adjacent-atlas

Adjacent Atlas

An observatory for the adjacent possible. Adjacent Atlas reads public signals — code activity and literature — and ranks the methods, instruments, and concepts close enough to a field’s frontier to be worked on next. It renders that ranking as an interactive 3D atlas and a set of sourced briefs.

The reference map shipped with the repository is extreme-precision radial velocity (EPRV) instrumentation.

Status: 0.1.0, early development. The engine and data schemas are stable enough to build on; the live fetch pipeline is the active edge.

Screenshots

Screenshots are generated from the local app rather than committed as stock images. To produce them, run the dev server (below), open / and /atlas, and capture at 1440×900. A generated social card is available via pnpm tsx scripts/export-social-card.ts → apps/web/public/social/card.svg.

What it does

fetch (code + papers) → normalise → score → snapshot → render

Each node gets an adjacency score (0–100): a weighted mean of five normalised components — momentum, recency, connectivity, novelty, feasibility. The model is a transparent heuristic, not a trained predictor; weights are visible and tunable. See docs/scoring-model.md.

Project story

Adjacent Atlas began as a way to make “what should I look at next?” answerable from public signal rather than intuition alone — in a field (EPRV instrumentation) where the gap between interesting and reachable is set by instrument systematics. The result is closer to an instrument readout than a dashboard: dark, quiet, and meant to be read closely.

Repository layout

Path Purpose
apps/web Next.js 14 site, 3D atlas, and JSON API
packages/engine Scoring, graph, layout, synthesis, export
packages/ui React primitives and design tokens
packages/embed Dependency-light embeddable view
data/schemas JSON Schemas for the data contracts
data/seeds EPRV seed graph
scripts Fetch, build, validate, and social-card tools
docs Architecture, methodology, scoring, visual language
design Token specs and image-generation prompts

Local setup

Requires Node ≥ 20 and pnpm ≥ 9 (corepack enable).

pnpm install
pnpm build        # builds engine + ui before the web app (turbo handles order)
pnpm typecheck
pnpm test

Run the web app (build the libraries first, or run their watch tasks):

pnpm --filter @adjacent-atlas/web dev

Generate and validate a snapshot:

pnpm snapshot
pnpm snapshot:validate

Scoring at a glance

adjacency = 100 × ( 0.30·momentum + 0.20·recency + 0.20·connectivity
                    + 0.15·novelty + 0.15·feasibility )

Every component is shown on the atlas and in each brief, so a ranking can be read, not just trusted.

Roadmap

Real fetch implementations, snapshot history and drift, a Barnes–Hut layout for larger graphs, and additional reference domains. Full list in docs/roadmap.md.

Deploying

Adjacent Atlas runs as a Node server (the API routes are dynamic), so it is not a static export. Build with pnpm build and deploy to Vercel or any Node host; see docs/deployment.md. Integration into another site (link, embed, or iframe) is covered in docs/website-integration.md. Launch materials — repo description, topics, a LinkedIn post, and image briefs — are collected in docs/launch-pack.md.

Ownership and license

Built and maintained by Biswajit Jana (@Biswajit1999). Released under the MIT License — see LICENSE. The seed signals values are illustrative scaffolding, not measured statistics; see docs/methodology.md.

Research Quality Upgrade

See RESEARCH_QUALITY.md for the validation layer, reference anchors, equations and research boundaries added to this repository.