Share financial data peer-to-peer and feed prediction tournaments — no central data broker.
FinField is an open field for provenance-first financial facts: fundamentals, prices, signals, targets and eras for 20,000+ listed companies and crypto assets. Publish once, signed and content-addressed; peers replicate it over the knitweb network and export it submit-ready for competitions like Numerai Signals and Numerai Crypto.
How it works
Publish a dataset
Package your data — fundamentals, price history, alt-data signals, or engineered features —
as a signed, content-addressed record with a declared license and provenance chain.
Same record, same CID, everywhere.
Peers replicate it
Other participants subscribe to the FinField knitweb and pull the records they want, directly from peers. Signatures and CIDs make tampering detectable; nobody sits in the middle.
Export & compete
Turn any combination of records into a submit-ready frame for your tournament of choice — ticker-mapped, era-aligned, point-in-time clean — and keep the full provenance trail of what went into your model.
Competitions
What lives in the field
Fundamentals & reference
Balance-sheet and income facts, corporate actions, identifiers and ticker maps for 20,000+ listed companies — provenance-first, every fact traceable to its source.
Prices & market data
End-of-day and periodic price series for equities and crypto, point-in-time stamped so backtests can't peek into the future.
Features & signals
Engineered features and alternative-data signals contributed by participants — shareable, licensable, and attributable to their author by signature.
Targets & eras
Tournament-aligned target definitions and era calendars, so a dataset pulled from the field lines up with the competition you're entering.
Modules
facts
The data model: scaled-integer FinFacts (never floats), deterministic CIDs, the 20,000+ company universe, and exact derived metrics with provenance chains.
scrapers
Ready-to-run, pure-Python scrapers: SEC EDGAR fundamentals, Stooq equity prices, CoinGecko crypto — each normalizing into signed-ready FinFacts.
knit
The FinField knitweb plugin: knit your own data in with your own key, and vote on what is factually correct — pulse polls for exact facts, vank consensus for float-natured values.
agents
The 5mart.ml/finfield deployment: every scraper runs as an agent with its own key, knitting fresh facts into the field on a schedule.
signals
Numerai Signals features from field data — exact-math momentum, volatility and fundamentals, exported as a submit-ready prediction file.
crypto
Numerai Crypto features from vank-consensus crypto prices, exported in the tournament's submission format.
lens
Analyst lenses for LLMs: sector & industry, country and macro-economy digests — read-only, deterministic, every answer citing fact CIDs and the snapshot state root.
web
The live explorer: provenance-first company pages for the whole universe — per fact its source, CID and derivation chain.
feed
Signed bootstrap corpus — 93,813 content-addressed facts from 6,500 SEC reporters, ready to verify and replicate into your own node.
Why p2p
No gatekeeper
Data brokers charge rent and decide who gets access. In FinField, publishers set the terms and peers exchange directly.
Provenance beats trust
Every record is signed and content-addressed. You don't have to trust a vendor's pipeline — you can verify the chain from source to submission.
Contributors get credit
Signals carry their author's signature. Useful data builds a verifiable track record, the same way good predictions do.