009-Link relationship semantics
Type: adr · Status: accepted · Tags: links
Status: accepted Date: 2026-02-21
Context
Untyped links ("see also", "related") tell a reader that two notes are connected but not how. An agent navigating under context pressure needs to decide whether following a link will advance its task — and "related" gives no signal for that decision. Wiki-style hyperlinks are cheap to create but expensive to follow blindly.
Ars Contexta proposes propositional link semantics drawn from concept-mapping research: causes, enables, contradicts, extends, specifies, supports. The key distinction is between mind mapping ("these relate somehow") and concept mapping ("this extends that because..."). We needed a vocabulary small enough to remember and use consistently, rich enough to support agent navigation decisions.
Decision
Every link in the KB must articulate the relationship using one of these types:
- extends — builds on, adds a dimension to, refines
- grounds (also: foundation) — provides the theoretical or evidential base
- contradicts — conflicts with, challenges, creates tension
- enables — makes possible, is a prerequisite for
- exemplifies (also: example) — is a concrete instance of
The relationship appears in the prose surrounding the link. In body text: "since title" or "because title". In Relevant Notes footers: "- title — extends: ..." with an explicit relationship word and a context phrase.
"Related" is not a relationship. If you cannot name the relationship, the link may not be worth making.
Consequences
Easier
- Agent navigation — an agent can prioritize links by type: follow "grounds" when verifying a claim, follow "contradicts" when looking for tensions, skip "exemplifies" when time is tight.
- Graph maintenance — typed links are testable. A "grounds" link to a note that doesn't provide evidence is a detectable error. Untyped links are unfalsifiable.
- Traversal as reasoning — when titles are claims and links carry relationship types, traversing the graph reads as an argument chain, not a random walk.
Harder
- Authoring cost — every link requires a relationship judgment. This is intentional friction — it prevents decorative linking — but it slows writing.
- Vocabulary drift — the vocabulary must stay small and stable. Adding types (e.g. "supersedes", "specializes") requires explicit decision, not gradual accumulation.
- Coverage — some genuine relationships don't fit cleanly (temporal succession, mutual dependency, "same phenomenon from a different angle"). The vocabulary is deliberately coarse; edge cases use the closest fit with a clarifying phrase.
Relevant Notes:
- Ars Contexta — source: propositional link semantics vocabulary we borrowed and adapted
- title-as-claim-enables-traversal-as-reasoning — enables: claim titles plus typed links make traversal read as reasoning
- link-strength-is-encoded-in-position-and-prose — extends: relationship type is one axis; position and prose encode a second axis (commitment strength)
- agents-navigate-by-deciding-what-to-read-next — grounds: the navigation-decision model that typed links serve