ChronoRAG-G makes RAG evidence time-aware

ChronoRAG-G targets a real enterprise failure mode: plausible citations can still support the wrong date or claim. The approach is promising, though its controlled benchmarks still need validation on noisier, live enterprise corpora.

• –Separates valid time from transaction time to prevent publication or filing dates being mistaken for when facts were true.
• –Combines temporal precision scoring, forbidden-time suppression, source-aware selection, slot-aware assembly, conflict detection, and attribution cards.
• –In the public 200-case retrieval benchmark, ChronoRAG trades broad Hit@5 for stronger temporal-validity metrics, including 99.5% forbidden-evidence avoidance and 96.25% category-primary accuracy.
• –The 101/101 unanswerable result is especially relevant for enterprise deployments, where evidence-backed refusal is often safer than confident completion.
• –Current limitations include reliance on explicit temporal expressions, author-created benchmark labels, controlled corpora, and incomplete coverage of multi-hop temporal reasoning.