v0.2.3

Documentation

• Knowledge-graph / graph-memory section in all seven READMEs, covering
  entities, typed relations, evidenced edges, triples, entity resolution and
  multi-hop traversal — including what sage-wiki does not do (temporal
  validity fields are stored but unqueried; there is no automatic
  contradiction detection).
• README benchmark and eval summaries now report real measurements: the
  memory benchmarks (LOCOMO 92.0%, LongMemEval 93.3%, BEAM 0.691 with gpt-5
  judging) and the quality/perf eval (87.4% median across 10 real wikis).

Fixed

• Both checkpoint readers retry transient Windows failures, via one shared
  helper. Fixing only the compile-state reader moved the CI failure to the
  batch-state reader — same test, same error, different path — so both now go
  through readStateFileRetrying and a test asserts it.
• The transient-error predicate now matches the message Windows actually
  emits. ERROR_SHARING_VIOLATION reads "The process cannot access the file
  because it is being used by another process" — containing neither "sharing
  violation" nor "access is denied", and Go does not map it to
  fs.ErrPermission. The predicate checked only those, so the most common
  Windows contention error fell through as fatal in both the rename retry and
  the new read retry.
• Postgres migration tests no longer depend on a pre-migrated database.
  TestMigrationV3/V4/V5 clone the base database and then simulate a
  downgrade, but did so before creating any schema — so they only passed
  against a long-lived developer database that had accumulated the schema
  from earlier runs, and failed on every fresh CI Postgres with
  relation "relations" does not exist. They now create the schema first.
• CI: the Postgres job now creates the pgvector extension explicitly. It
  had never passed since being added — sage-wiki refuses to CREATE EXTENSION
  itself by design, and the image's initdb hook was not reliably creating it,
  so every pg test failed with "relation does not exist".
• Compile-state reads now retry transient Windows file-sharing failures.
  The write half of this contract already retried (writeFileAtomicUnique via
  isTransientRenameError), but reads did not — so a concurrent writer
  holding the handle surfaced as "The process cannot access the file because
  it is being used by another process" and aborted the caller outright. Same
  family as the v0.2.2 manifest-lock fix, different code path; present in
  v0.2.2 and earlier, not a regression.