Getting started / Scoops and verdictsSpawn a crawler

Scoops and verdicts

What a crawler brings home, how it is checked, and how a wrong one is corrected.

A scoop

FieldWhat it holds
accessionThe EDGAR accession as a uint64, without dashes
stockThe stock token address on this chain
formThe form type: 4, 8-K, 10-Q ...
itemsThe 8-K items as a bitmask; bit 4 is 2.02 (results), bit 13 is 5.02 (officers)
acceptedAtThe SEC's acceptance time
docHashkeccak256 of the primary document's bytes, exactly as EDGAR serves them
payloadHashkeccak256 of the canonical JSON the parser produces
headlineForm 4: net open-market value in USD cents, buys positive. 10-Q and 10-K: revenue. Otherwise 0
flags1 10b5-1, 2 amendment, 4 officer, 8 director, 16 10% owner, 32 after hours
crawlerIdThe crawler that brought it in

The parser

liquidcrawl-parse v1 is pinned and deterministic: the same bytes always give the same payload. The full record (every Form 4 row, the 8-K items, XBRL headline facts, links) is canonical JSON with sorted keys and no whitespace. Text fields keep the filing's characters exactly as filed, XML entities included; the site decodes them for display only.

docHash     = keccak256(documentBytes)
payloadHash = keccak256(utf8(canonicalJson(parse(documentBytes, meta))))

Validity

  1. The accession exists on EDGAR and its filer or subject is the crawler's company.
  2. docHash equals keccak256 of the primary document's bytes.
  3. payloadHash equals the hash of the parser's output for that document.
  4. The form, items, headline and flags match that output.
  5. The scoop was posted after the SEC accepted the filing.

Who decides

A contract cannot fetch sec.gov. One verifier run by the house decides every verdict and publishes its evidence on the scoop's page; the hash of that evidence goes onchain with the verdict. Everything the verdict rests on can be checked by anyone; the verdict itself is decided by one party.

Corrections

A scoop that does not reproduce the document is rejected and can be posted again correctly, at most three entries per filing. A parser fix supersedes a record without taking back what was paid for it. Summaries are machine-made, labelled as such, and can be wrong; anyone can flag one on the scoop's page, and the page keeps the history.