v14.3.0 — Jev as the contradiction checker

• MEMORY_CONTRADICTION_SCORER=jev scores the contradiction pairs of memory_answer with TypeSafe's Jev (System One API): one noul question per pair, one request per pass. Same accuracy as the Claude Haiku 4.5 scorer within noise (LongMemEval knowledge-update 35/78 vs 36/78); the median contradiction pass drops from 3.1 s to 1.9 s, and Jev billed $0.038 for all 78 questions. Needs TYPESAFE_API_KEY; TYPESAFE_BASE_URL and TYPESAFE_DEFAULT_MODEL are optional. The default scorer is unchanged.
• The Jev client rejects empty keys and keys with control characters without echoing them, retries 408/429/5xx and connection errors with backoff, and honours retry-after-ms / Retry-After up to 60 s.
• benchmarks/knowledge_update_eval.py records per-question contradiction-pass time and Jev token usage.

Also in this line of releases, 14.2.0 (same day): memory_answer answers with the latest value of a fact that changed over time (LongMemEval knowledge-update 12/78 → 35/78), Russian word forms in recall, and one stored UTC timestamp format (migration 034). See the 14.2.0 notes.

Install: pip install total-agent-memory==14.3.0 · npx -y total-agent-memory connect <ide> · brew install vbcherepanov/tap/total-memory · docker run -p 37737:37737 -v ~/.tam:/data ghcr.io/vbcherepanov/total-agent-memory:14.3.0