# Changelog

## Score

- CAI 64 → 64 (-0.0)
- Rubric changed (rubric-2026.10.5 → rubric-2026.11.2) — scores are not directly comparable.

## Lenses

- Code Health 100 → 100 (+0.0)
- Architecture 98 → 98 (-0.3)
- Maturity 43 → 43 (+0.0)
- Readiness 67 → 67 (+0.0)
- Security 95 → 95 (+0.0)

## Resolved (5)

- Documentation: no usage examples (README.md)
- Similar to the read methods, there are four subscription methods differing by scope (Stream/All) and mode (Live/FromCheckpoint). The return types also differ (`EventStoreSubscription` vs `EventStoreCatchUpSubscription`), forcing the caller to know the exact method signature to handle the return type correctly.
- The API exposes four distinct methods for reading events that differ only by stream scope (All vs Stream) and direction (Forward vs Backward). This creates a combinatorial explosion of methods (4) for a single logical operation (reading a slice of events).
- The PersistentSubscription class exposes 8 methods for acknowledging and failing events, covering combinations of single/multiple and object/ID. This is a classic API bloat issue where the same logical operation (ack/nack) is repeated for different input types.
- There is a duplication of intent between 'Raw' and 'Structured' metadata operations. While the data types differ (string vs object), the method naming pattern `setStreamMetadata` vs `setRawStreamMetadata` is inconsistent with standard library patterns which usually prefer a single method with an optional format flag or distinct domain objects. More critically, `getStreamMetadata` returns a `StreamMetadataResult` wrapper, while `getRawStreamMetadata` returns `RawStreamMetadataResult`. The parallelism is good, but the existence of two completely separate code paths for metadata (one typed, one raw string) without a unified interface or clear distinction in usage complexity is confusing.

## New (1)

- Documentation: no build, run or usage example (README.md)
