adjacent-atlas

Field notes — limitations of the scoring

The adjacency score is a useful lens and a blunt instrument. Its limitations are not footnotes; they are the honest boundary of what the tool can claim.

It measures attention, not merit

Every input is a proxy for how much public activity surrounds a technique. A foundational, settled method can look quiet; a fashionable dead end can look busy. The score conflates these on purpose, because the question is “what is moving”, but that means a high score is a prompt, never a verdict.

The signals are coarse proxies

GitHub repository counts and OpenAlex keyword-search counts are noisy. A search for a technique will sweep in loosely related work and miss work described in other words. These are not curated bibliometrics and must not be read as citation counts or as any quantitative claim about a specific paper.

The weights are a point of view

Momentum at 30%, recency and connectivity at 20%, novelty and feasibility at 15% — these are defensible, not objective. Change them and the ranking changes. The method page shows the weights precisely so a reader can disagree with them.

The graph is small and authored

Edges and their weights are hand-set. With a small graph, connectivity is sensitive to a handful of decisions. This is fine for a readable map of one field; it is not a derived network.

No causal or predictive claim

Nothing here forecasts. A high opportunity score does not mean a technique will succeed; it means it is young, moving, and buildable by the proxies above. Treat it as a reading list, not a bet.