github AxDSan/mnemosyne v2.7.0

6 hours ago

What's Changed

  • fix: consolidate_fact PRIMARY KEY collision (silent data loss on long content) by @kohai-ut in #83
  • fix(E3.a.3): cross-tier recall dedup — collapse (summary, source) duplicates by @kohai-ut in #88
  • fix: cascade-clean orphaned vec_episodes rows on import_from_dict force-overwrite by @kohai-ut in #87
  • fix(E7+E8+E9): gate harness bypasses behind MNEMOSYNE_BENCHMARK_PURE_RECALL by @kohai-ut in #90
  • fix: race-pattern fix on VeracityConsolidator sibling methods (E2.a.6) by @kohai-ut in #85
  • E5.a: rewire polyphonic vector voice to dense memory_embeddings store by @kohai-ut in #80
  • E2: remember_batch enrichment parity with remember() (unblocks Arm C) by @kohai-ut in #82
  • fix(C30+C31+C32): pre-experiment telemetry + env-parser cleanup (stacks on #90) by @kohai-ut in #91
  • refactor: DRY consolidate_fact onto _serialized_write helper by @kohai-ut in #86
  • fix: pre-experiment fidelity bundle (E4.a.1 + E2.a.10 + C29) by @kohai-ut in #89
  • docs+harness: benchmark infrastructure + merge-regression recovery by @kohai-ut in #92
  • feat: 9 A/B ablation toggles for the BEAM-recovery experiment (Gap B) by @kohai-ut in #94
  • feat: paired-outcomes JSONL + linear voice_scores parity (Gaps E + G) by @kohai-ut in #95
  • docs: output-file schemas + analysis recipes for AI-assisted interpretation by @kohai-ut in #96
  • feat: per-question voice attribution + BEAM_OPTIMIZATIONS env-parser migration (pre-test gap closure) by @kohai-ut in #97

Full Changelog: v2.5.0...v2.7.0

Don't miss a new mnemosyne release

NewReleases is sending notifications on new releases.