Temporal
Time-Aware Ranking
Aletheia ranks by relevance and freshness so outdated context does not dominate.
Temporal scoring
After retrieval and optional reranking, Aletheia applies temporal policy. Expired memories are filtered; surviving memories can be decayed based on age and kind.
This reduces stale recall while preserving long-lived facts and preferences.
TTL and decay
- TTL is a hard cutoff.
- Decay is a soft demotion.
- Policy is kind-specific, not global.
final_score = relevance_score * freshness_weight(kind, age)
if age > ttl(kind): dropPractical tuning
Tune using replay datasets with known truth changes. If historical facts still outrank new facts, shorten half-life for conversational memory or increase supersession penalty.