Methodology
How Bixel counts, verifies, and proves
Bixel publishes verified private company and market data, with receipts. This page defines every number we publish about ourselves, what each tier of coverage claims, and how to check us.
The numbers
The numbers, with their definitions
— tracked companies. A tracked company is an apex domain where our mechanical checks verified a live, company-shaped website: the domain resolves in public DNS, a real server answers over a valid security certificate, and the page served is genuine content rather than a parking template or an empty shell. A deliberately fake subdomain is tested as a control, so a domain that answers yes to everything is caught. Every tracked company carries the dated observation records behind its status, and the discovery receipt naming the public archive queries that surfaced it. Nothing else is claimed about these companies. No human or AI judgment is involved at this tier.
14,341 verified members. A member passed an individual verification: the site’s actual content was read and judged to be a real operating software company rather than an agency, a lookalike, or a dead project. Each membership carries a receipt recording what was read, when, and under which version of our verification standard. Membership currently covers software companies, because software is the field our measurement framework covers most deeply. The definition is designed to extend to other industries as measurement frameworks for them are built.
Deep coverage is the smallest tier: companies covered across every dimension we measure, refreshed on a fast cadence with full change history. Promotion is driven by demand for the company’s data, not by our guesses.
Companies that die are marked dormant or dead in public view, with dated evidence. A count that only grows is an obituary pile, not a dataset. Our tracked count is a count of companies verified alive on a defined cadence, and the die-off is part of the published data.
Claims
What each tier claims, exactly
A tracked company claims: a live, company-shaped website existed at this address on this date, and here is why we track it. A verified member adds: a person-auditable reading judged this a real software company, on this date, under this standard. Deep coverage adds the measured dimensions, each fact with its own evidence. No tier ever claims more than its evidence supports, and a tracked company publishes zero facts.
Honesty
How verification stays honest
Judgment is minimized: free mechanical evidence decides most cases before any paid reading happens. Judgment is audited: a fixed percentage of every automated decision lane is independently re-checked by a stronger reader, continuously, and the agreement rates are published below, including when they drop. Judgment is gated: every pipeline change must pass a fixed, human-ratified reference set before it ships. Humans review samples: every verification batch ends with a human spot-review before new members are minted.
Accuracy scorecard
Membership verification (free-admit lane)
A fixed share of automated membership admissions is independently re-read by a stronger model, continuously. Agreement means the second reading reached the same admit decision.
99.1% agreement across 117 independently audited decisions (2026-07-27 to 2026-08-16).
These numbers render from the recorded audit rows at page build time. They are never typed in, and history stays published when a rate drops.
Receipts
Receipts, and what they prove
Every fact points at captured evidence, carries its date, its source, and whether it was independently observed or stated by the company itself. Evidence is content-addressed and append-only: corrections are new dated records, never edits, and wrong facts appear in our public corrections log with reasons.
Our evidence logs are hash-chained and anchored to the Bitcoin blockchain through open timestamping. Anchoring proves the records existed by a given date and have not been rewritten since. It does not, by itself, prove the records are true. Truth comes from reproducibility: the checks are mechanical, the underlying archives are maintained by independent parties, and anyone can re-run them. We state both halves because the distinction matters.
What we have not verified is marked unknown, and unknown never quietly becomes fact. When a company claims something about itself, the claim is labeled as a claim; where we independently observe supporting evidence, the data shows claimed versus observed side by side. A company’s legal name starts as the company’s own statement and is upgraded through explicit, evidence-backed levels of corroboration. The level always shows.
There is deliberately no per-fact confidence score anywhere in the product. A number between 0 and 1 invented per fact would be decoration. Evidence states and the measured accuracy above are the real thing.
Freshness
How fresh the record is
Every fact displays the date of the capture that last supported it. A fact whose page we re-observed recently is current as of that look, even when nothing changed; stability is data, not staleness. When our own re-observation window for a dimension passes without a fresh look, the record says so in place, dated. These are the windows we hold ourselves to:
| Dimensions | We look | Verification note appears after |
|---|---|---|
| Pricing, features, positioning, stack | twice weekly | 90 days without a look |
| Hiring, reliability | every 30 days | 60 days without a look |
| Security, legal terms | every 90 days | 180 days without a look |
| Funding (filing-derived) | on filing or announcement | never (dated events) |
| Legal identity, location | yearly | 730 days without a look |
Conduct
How we read the web
Crawler rules are honored. Sites that block us are recorded as blocked, never evaded. Takedown requests are answered. Most of our strongest signals require no page access at all: they come from public DNS, certificate infrastructure, regulatory filings, and independent archives.