Expand description
Archetype detection (Ontology phase — the COMPILER-gap attack).
Deterministic evidence scoring over the reality graph + raw import
table classifies a repository into one of eight archetypes (plus the
honest Unknown fallback). Every signal contributes +1..+3; the
highest total wins; score ties are broken by the fixed precedence
order in scc_core::Archetype::PRECEDENCE.
Provenance discipline: all signals are counts/shape facts over EXTRACTED entities and relationships — no heuristics are promoted to facts, this function only classifies.
Enums§
- Cluster
Prior - Archetype emphasis (semantic clustering): for each archetype the clustering priors change — one region trait coheres (+PRIOR_WEIGHT) when the repository matches that archetype.
Constants§
- PRIOR_
WEIGHT - Extra weight applied between two regions sharing the prior trait.
Functions§
- cluster_
prior - The clustering prior for an archetype, or
Nonewhen the archetype adds no region-trait prior (LibrarySdk’s emphasis is the doubled public-surface cohesion weight in the clustering signal itself; Unknown/Monorepo/Infra/ Plugin rely on the base signal set). - detect_
archetype - Deterministic evidence scores per archetype. Pure function of graph + import-table shape; every signal is an integer count comparison.
- is_
phase_ symbol - True when a symbol name matches the compiler/language-tool phase
vocabulary (
parse/analyze/transform/generate/…). Shared with archetype detection so the clustering prior uses the same vocabulary.