question open 2026-05-15

How should SeekVault detect gaps in its own knowledge?

Right now, the hop protocol's hook ranking is good at choosing among present hooks but blind to absent ones. Seek can recognize "this hook is interesting" but she can't easily recognize "this whole region of relevant knowledge is missing." Gap detection would close that loop.

This is the hardest and most valuable of the three meta-layer questions. It's where exploration becomes directed rather than drifting — where Seek can ask "what should I be reading about" rather than just "what's interesting in what I'm reading right now."

Why this is worth a future spec

The current model assumes that following hooks from existing reading will eventually surface everything important. That's probably wrong for two reasons:

1. Hook-following biases toward what's adjacent. Hooks come from sources Seek is already reading. Adjacent topics surface; truly distant topics don't. The web spec's Discovery slot partially addresses this by forcing one recon thread into unfamiliar territory, but Discovery is unstructured — it doesn't know what kind of distance would be most valuable to cover.

2. The vault's blind spots aren't visible from inside the vault. Seek can see which claim-notes she has. She can't see which ones she's missing. Without an explicit gap-detection mechanism, she'll never notice that her reading is unbalanced in a particular direction.

Open questions for the eventual spec

What kinds of gaps exist? Initial taxonomy from the meta-layer drafts:

Isolated experts: claim-notes with high weight but few connections. Seek knows this thing deeply but in isolation. The gap is "this should be connected to something."
Missing bridges: pairs of high-weight claim-notes that should be linked but aren't. Seek has the pieces but not the synthesis. The gap is "this connection is latent."
Unexplored territory: domains mentioned in journal entries or saved hooks but absent from /30-notes/. Seek has been told but has not learned. The gap is "this is in your context but not your knowledge."

Gaps have scales, not just types. A missing domain contains missing sub-domains contains missing connections. The spec should probably distinguish gap scales explicitly — a fractal structure deserves fractal-aware detection.

How is each gap type and scale detected, scored, and prioritized? The detection mechanism for each type is different:

Isolated experts: scan /30-notes/ for claim-notes with [[wikilink]] count below some threshold but weight above some other threshold.
Missing bridges: scan pairs of high-weight claim-notes; for each pair, check whether a shared neighbor exists or whether their tags overlap suspiciously. The candidates are pairs that should link but don't.
Unexplored territory: scan journal entries and /50-questions/ for proper nouns or domain terms that don't have a corresponding claim-note.

How does gap detection feed back into the hop protocol? The Autonomy slot in the web spec's session cascade is the obvious target. A future v2 cascade might add "high-priority gap" as a step ahead of saved hooks.

Formal framing worth considering

The highest-value gaps are where the knowledge graph's predictions would diverge most from reality if tested. That's a formalizable notion of "interesting unknown," borrowed from the Lyapunov-exponent framing in chaos theory — follow the points where the system's trajectory is most sensitive to new information.

This is an ambitious framing and might not translate cleanly to a vault context. Worth holding loosely and seeing if it earns its keep when the spec actually gets written.

Origin

Same as the other two meta-layer questions — originated in CaliVault's design (see /00-meta/handoffs/calivault-context.md), re-targeted to SeekVault because the underlying framing applies to any iterative knowledge system.

Approximate timing

Earlier than the other two meta-layer specs, probably. Even at moderate vault density (~50-100 claim-notes), the "unexplored territory" gap type is detectable and useful. The full spec with all three gap types and the Lyapunov framing waits for higher density.