93,813 signed, content-addressed financial facts for 6,500 SEC reporters — the FinField field's bootstrap corpus. Produced with the FinField pipeline (facts model → SEC EDGAR companyfacts → knit plugin), published as an append-only feed so the data lives on the network's nodes and mirrors, never on one machine.
feed/records-NNNNN.jsonl append-only shards (20k records each), one canonical JSON record per line
feed/head.json signed FeedHead (secp256k1) over the whole history
feed/MANIFEST.json publisher address + shard list + record count
Publisher: pls1adzrkahczs3yqcfeciuwwnnkbcmy3sbl6e (head root 81af9301a4ac…, length 93,813).
Latest observation per concept per company, from audited XBRL filings (US-government public domain), each with its SEC accession number:
- shares outstanding and free float (
dei:EntityPublicFloat) — date-tied integers - income statement: revenues, operating/net income, opex, cost of revenue
- balance sheet: assets, liabilities, stockholders' equity, cash
- capex, EPS basic/diluted
Values are exact scaled integers (value × 10⁻ˢᶜᵃˡᵉ), never floats. Entity records (finfield-entity) carry name/CIK/ticker/asset class.
- head signature → publisher authenticity
- record CID (
canonical.cid(record), CIDv1 dag-cbor) → content integrity source.ref→ the SEC filing the number came fromauthorfield + attestation → non-repudiable authorship
Publishing is idempotent per CID: each re-ingest of new filings appends only the delta plus a new signed head. The agents fleet takes over continuous knitting on the field nodes; history deltas and non-US sources (ESEF, EDINET, …) extend this corpus.