Resources

Research

Paper-linked notes and benchmark results from the HeurChain team. Every number is real and every scope caveat is stated; drafts stay out of the index, sitemap and feed until verified. Tags are for discovery.

Six memory systems compared on LoCoMo DEV

· HeurChain team

Six systems, two runs each, 543 LoCoMo DEV questions: HeurChain + date writing 72.5%, Mem0 71.3%. DEV only; gaps inside run noise.

benchmarkinglocomomemory-systems

Why temporal answers fail on evidence already retrieved

· HeurChain team

Temporal misses are answer-side (dates unresolved); multi-hop misses are retrieval-side (evidence missing). DEV-only internal analysis.

benchmarkinglocomotemporal-reasoningmulti-hop-reasoningretrieval