This method note describes a five-layer classification architecture built for a personal research archive of ~180 pieces spanning blog essays, creative works, translations, and forensic tools. The system was designed against two failures: static taxonomies that flatten cross-cutting work into single categories, and algorithmic recommendation systems that dissolve authorial intent into engagement metrics. The architecture treats classification as living induction rather than static labeling. Its five layers — tier gating, multi-cluster assignment, cluster-specific relevance scoring, intellectual-operation typology, and a recursive corpus induction loop — were developed through multi-round collaborative design between the author and AI collaborators, with each layer's categories discovered from cleavages already present in the corpus rather than imposed from external taxonomy. The note's primary contribution is the intellectual-operation typology: nine inductively identified modes of intellectual work (autotheoretical naming, narrative-form critique, formal scholarly paper, critical close-reading, structural diagnostic, incident-unfolding cultural critique, method/tool-building, declarative ontological reclamation, manifesto-testimony) detected via signal fingerprints and calibrated across four iterative phases. A secondary contribution is the dual-signal substance scoring formula that combines regex-detected structural rigour with LLM-evaluated literary value, solving the problem that academic formatting does not equal academic substance. The system is offered not as a universal standard but as a transferable methodology for independent researchers building self-governed archival infrastructure outside institutional gatekeeping.
Julia Chika Kham (Tue,) studied this question.