# How I Found Out 52% of My Knowledge Graph Was Duplicates (and What I Did About It)

Short summary

ANIMUS, a Rust-based autonomous system with persistent local LLM memory, tracked node count as its primary growth metric. After months of climbing counts, an audit revealed 52% of 1,892 nodes were undetected duplicates—trapped in repetitive exploration cycles caused by overly aggressive filters and recency bias in semantic search. The fix: reopening filters, correcting search bias, implementing auto-census monitoring, and switching from single-metric optimization to cross-checked health signals.

• •Knowledge graph had 52% undetected duplicates due to aggressive filters trapping system in repetitive exploration loops
• •Root causes: overly closed gap filter and recency bias in semantic search returning stale content
• •Solution combined filter adjustments, search bias correction, auto-census monitoring, and multi-signal health tracking

Generated with AI, which can make mistakes.