Editorial policy
Editorial policy for the Bookmaker Atlas
1,869 brands profiled2026-06-10 last full recomputeMonthly recompute cadence0 hand-tuned brand scores
The policy at a glance
- No operator pays for placement, score adjustments, or removal of negative information.
- Scores are the deterministic output of the published pipeline; nobody hand-tunes a brand.
- Material corrections are logged with date and affected pages; cosmetic fixes ship silently.
- Corrections need a verifiable source: regulator filing, court document, or official disclosure.
Spotted an error? The corrections route takes you through it.
Who edits the Atlas
The verdict on every brand is the deterministic output of the scoring pipeline. The same code, against the same sources, on the same monthly cadence. There is no senior reviewer who can hand-tune one operator's rating.
Reviewed by Atlas Editorial
The byline on every leaf reads “Reviewed by Atlas Editorial” because the verdict is code, not opinion. The data side is engineered and audited by Erik Andersson, who has built the pipeline from the ground up. All 1,869 brands are scored in the same build pass.
No operator pays for placement
Scoring weights and red-flag overrides live in a private repository, reviewed at every change. No field is reachable from a CMS or admin panel. If a brand wants their score reconsidered, the path is the same as any reader's: send evidence and we check it against the pipeline.
Update cadence
Different signals move at different speeds. Each leaf shows its own “Updated” date so you can see exactly when each fact last refreshed.
Monthly
Full Trust Score v2 recompute
Latest recompute: 10 June 2026.
Weekly
Regulator-register refresh
Local regulator registers, enforcement actions, and aggregate public-reputation trends.
Daily
DNS, security, threat intel
Email-authentication posture, DNS hygiene, web-security scans, and certificate-transparency activity.
On change
Corporate ownership
Refreshes when public corporate-ownership and sanctions databases publish a change. Can land any day.
Lag exists. A licence revoked yesterday may not appear until the next refresh of that source.
How fact-checking works
Every claim traces back to a public source you can verify yourself. No paraphrasing. No interpolation. No padding sparse data with generated prose.
Deterministic ingest only
Every claim on a leaf page is traceable to a JSON file in our data layer. No prose is generated by language models. No facts are paraphrased away from their source.
Multiple-source cross-checks
Licence claims validate against the regulator's official register. Corporate ownership against multiple public corporate-registration databases. Security and threat signals against independent web-security scanners and threat-intelligence feeds.
Public sources only
No private databases, no paid leaks, no operator-supplied data. Every source is something a determined reader could verify themselves.
Sparse data is shown sparse
When a brand has no public reviews, no security audit, or no regulator registration, the page says so explicitly with the audit trail. We never invent or interpolate.
Corrections policy
Mistakes happen. When they do, the fix is structural, not cosmetic.
Send the correction with a verifiable source: regulator filing, court document, primary news article, or official disclosure.
If the underlying source data was wrong, we fix the data file. The correction propagates to every page that references it on the next build.
If the source was correct but our pipeline misinterpreted it, we fix the pipeline. Same propagation.
Material corrections (licence tier change, ownership chain update, false enforcement attribution) get logged with date and affected pages.
Cosmetic typos and broken links are fixed silently in the next build.
How to report a correction
Send corrections via our contact page. Include the page URL, the claim you want reviewed, and the source you would like us to weigh against ours.
We acknowledge every report. We do not promise to act on every report, but we do promise to read every report and document why we declined any we declined.
What we will not do
Two policies that the industry rarely puts in writing. We put them in writing.
AI and content generation
Page narratives are generated deterministically from verified data fields: numbers are computed by the scoring pipeline, entity names come from licence registers, dates come from source timestamps. No language model is asked to invent or paraphrase facts. AI tools are used as engineering assistants for code and pipeline work, with human oversight at every step. Every visible sentence traces back to a data field a human reviewed.
Conflict of interest
SharkBetting earns affiliate revenue elsewhere on the site (oddsmatcher signups, exchange referrals). Atlas pages take no affiliate kickback and no operator pays for placement here. The scoring code is blind to commercial relationships: the same code runs on every brand, and component weights are not adjusted per brand. Operators we earn revenue from elsewhere can and do score lower in the Atlas than operators we do not.
We do not accept payment to remove unfavourable information, alter scores, or downplay enforcement actions. Requests of that kind get forwarded to the operator's regulator if they include implicit pressure.
Editorial standards we hold ourselves to
Five lines we will not cross, regardless of who is asking.
Every quantitative claim on a leaf page must be traceable to a JSON file in the repo.
Every regulator-tier classification must be sourced from the regulator's public register, not a third-party aggregator.
Every enforcement-action mention must link to the regulator's notice, not a press summary.
Empty data is shown empty. Sparse leaves are not padded with generated prose.
The same template renders all brands. There is no "hero treatment" reserved for friendly operators.
Frequently asked questions
Do bookmakers pay for placement in the Atlas?
No. No operator pays for placement, score adjustments, content edits, or removal of negative information on Atlas pages. SharkBetting earns affiliate revenue elsewhere on the site (oddsmatcher signups, exchange referrals), but Atlas pages take no kickback and the scoring code is blind to commercial relationships. Operators we earn revenue from elsewhere can and do score lower in the Atlas than operators we do not.
Do you use AI to write bookmaker reviews?
Page narratives are generated deterministically from verified data fields: numbers are computed by the scoring pipeline, entity names come from licence registers, dates come from source timestamps. No language model is asked to invent or paraphrase facts. AI tools are used as engineering assistants for code and pipeline work. Every visible sentence traces back to a data field a human reviewed.
How do I report a factual error?
Use the contact page with the page URL, the claim you want reviewed, and a verifiable source (regulator filing, court document, primary news article, or official disclosure). If the underlying source data was wrong, the correction propagates to every page that references it on the next build. We acknowledge every report.
How often is the Trust Score recomputed?
Every brand is rescored monthly in one deterministic build pass. Regulator-register data refreshes weekly. Security, DNS, and threat-intelligence signals refresh daily. Corporate-ownership changes land on the day the underlying public source publishes them.
How is editorial independence enforced?
Scoring weights live in a private repository and are reviewed at every change. No field is reachable from a CMS or admin panel, so no individual reviewer can hand-tune one operator's rating. Every brand is scored by the same code, against the same sources, on the same cadence.
See also: how the Atlas dataset is built, and the full Trust Score methodology.
Read deeper, or browse the data
The methodology page walks through the scoring math line by line. The master hub lists every brand we track.