METHODOLOGY

How DEHY gets the data.

Every value on every page traces back to a specific field in the filer's SEC Form 4 XML. No third-party data. No inferences.

Source

The U.S. Securities and Exchange Commission's EDGAR system. We use two public endpoints:

  • Recent filings Atom feed — browse-edgar?action=getcurrent&type=4&output=atom. Polled every 30 seconds for new Form 4 filings.
  • Primary document XML — /Archives/edgar/data/{cik}/{accession}/...xml. Fetched once per filing, parsed once, stored.

Latency

Discovery runs every 30 seconds. Extraction is queued immediately on discovery. In practice the median end-to-end latency from EDGAR publication to your feed is well under 60 seconds. We do not guarantee an SLA on the free tier.

Form 4 fields we extract

Every Form 4 page on DEHY reflects these fields parsed from the SEC's XML, never inferred:

  • Issuer name, CIK, and trading symbol
  • Reporting owner name, CIK, and relationship (director, officer, 10% owner, officer title)
  • Per-transaction: security title, transaction date, transaction code, shares, price per share, acquired/disposed code, post-transaction holdings, direct/indirect ownership
  • Derivative transactions: underlying security, conversion price, exercisable and expiration dates
  • Footnotes as filed

What we do not do

  • We do not infer the "real" intent of a transaction. A 10b5-1 plan sale and a discretionary sale both appear as code S. We surface the code; you draw the conclusion.
  • We do not adjust prices for splits or dividends. The price shown is the price reported on the filing.
  • We do not enrich with non-EDGAR data sources (price feeds, news, social signal) on the free tier. Pro will add explicit, sourced enrichments.
  • We do not generate "buy" or "sell" signals. Form 4 is a disclosure of fact, not an opinion.

The DEHY Score

On top of the raw fields, every Form 4 carries a reproducible 0–100 score. It is a deterministic function of five sub-factors, each itself 0–100, weighted into the headline:

  • Historical pattern (25%) — first-time filers score high; insiders who file frequently are damped.
  • Magnitude (25%) — shares transacted as a share of the insider's total post-trade holdings.
  • Role weight (15%) — CEO / CFO carry the most signal; directors and other officers, less.
  • Plan dampener (20%) — sales filed under a 10b5-1 plan are treated as scheduled and damped; discretionary sales score full.
  • Cluster signal (15%) — other insiders trading the same direction within fourteen days amplify the score.

The algorithm is versioned ( algoVersion) so future refinements can be replayed across historical filings without losing the original score. Same inputs always produce the same output — we do not call out to an LLM to score Form 4s.

Errors and corrections

EDGAR data occasionally contains errors that filers later correct via Form 4/A (amendment). We ingest amendments as separate filings and link them to the originals. We do not silently overwrite an original filing with its amendment.

Retention

Every filing we ingest is kept indefinitely. Historical context — a reporter's previous filings at an issuer, their net position change over time — depends on this.