Nonaka & Takeuchi's SECI model (1995) maps tacit-to-explicit knowledge through four stages: Socialization → Externalization → Combination → Internalization. Most AI memory systems skip stage 1.
The externalization trap: writing structured docs on first signal creates expensive artifacts that decay within weeks as the situation evolves.
Luhmann's Zettelkasten used a "promotion gate" — fleeting notes are cheap; permanent notes earn their place by surviving repeated use, not upfront intent.
The 3+ corroborating observation rule: a concept only graduates to your Knowledge Library after at least three independent instances confirm the pattern matters.
Wrong + cheap = acceptable. Wrong + expensive = system decay. The architecture inverts risk by making early-stage artifacts disposable by design.
An untrustworthy knowledge base is worse than none — it creates invisible debt as people spend time consulting docs they've silently stopped believing.
"Structure without evidence is just prediction. And in a fast-moving domain, prediction decays fast. Keep knowledge fluid until patterns emerge — then, and only then, externalize into something you're willing to maintain."
Originally written about Toyota. Maps perfectly onto AI agent memory architecture.