Methodology

How this research is produced

Every claim on this site traces back to a primary or strong secondary source. Here's how company profiles are researched, verified, and written.

Source standards

No claim without a source

Every operational fact — how a company structures meetings, makes decisions, or handles conflict — must link to a primary source (official documentation, founder writing, verified employee accounts) or a strong secondary source (credible journalism, conference talks, books with named attribution).

Statements that can't be directly sourced are labeled as inferences and placed in a clearly marked section at the end of each profile.

Source quality scale

Four tiers of evidence

Primary

Official company documentation, employee handbooks, founder-written content, public roadmaps, internal documents shared publicly.

Strong secondary

Credible journalism with named sources, conference talks by company leaders, books with direct attribution, verified podcasts.

Weak secondary

Aggregated reviews, indirect reporting, industry analysis without primary access.

Tertiary

Opinion pieces, anecdote-only accounts, unnamed sources — used for context only, never as evidence for operational claims.

Research workflow

How a profile gets built

Source collection

Gather 8-15 primary and strong secondary sources per company. Build a source pack before any writing begins.

Evidence extraction

Pull direct quotes, operational facts, and examples from each source into a structured evidence table.

Profile structuring

Organize evidence into the 14-section company profile schema — snapshot, philosophy, communication model, planning cadence, and more.

Writing

Draft the profile from the evidence table. No invention, no generic advice, no filler.

QA review

Every claim is cross-checked against its source. Inference labels are applied. The profile is either approved, revised, or rejected.

Show the receipts

Inside a QA review

Every published company page on this site has a corresponding QA review file on record. The reviewer (the QA agent, plus an editorial pass) walks the page claim by claim against the source pack, marks each as Supported, Supported with edit, or Inference (labeled), and notes any reconciliation the page needed to ship. The format below is a real excerpt from the GitLab review — abridged to five representative rows out of thirty-one.

#ClaimStatusSource / Note
1"GitLab has 1,500+ team members across 65+ countries with no offices"Supportedhandbook.gitlab.com/handbook/company/culture/all-remote/guide
6Three-exchange rule: move to call after 3 async exchangesSupportedhandbook.gitlab.com/handbook/company/culture/all-remote/asynchronous
12Cadence: weekly / monthly (~4x per month) / quarterly OKRs / annual / 3-year strategySupported (with edit)Published page read "~4x per week" — corrected to "~4x per month" during reconciliation.
16"If it's not in the handbook, it doesn't officially exist."Supported — direct quotehandbook.gitlab.com/handbook/company/culture/all-remote/handbook-first-documentation
29Stratification at scale (false intimacy from flat-feeling tools)Inference (labeled)Page-level inference — appears in Inferences section, not body.

Two operating rules fall out of this format. First, claims marked Inference only appear on the published page if they are also clearly labeled there — the reader sees the same label the reviewer wrote. Second, claims marked Supported with edit trigger an actual page edit; the QA review is the place where errors get caught before publish.