Methodology and verification sources

How to read an operator page

• Last verified is the day a fact on the page last changed when we re-read its source; the content hash beside it identifies that exact set of facts.
• First-hand tested appears only when at least one published value was read from our own capture of the operator’s pages, by the desk or reviewer named on the page.
• Independent sources counts the review sites we consulted for that operator; each is named, dated and linked where it is cited.
• The fact table at the end of a page lists every published fact with its source, the day it was recorded, the day the source re-confirmed it unchanged, and whether it is inside its freshness window. A unified profile (below) carries no fact table: its Sources block names each lane once instead.

Unified operator profiles

Some operator pages take a unified layout that merges every lane we hold for that operator into a few reader-facing modules. The method behind each module lives here, not on the page.

• The summary cards at the head of the page show the minimum deposit, the withdrawal time, the payment methods and the mobile app rating. Each card is one lane’s own figure, named on the card: a first-hand capture, a stated figure on the operator’s own banking page, or a review site’s own listing. Where no store listing exists the fourth card shows the game types instead. Nothing is averaged across lanes.
• One banking table unites the payment methods named by up to three lanes: our first-hand capture of the operator’s pages, the operator’s own banking page as our capture lane read it, and a review site’s Ontario page. Each method appears once. A direction is marked only where a lane states it; a dash means not stated, never not accepted. An operator’s umbrella word for card payments is folded onto the card schemes it names elsewhere, so a reader never meets the same rail twice under two names. The lanes behind the table are named once, in the Sources block, with the day each was read.
• Game types and studios are chips, each carrying the lanes that evidence it: the first-hand fact, the sections the operator’s own lobby page names, the register’s own product row. A lobby section that names no vertical adds nothing.
• What players report is a short summary of patterns in the review-site data we hold for that operator: the sites’ aggregate figures, their complaint counters and staff summaries, a handful of player reviews, and one site’s section scores. It is drafted with the claude-sonnet-5-5 model, in one request, from that data alone. Before it is published it must pass our checks: no inducement vocabulary, no ranking word, no figure at all, no link, and no run of words copied from any source we consulted. A draft that fails is not published, and the record names the sites and snapshot days it rests on.
• Independent review sites are shown one block per site, on that site’s own scale, with its own section scores where it publishes them, our counts over its player reviews, and its complaint counters once. No figure is converted or averaged with another site’s.
• Sources, once. The Sources block lists the register, each data source, the merged lanes with their dates, the review sites consulted and the archived capture set, each exactly once, and closes with the link to this page.

The sources, lane by lane

The iGaming Ontario register

Registration status, the registered operator entity and the official website come from the official iGaming Ontario register and from nowhere else. The register is re-read daily; a fact that re-reads unchanged keeps its original date and gains an “unchanged at” day.

First-hand capture

Our research desk captures each operator’s own public, pre-login pages in journey order — homepage, lobby, payments, support, terms — and every published first-hand value is verified against a verbatim quote in that capture. Account, identity and residency gates are respected and never bypassed, and no real-money play takes place. The full procedure is on the first-hand testing page.

The operator’s own pages, as published

A separate capture lane reads an operator’s public lobby page, banking page and, where one exists, its published return-to-player sheet, and records what those pages said on the day they were read: the sections and studios a lobby names, the payment methods and stated limits on a banking page, the RTP a studio publishes for a game. These readings sit beside the first-hand record, never inside it, and are always dated and named by host. A stated processing time is recorded but not published, because a headline is not a measurement.

Independent review sites

We consult independent review platforms, attribute them, and never reproduce them. A published rating is shown as that site’s own figure on its own scale with the day we read it and a link to the page we read; we never convert a score onto another scale and never average two sites together. What we work out ourselves — counts over a site’s full review set, complaint statuses, the patterns a set of reviews shows — is labelled as ours. Where a short summary of what players report is drafted with the help of a language model, it is drafted only from the review data we hold for that operator, names the sites it draws on, carries no figure of its own, and is checked before publication against our vocabulary rules and against the text of every source we consulted, so no sentence of a source is reproduced.

Public records

Where a registrant belongs to a listed group, the parent filer’s public filings give context — the group, its listing and its latest reports — and the registrant’s domain registry record and public DNS records are read as published. Each line names its source and the day it was read. None of it is a rating.

What we read but do not publish

Some sources we read only to check our own work: app-store listings, legal-entity identifiers, breach registries and general knowledge bases. Their licences or their nature make them suitable for internal quality assurance, not for publication, so no figure from them appears on any page.

Freshness

Every fact carries a freshness window set by how fast it moves: register facts are re-read daily, first-hand facts are recaptured on a fixed cadence, and independent review readings are re-read on a 35-day cycle. A reading older than twice its window comes off the page rather than being shown as current; a source whose readings have all aged out comes off whole. The fact table marks each fact fresh or stale against its own window.

Corrections

Our crawler identifies itself as PerceptBot and carries a contact address in its user-agent string on every request. Write to it to have a reading corrected, to change how we attribute a source, or to ask us to stop consulting a site. A withdrawn source is removed on the next daily build.

Read the first-hand testing methodology →