Adjacent Atlas answers one question: within a field, which methods, instruments, and concepts are adjacent to the current frontier — close enough to be the next thing worked on — and which of those already have momentum?
Two public signal sources:
These are normalised into a graph of nodes (concepts, methods, instruments, datasets, implementations, results) connected by weighted edges (relates, depends, cites, implements, supersedes).
Adjacency is not novelty alone, nor popularity alone. It is a deliberate blend:
something is adjacent when it is moving (momentum, recency), connected to the
rest of the field (connectivity), not yet saturated (novelty), and buildable
(feasibility). The scoring model in docs/scoring-model.md makes this concrete.
The shipped graph is extreme-precision radial velocity (EPRV) instrumentation. It was chosen because progress there is gated by instrument systematics as much as by ideas, so the difference between “interesting” and “reachable” is sharp.
The bundled seed’s signals values (activity series, implementation counts,
maturity) are illustrative scaffolding that exercise the model across a realistic
range. They are not measured statistics. Live snapshots replace them with values
from the fetch scripts.
The scoring is a transparent heuristic, and its boundary matters as much as its
output. The honest limitations — attention versus merit, proxy noise, the
subjectivity of the weights, and the absence of any predictive claim — are set
out in field-notes/scoring-limitations.md.
Authored, researcher-voice context lives alongside this document:
These are commentary, not data. The seed’s activity numbers are illustrative;
real snapshots come from the ingestion pipeline (docs/data-sources.md).