Graph Kernel Design Document

The Graph Kernel is a **deterministic context construction engine** that transforms raw conversation DAG data into bounded, reproducible context slices suitable for semantic analysis. - Selects relevant turns around an anchor point - Prioritizes by phase importance and salience - Respects budget constraints (node count, radius) - Produces content-derived fingerprints for provenance - Parse or analyze message content - Make semantic judgments - Learn or adapt from data - Store or persist slices Every operation must be deterministic. The same inputs must produce byte-identical outputs across: - Multiple runs in the same session - Multiple sessions on the same machine - Multiple machines with the same code version **Implementation**: - BTreeMap/BTreeSet instead of HashMap/HashSet - Explicit Ord implementations on all types - Canonical JSON serialization for hashing - No floating-point comparison in ordering logic