Methodology

Every number on EarlySeed traces to a source you can click. This page states exactly what we cover, what we do not, and what we measured about our own accuracy — including the parts that are unflattering.

Where the data comes from

The spine is SEC Form D filings (tier T0 — legally mandated, structured, registry-timestamped). A Form D is generally filed within 15 days of a first sale of securities. We poll EDGAR, parse filings deterministically, and never let a language model originate an amount, date, name, or investor. Web research (company domains, press coverage) is layered on as enrichment with its own tier and timestamps.

The founding measurement (2026-07-29, 100-round golden set)

• Form D → company match rate: 67% — two-thirds of early-stage filers could be confidently matched to a real company website via officers and location.
• Median lead time vs press: −0.5 days. Our original hypothesis — that filings systematically beat press by 5+ days — was false for rounds that get press, and we retired that claim. Announced rounds are announced at close; the filing follows.
• 85% of matched early-stage rounds had zero press coverage at all. Not late coverage — none. That is the actual reason this product exists: most early rounds are invisible outside filings.

Stage classification

Form D carries no round label, so every stage is inferred, marked as such, and auditable (rule stage-rules-v3): Pre-seed ≤ $1.5M · Seed $1.5–6M · Series A $6–30M, with ordinality bumps from known prior rounds. The amount basis is amount sold when real money moved, else total offering. Structurally excluded at publish (physically absent from the data your browser downloads): pooled investment funds, non-equity offerings, business combinations, real-estate/banking/insurance/resource-extraction offerings, listed-issuer and investment-vehicle name-pattern entities (funds, SPVs, series vehicles), anything above $30M, and any 3rd+ priced round.

Amounts are three separate facts

Total offering (what the filing permits), amount sold (what had been sold at filing time), and press-reported figures are stored and displayed separately, never blended into one "raised" number. "Indefinite" offerings stay Indefinite. Unknown values render as "not found" — we never round a gap into a guess.

Announcement status is three-state

A round shows ⚡ Unannounced only when we actually checked for coverage and found none. Rounds we have not yet checked say "not yet checked." We do not claim silence we have not verified.

Known coverage limits, stated plainly

• US-only registry at present (UK Companies House is the next T0 source; non-US coverage is currently signals-only).
• Not all raises file a Form D; some rely on other exemptions. Form D is a floor, not a census.
• Filing lag: up to ~15 days after first sale, plus issuer behavior.
• Domain matching currently covers the audited golden sample; most companies show "domain not found" until enrichment widens. Match rate on the audited sample: 67%.
• Officer names come from filings in a business capacity. We store name, role, and public link only — never email, phone, address, or scraped biographies. Removal requests are honored.

Crawl conduct

Identified user-agent with a contact address, ≤5 requests/second against EDGAR (their policy allows 10), robots.txt respected, no auth walls or paywalls ever bypassed, immutable raw store so every published value can be replayed from its original document.

EarlySeed describes what is knowable from public sources. Nothing here is investment advice, a recommendation, or a score of company quality.