Ten deep dives on temporal validity, stale-fact errors, code evolution, and local context compression — grounded in our paper series and arXiv:2606.26511.
Explore the research portfolio →Featured
MemStrata is the local context layer for AI coding tools: median 42% fewer input tokens vs naive RAG, money-back on Pro, and research-backed temporal memory so cheaper context is still correct context.
Apples-to-apples local 7B evaluation: MemStrata’s deterministic supersession hits 1.000 on MemArch supersession/poisoning/TEMPO axes where flat RAG and several agent-memory systems still serve stale values.
Oracle experiments on synthetic world-knowledge facts hit 1.000 accuracy / 0.000 stale. The mechanism is perfect; remaining error is extraction. Here is what Paper 5 proves and what remains.
You cannot supersede logic you cannot extract. Paper 4’s structural-staleness approach marks scope-level code changes and warns at retrieval — without faking understanding or breaking the deterministic moat.
Across 707 SWE-bench fixes, deterministic temporal memory natively covers ~19% of evolution. Contract extraction lifts to ~24%. The remaining ~76% is logic — and overclaiming extraction is a research risk.
On 130 marker-free scenarios extracted from SWE-bench Lite/Verified buggy→fixed histories, MemStrata temporal_v6 hits 0.908 accuracy vs ~0.57 RAG — with stale-fact-error 0.023 vs 0.262.
If your evolving-knowledge benchmark marks outdated chunks as “[OUTDATED]”, baselines win by reading the label. MemStrata’s marker-free protocol removes that contamination — and scores drop up to 14 points when markers are removed.
Calibration results show cosine similarity cannot separate contradicted facts from duplicated ones. Contradictions are more embedding-similar than rephrases. Here is the data and the design implication.
MemStrata maintains temporal validity with deterministic (subject, relation, object) supersession in a bi-temporal ledger. No similarity threshold. No LLM on the read path. ~2.1s retrieval latency.
When knowledge evolves, RAG retrieves both stale and current facts with near-identical embeddings. Learn why stale-fact errors are structural — not a tuning problem — and what temporal validity changes.
Install MemStrata locally and see token savings + temporal memory on your own codebase.
Download free