Research

How research works

Two waves across public registers and the company's own filings, with provenance on every field.

Two waves, in that order

Why registers are keyed by name

A register cannot be asked who owns a domain. So the first wave reads everything reachable from the domain itself, including the legal notice, which by law states the legal name, the register number and the VAT ID. The second wave feeds those identifiers to the registers, which turns a fuzzy name search into an exact lookup. That is the difference between attaching the right parent company and attaching a stranger's.

Wave 1: from the domain

  • Legal notice
  • Website
  • Job boards
  • Technology
  • Mail platform
  • GitHub
  • Wikidata
  • Domain registration
  • Wayback

Identifiers

  • Legal name
  • HRB number
  • VAT ID
  • LEI

Wave 2: by identity

  • GLEIF
  • VIES
  • Companies House
  • SEC EDGAR
Dossier with provenance

Wave 1: from the domain

Everything reachable from the domain itself, read before any register is asked.

Impressum
Legal name, register number and VAT ID, which the legal notice states by law
Website
Description, people on team pages, locations
Job boards (six ATS)
Open roles, functions, hiring velocity
Technology fingerprint
The stack the site runs on
Mail platform
The mail provider behind the domain
GitHub
Engineering activity
Wikidata
Verified identifiers, including the LEI
Domain registration
Age of the domain
Wayback
History of the site

Wave 2: by identity

The identifiers from wave one turn a fuzzy name search into an exact lookup.

GLEIF
LEI, parent and child entities
EU VIES
VAT confirmation
UK Companies House
Officers, SIC codes
US SEC EDGAR
Filings, XBRL
Live research running on a company, wave by wave

The provenance model

Every established field is stored as a record, not a value: what was found, where, at which URL, with what confidence, and when it was observed.

{
value,
source,
url,
confidence,
observed_at
}

Not established

Every field no source could confirm is listed under "not established" on the dossier. The list is part of the output, not an absence from it.

Precedence when sources disagree

  1. 1Register0.95
  2. 2Impressum0.90
  3. 3Wikidata0.70
  4. 4Website0.60

Refusals

Ambiguous name matches are refused rather than guessed. A single-word name without a country is refused. A refusal is recorded on the dossier with the reason, so you know what was tried.

Scoring

Fit
Computed against your profile, renormalised over the criteria you set, with a reason per criterion. Null without a profile.
Signal
Dated evidence with a 60-day half-life. Saturating rather than capped: the tenth signal still counts, a little.
Tier
Strike, engage, nurture or watch, set by fit and signal together.

What we do not do

  • No purchased contact lists
  • No scraping of private profiles
  • Person research limited to professional roles
  • EU-hosted
  • GDPR-first
  • DPA on every plan
Visit the trust centre

See Caicle on the accounts you actually sell to

Start a free trial in minutes, or send us three domains and we will run them live with you.