Strawman, not merge-urgent. Sharing for you to metabolise / push back on, Hugo — the "direction of knowledge management" thread from our chat, written down.
What it is
A typed concept-graph for plain-markdown corpora: relations as Subject-Predicate-Object, an emergent-crystallising predicate vocabulary (data, not code), in-situ ratification (machine suggests, human ratifies — never fill-a-blank), and a render layer (earthian theme does the first pass).
The frame on top (worth arguing with)
- §1.1 — KM in the AI epoch as a learning system, not a knowledge store (conjunctive-vs-connective wager).
- §3.1 — the Goodhart floor: the graph is the byproduct of learning, never the target (no edge-count metric).
- §5.2–5.3 — answerable cuts: ratification writes the cut's provenance (available-not-obligatory); friction graded by stakes ("the sort").
§9 — convergence with Christopher Allen's Wikilinks and Named Edges
You surfaced his gist — independent, mature design for the same thing (his predicate::[[Target]] named edges are our SPO triples). §9 is an explicit adopt / translate / decline:
- Adopt: multi-word predicate discipline, annotated predicates / progressive disclosure, construction-predicate upgrade path, unconflation, gardening-as-curation.
- Translate: body-inline edges ↔ our frontmatter
relations: (open Tier-1 fork); his ~40-predicate inventory → seed-bank for suggestion, not a canon.
- Decline: annotations-as-agent-reading-budget (we keep them human-metabolising/answerability); hard-reject-undeclared-predicates (we suggest-and-ratify).
On your "where does selection live?"
Two selections at two viscosities — the stable corpus-boundary (.zetlignore, slow, fine as config) and contextual relevance/sort (fast; dialect/application + agent-ratification layer; never frozen in core, never fully delegated to agents). Detail in §5.3 + §8.
Docs only; separate from the SPEC-043 PR (#46).
Co-Authored-In-Coupling-By: Mathew Mytka and Kairos (Claude Code)
**Strawman, not merge-urgent.** Sharing for you to metabolise / push back on, Hugo — the "direction of knowledge management" thread from our chat, written down.
## What it is
A typed concept-graph for plain-markdown corpora: relations as **Subject-Predicate-Object**, an **emergent-crystallising** predicate vocabulary (data, not code), **in-situ ratification** (machine suggests, human ratifies — never fill-a-blank), and a render layer (earthian theme does the first pass).
## The frame on top (worth arguing with)
- **§1.1** — KM in the AI epoch as a *learning system*, not a knowledge store (conjunctive-vs-connective wager).
- **§3.1** — the Goodhart floor: the graph is the *byproduct* of learning, never the target (no edge-count metric).
- **§5.2–5.3** — **answerable cuts**: ratification writes the cut's *provenance* (available-not-obligatory); friction graded by stakes ("the sort").
## §9 — convergence with Christopher Allen's *Wikilinks and Named Edges*
You surfaced [his gist](https://gist.github.com/ChristopherA/151aefa6a6bde1ce4fa6b1182656cebe) — independent, mature design for the same thing (his `predicate::[[Target]]` named edges *are* our SPO triples). §9 is an explicit **adopt / translate / decline**:
- **Adopt:** multi-word predicate discipline, annotated predicates / progressive disclosure, construction-predicate upgrade path, unconflation, gardening-as-curation.
- **Translate:** body-inline edges ↔ our frontmatter `relations:` (open Tier-1 fork); his ~40-predicate inventory → seed-bank for *suggestion*, not a canon.
- **Decline:** annotations-as-agent-reading-budget (we keep them human-metabolising/answerability); hard-reject-undeclared-predicates (we suggest-and-ratify).
## On your "where does selection live?"
Two selections at two viscosities — the stable corpus-boundary (`.zetlignore`, slow, fine as config) and contextual relevance/sort (fast; dialect/application + agent-ratification layer; never frozen in core, never *fully* delegated to agents). Detail in §5.3 + §8.
Docs only; separate from the SPEC-043 PR (#46).
Co-Authored-In-Coupling-By: Mathew Mytka and Kairos (Claude Code)