was REMOVED FOR PUBLICATION; it is not missing from the analysis. The complete artifact is available from the repository owner.**

## Score

- CAI 62 → 61 (-0.6)
- Rubric changed (rubric-2026.10.5 → rubric-2026.11.2) — scores are not directly comparable.

## Lenses

- Code Health 73 → 73 (+0.2)
- Architecture 92 → 89 (-3.6)
- Maturity 70 → 75 (+5.4)
- Readiness 54 → 55 (+0.5)
- Security 59 → 53 (-5.7)
- Event Sourcing 100 → 100 (+0.0)
- Performance 83 → 100 (+16.9)

## Resolved (20)

- ADR has no Consequences section (docs/adr/0001-read-becomes-lazy-in-3-0.md)
- Documentation: no project overview (docs/superpowers/specs/2026-10-09-ca-certificate-fallback-design.md)
- Duplicate operation names for the same intent. Both methods perform appending events to a stream with identical signatures (StreamIdentifier/String, events, options).
- Duplicated block (27 lines × 2) (Sources/KurrentDB/Streams/Usecase/All/Streams.SubscribeAll.swift)
- High: security finding (details withheld)
- High: security finding (details withheld)
- Inconsistent naming for creation operations. Persistent subscriptions use `createPersistentSubscription`, while projections use specific types like `createOneTimeProjection`. This makes the API surface unpredictable.
- Inconsistent naming for listing subscriptions. `listAllPersistentSubscription` uses singular 'Subscription' while others use plural 'Subscriptions'. Also, `listAll` vs `list` is inconsistent with the pattern of `listPersistentSubscriptionsToAllStream`.
- Inconsistent naming for metadata operations. The client-level method uses `setStreamMetadata` while the fluent `Streams` object uses `setMetadata`. This breaks the pattern where `Streams` methods should mirror client methods or be clearly distinct.
- Inconsistent naming for metadata retrieval. The client uses `getStreamMetadata` while the fluent `Streams` object uses `getMetadata`.
- Inconsistent naming for stream deletion. The client uses `deleteStream` while the fluent `Streams` object uses `delete`.
- Inconsistent naming for stream tombstoning. The client uses `tombstoneStream` while the fluent `Streams` object uses `tombstone`.
- RPCError.rethrow (cognitive 16) (Sources/KurrentDB/Core/Error/KurrentError.swift)
- RPCError.rethrow (cyclomatic 17) (Sources/KurrentDB/Core/Error/KurrentError.swift)
- Repeated repair: Sources/KurrentDB/Core/Additions/Usecase/StreamStream.swift (Sources/KurrentDB/Core/Additions/Usecase/StreamStream.swift)
- Repeated repair: Sources/KurrentDB/Core/Additions/Usecase/StreamUnary.swift (Sources/KurrentDB/Core/Additions/Usecase/StreamUnary.swift)
- Repeated repair: Sources/KurrentDB/Core/Additions/Usecase/UnaryUnary.swift (Sources/KurrentDB/Core/Additions/Usecase/UnaryUnary.swift)
- Typo in property and method names. `cerificates` and `cerificate` are misspelled (should be `certificates` and `certificate`).
- generate-doc-snippets.extract (cognitive 20) (scripts/generate-doc-snippets.py)
- generate-doc-snippets.main (cognitive 19) (scripts/generate-doc-snippets.py)

## New (25)

- Classify ADR enforcement: ADR 0001 — `read()` delivers events lazily in 3.0 (docs/adr/0001-read-becomes-lazy-in-3-0.md)
- Duplicated block (10 lines × 2) (Sources/KurrentDB/Streams/Usecase/All/Streams.SubscribeAll.swift)
- Duplicated block (12 lines × 2) (Sources/KurrentDB/Streams/Usecase/All/Streams.SubscribeAll.swift)
- Duplicated block (5 lines × 2) (scripts/Sources/Tooling/BenchToBMF.swift)
- High IaC: WD-COMPOSE-0002 (server/docker-compose.single.yaml)
- High: security finding (details withheld)
- High: security finding (details withheld)
- High: security finding (details withheld)
- High: security finding (details withheld)
- KurrentError.streamsErrorDetails (cognitive 18) (Sources/KurrentDB/Core/Error/KurrentError+StreamsErrorDetails.swift)
- Low coverage: scripts/Sources/Tooling/Entry.swift (scripts/Sources/Tooling/Entry.swift)
- Medium IaC: WD-COMPOSE-0002 (server/docker-compose.single.yaml)
- Medium IaC: WD-COMPOSE-0002 (server/standalone/docker-compose.yaml)
- No assertions: caseUsesCurrentRevision (Tests/MockClientTests/CurrentRevisionStateRenameTests.swift)
- No assertions: missingStream (Tests/MockClientTests/DeleteWrongExpectedVersionTests.swift)
- No assertions: otherDomain (Tests/KurrentCoreTests/StreamsErrorDetailsMappingTests.swift)
- No assertions: rejectedReadIsFullyMapped (Tests/MockClientTests/ServerExceptionMappingTests.swift)
- No assertions: revisionConflictConstants (Tests/KurrentCoreTests/StreamsErrorDetailsMappingTests.swift)
- No assertions: testDeleteMissingStreamThrowsWrongExpectedVersion (Tests/StreamsTests/StreamsReadTests.swift)
- No assertions: testTombstoneMissingStreamThrowsWrongExpectedVersion (Tests/StreamsTests/StreamsReadTests.swift)
- …and 5 more

## Changes since last survey

- 70 commits — 36 feature/other, 34 fixes

## By area

- (root) — 21 commits
- .github/workflows — 18 commits
- Sources/KurrentDB — 18 commits
- Benchmarks/OfflineBenchmarks — 5 commits
- Tests/MockClientTests — 2 commits
- scripts/check-release-readiness.py — 2 commits
- Sources/_GRPCProtobufGenerated — 1 commit
- docs/index.html — 1 commit
- scripts/Sources — 1 commit
- server/standalone — 1 commit

## Notable commits

- fix: [FIX] accept only a bullet ending with (X.Y.Z) as a patch release's What's new entry, not any mention of the version
- fix: [FIX] build once per Swift version and share it with that version's KurrentDB test jobs, and upload the coverage Codecov has not received since March
- fix: [FIX] cancel the held call only once it waits in the wrapped transport, so the test covers cancelling a call in progress rather than one that never started
- fix: [FIX] check the markdown pass too before publishing it: its report must hold every benchmark the JMH export holds, since package-benchmark lists failures only in the export
- fix: [FIX] collect coverage through LLVM_PROFILE_FILE instead of swift test --enable-code-coverage, whose JSON export added about 140 s per run on Swift 6.4, and keep SwiftPM's git mirrors in the artifact
- fix: [FIX] convert coverage with llvm-cov instead of swift-coverage-action, which finds no .xctest in Swift 6.4's per-target test libraries
- fix: [FIX] copy the request metadata only for a call that carries grpc-timeout, so the other calls pay no copy, and say that the gRPC baseline runs behind the same erased transport
- fix: [FIX] decode the v2 Streams error details, so an AppendSession to a tombstoned stream throws resourceDeleted instead of a grpcError that is retried as a node failure
- fix: [FIX] deploy GitHub Pages with a build version unique to each run, so a release tag on main's commit replaces main's site
- fix: [FIX] describe the erased transport's forwarding as grpc-swift uses it: the method config and a stream on every RPC, a connect on every connection, nothing per message
- fix: [FIX] end a subscription to a tombstoned stream with resourceDeleted instead of delivering $streamDeleted and a raw RPCError
- fix: [FIX] fail the Benchmarks job when a benchmark throws, instead of uploading a report without it: package-benchmark only lists the failure and exits 0
- fix: [FIX] fail the coverage step when either test run left no profile, and name profiles %p-%m so a reused pid merges instead of truncating
- fix: [FIX] give each build artifact download 5 minutes and one retry, and pack Swift 6.4's build without its intermediate files
- fix: [FIX] go back to actions/deploy-pages: the Pages API rejects a build version that is not a commit SHA with 404
- fix: [FIX] keep a record's custom metadata and properties instead of dropping them: parse customMetadata with JSONDecoder and encode every property as JSON
- fix: [FIX] keep the CA key out of the standalone node, regenerate incomplete certificate pairs, and note the port and password encoding in the connection instructions
- fix: [FIX] list the largest .build directories in the pack step: du skips a directory it already counted under .build, so it printed only the total
- fix: [FIX] make the in-process server take each request whole and check its message count, keep the grpc-timeout header from it so only the client's deadline timer runs, and check the lazy read before measuring
- fix: [FIX] map a buffered read the server rejects, such as access denied, the way a lazy read maps it
- …and 50 more
