finweb

The FinField explorer — a framework-free static site over 20,000+ listed companies and crypto assets, live at finfield.github.io/web, plus the deterministic builder that compiles its data.

• web/ — the site: plain HTML/CSS/JS, no framework, no build step. Instant search over the whole universe; a company page that renders every curated annual fact and derived metric with its provenance: source kind, SEC accession (linked to the filing), fetch date, CID, and derived_from links you can click to walk a ratio back to the audited filing.
• src/finweb/build.py — the shard compiler: universe + cached SEC facts → web/data/ (per-company JSON shards, search index, meta). A pure function of its inputs: same universe + same cache → byte-identical output tree. Values stay exact scaled integers end to end; the browser renders them from the integers, no floats anywhere.

Build the shards

pip install -e ".[dev]"                       # finfacts comes from ../facts or git

python -m finweb.build --out web/data                              # identity-only universe
python -m finweb.build --out web/data --facts "AAPL US" --fetch    # + SEC facts (needs finscrapers)
python -m finweb.build --out web/data --facts-file web/facts_tickers.txt --cache .cache/sec
python -m finweb.build --verify web/data                           # recompute every CID offline

--verify re-hashes every fact in the tree and checks each one carries a complete source block — anyone can audit a deployed site without trusting the builder. A size guard fails the build before it can outgrow static hosting.

Serve locally

python -m http.server -d web    # then open http://localhost:8000/

All asset and data URLs are relative, so the site works from any mount path (GitHub Pages serves it under /web/). The built web/data/ tree is not in this branch — it is deployed to the gh-pages branch.

Tests

python -m pytest -q     # offline: synthetic universe + canned facts, no network

Part of the FinField field: facts · scrapers · knit · agents · signals · crypto