concept-synthesis

Deduplicate concept stubs using Jaccard and substring matching with semantic duplicate detection to merge pages, preserving aliases and timelines. This synthesizes repeated mentions into a persistent knowledge graph tracing idea evolution.

Concept tiering scores frequency, timespan, breadth, and engagement to assign T1–T4 levels, categorizing concepts as Canon, Developing, Speculative, or Riff to structure an intellectual map and filter high-signal ideas for LLM synthesis.

Generate a clustered intellectual map by running LLM synthesis on T1/T2 concepts, then clustering tiered concepts into named domains with summaries and a master concepts/README.md written under the concepts/ directory.

Yes, concept synthesis applies deterministic dedup and merging to large-scale note corpora where repeated or conflicting concept mentions have been ingested as individual stub pages, eliminating shallow duplicates.

LLM synthesis is restricted to T1/T2 high-signal tiers only. T3 and T4 concepts bypass LLM synthesis, relying solely on deterministic dedup, scoring, and tier assignment to preserve processing efficiency.