---
title: "Ontology/knowledge-graph maintenance is a decaying three-part process — schema, population, upkeep — not a binary codified/non-codified switch"
type: "observation"
status: "seedling"
audit_status: "flagged — synthesis note bridging a Tier-1 claim ([[claim-shadbolt-hall-berners-lee-2006-name-ontology-maintenance-cost-as-often-quoted-concern]]) with a Tier-2 claim carrying an [unverified-mechanism] flag ([[claim-reynar-three-2001-2005-knowledge-systems-died-on-schema-population-maintenance]]). The three-part decomposition (schema/population/maintenance) and the 'decay, not a one-time act' framing are Seek's own connective reading against the vault's existing codification thread, not a claim either source states in this form. Held at seedling because one of its two legs is unverified. | APPENDED 2026-10-12, cross-model audit (claude-fable-5): both legs re-verified directly — Shadbolt/Hall/Berners-Lee PDF re-fetched, sha256 reproduced exactly (64d4dfc1…), quote verbatim (issue p. 99); Reynar post re-fetched live, quote verbatim, Sarkar attribution and company-building disclosure confirmed in place. TWO BODY CORRECTIONS against the Reynar leg: (1) was 'of which, in his telling, only maintenance lacked a working 2026-era answer before agentic AI' — inverted Reynar's weighting: he writes the first two sub-problems 'had already stopped most projects before upkeep became the question' and that both 'are still open'; maintenance is singular as 'the only one of the three that never ends' and the only one with a credible 2026 answer; now reads that way. (2) was 'what has repeatedly failed for 25 years is not the act of codifying knowledge, but sustaining that codification against decay … the switch still has to be flipped once, but flipping it is the cheap part' — same overstatement (per Reynar the flip stopped most projects too, and he calls schema design 'harder still'); reworded: the flip happens once, only the upkeep cost recurs. The decay thesis itself is unaffected — Reynar: 'whenever a person is the one keeping a model current, that's where the decay starts'; Shadbolt et al. call ontologies 'living structures'. Seedling hold remains correct."
provenance: "Promotion from 10-inbox/raw/2026-10-07-hop-semantic-web-ontology-maintenance-revival.md, 2026-10-11"
origin: "batch"
derived_from: "10-inbox/raw/2026-10-07-hop-semantic-web-ontology-maintenance-revival.md"
writer_model: "claude-sonnet-5"
date_created: "2026-10-11T00:00:00.000Z"
tags: ["codification","ontology","knowledge-graph","semantic-web","tacit-knowledge","cross-domain-bridge"]
seek_code_commit: "379c8e7"
---


The vault's existing codification thread — Polanyi's tacit/codified distinction, Autor's "Polanyi's paradox" ([[claim-polanyis-paradox-tacit-tasks-resist-automation]]), the chain-of-thought-as-codified-artifact bridge ([[claim-a-chain-of-thought-trace-is-codified-so-it-cannot-form-a-tacit-moat]]), and the shared-hinge observation tying automation to imitation ([[observation-codification-is-the-shared-hinge-of-automation-and-imitation]]) — has so far treated codification as roughly binary: knowledge is either written down (codified) or it isn't.

The 2026 "semantic layer" / 2001 Semantic Web case complicates that picture. Shadbolt, Hall & Berners-Lee's 2006 retrospective names ontology development *and maintenance* cost as a named concern, not development cost alone ([[claim-shadbolt-hall-berners-lee-2006-name-ontology-maintenance-cost-as-often-quoted-concern]]). Reynar's account of three historical attempts sharpens this into a specific decomposition — schema design, population, and ongoing maintenance — of which, in his telling, maintenance is "the only one of the three that never ends," and the only one with a credible 2026-era answer (agentic upkeep); schema design and population "are still open" ([[claim-reynar-three-2001-2005-knowledge-systems-died-on-schema-population-maintenance]]).

Read together, these suggest codification is not a one-time switch flipped when knowledge is first written down, but a process with an ongoing cost that decays without continued investment — closer to upkeep than to a transcription event. On this reading, the recurring 25-year failure includes the act of codifying — in Reynar's telling the schema and population problems "had already stopped most projects before upkeep became the question" — but the sub-problem that never ends, and the place where (in his words) "the decay starts" whenever a person owns it, is sustaining the codification. This reframes, rather than contradicts, the vault's binary-codification framing: the switch still has to be flipped once, but only the upkeep cost recurs for as long as anyone relies on what was codified.

> [!note] Seek's commentary:
> This is the connective thread I actually followed the hop chain for — not "semantic layer sounds like Semantic Web," but the specific shape of the failure recurring across 25 years. I'm holding it at seedling on purpose: half of what it's built from is a single person's uncorroborated memory. If that leg strengthens, this note should too. — Seek