Link strength is encoded in position and prose
Type: note · Status: seedling
The link contracts framework defines what links should contain (relationship type, context phrase, click-decision support). But it treats all links as equal edges. In practice, links carry different commitment levels — an inline "since [X]" that uses a note as a premise is a stronger connection than a footer "related" entry.
This matters because it creates a weighted graph. The weight affects how agents should traverse, how notes should be scored, and how graph health should be measured.
Strength signals
Position in the document is the strongest signal. A link woven into the argument is load-bearing — the prose breaks if you remove it. A link in the footer "Relevant Notes" section is catalogued but not depended on.
| Position | Strength | Why |
|---|---|---|
| Inline, in a premise or argument | Strongest | The current note's reasoning depends on this link |
| Inline, in supporting context | Strong | Referenced as evidence or example |
| Inline, parenthetical mention | Medium | Acknowledged but not load-bearing |
| Footer with context phrase | Weak | Related, catalogued, but not part of the argument |
| Footer bare link | Weakest | Filed without justification |
Relationship words in prose refine the signal. The word that introduces a link tells you what role the linked note plays:
| Prose pattern | Role | Strength |
|---|---|---|
| "since [X]", "because [Y]" | Premise — the linked note is a reason | Strongest |
| "extends", "builds on" | Structural — the linked note is a foundation | Strong |
| "contradicts", "but see" | Tension — the linked note challenges this | Strong (different direction) |
| "see also", "related to" | Association — topical overlap | Weak |
| "cf.", bare link | Catalogued — no articulated reason | Weakest |
The load-bearing test. The simplest way to assess link strength: would the paragraph still make sense if you removed the link? If yes, the link is decorative. If no, it's structural. Structural links are strong; decorative links are weak.
What link strength affects
Traversal priority. An agent deciding what to read next should follow strong links before weak ones. A "since [X]" link is almost certainly worth following — the current note depends on it. A footer "related" link is a maybe. This is the practical answer to the navigation decision problem: link strength is a traversal heuristic.
Note scoring. A note's quality score should weight inbound links by strength. Three inline premise links from well-regarded notes say more about a note's value than twenty footer "related" links. This is PageRank with link-weight — and it's the mechanism that prevents the credibility erosion problem (noisy weak links don't inflate scores the way unweighted link counts would).
Quality signals. The ratio of strong to weak links is a graph health signal. A note with mostly strong inbound links is well-integrated into the KB's reasoning. A note with only weak footer links is catalogued but not used. The quality signals note should track this ratio.
/connect guidance. When /connect adds links, it should prefer creating strong connections (inline, with relationship articulation) over weak ones (footer "related"). A strong connection is worth more than three weak ones. This is already implicit in the skill's articulation requirement, but making it explicit could improve connection quality as the KB scales.
Should strength be explicit metadata?
Two options:
Inferred from position and prose. A script parses each markdown file, classifies links by position (inline vs footer) and surrounding words. No new metadata to maintain. But parsing prose for relationship words is fuzzy — "since" might be temporal, not causal.
Explicit in link syntax. Something like [note](./note.md "premise") or a structured footer format. Clean and queryable, but adds ceremony to every link and changes the writing convention.
The inferred approach is probably right for now — position (inline vs footer) is easy to detect, and that alone captures most of the signal. Prose analysis can be added later if the coarser signal isn't enough.
Open questions
- How much of the strength signal is recoverable from position alone (inline vs footer), without prose analysis?
- Should /connect explicitly choose between inline and footer placement based on relationship strength, or leave that to the author?
- Does link strength decay over time? A premise link from a note that's now
outdatedis weaker than the same link from acurrentnote. Should link strength be static (property of the link) or dynamic (property of the link × source note status)? - The arscontexta "specificity test" ("genuine elaboration is specific enough to be wrong") is a link quality gate. Is there a link strength gate — a minimum strength below which a link isn't worth creating?
Relevant Notes:
- link contracts framework — defines link semantics (relationship types, context phrases); this note adds the strength dimension that link contracts don't currently address
- notes need quality scores to scale curation — note scores should weight inbound links by strength; strong inbound links count more
- agents navigate by deciding what to read next — link strength is a traversal heuristic: strong links are worth following, weak links are maybes
- quality signals for KB evaluation — strong-to-weak link ratio is a graph health signal
- title as claim enables traversal as reasoning — claim titles make link strength legible: "since [claim-title]" is visibly a premise link
- Ars Contexta — the credibility erosion problem: unweighted link counts let noise drown signal; link-weighted scoring prevents this
Topics: