Source ratings and evidence availability are different measures. Neither is a guarantee of broker quality or suitability.
01 · What is rated today
The current release reports the numeric Trustpilot rating shown on each linked brand profile, on its original 1–5 scale. We retain the source’s total published review count and observation date. We do not convert the score into a satisfaction percentage or present it as an FXNumbers assessment.
TrustScore uses the provider’s weighting methodology; it is not a simple arithmetic average. The headline count is not its effective statistical sample. We therefore do not calculate a confidence interval from the score and total count alone.
FXNumbers does not yet publish a composite rating, a trend, a cross-source sentiment score or entity-specific assessments. A later method will receive a new version before use.
02 · The five-segment evidence indicator
The indicator is an editorial coverage heuristic, not statistical confidence. It measures how much of our declared evidence requirements is available. It does not change the source rating.
E = 100 × min(V, S, F)
V · Reported volume
min(N / 1,000, 1), where N is the published review count for the displayed source. Reviews are not assumed to be unique customers.
S · Source coverage
min(J / 3, 1), where J is the number of distinct, eligible review-platform publishers actually recorded for the brand. Syndicated copies do not count as new sources. The current release has J = 1 for every brand; this is not proof that reviewers are independent.
F · Observation freshness
1 through 7 days; 0.5 from 8–30 days; 0.25 from 31–90 days; 0 after 90 days. Age refers to when the aggregate was checked, not when individual reviews were written.
The thresholds are editorial design choices, not empirically validated cut-offs. The weakest component caps the result. Current single-source profiles cannot exceed 33.3%, however large their review count. Segment i fills by clamp(5 × E/100 − i, 0, 1), for i = 0…4.
Labels: 0% = unavailable; above 0 and below 50% = limited; 50–under 80% = developing; 80–100% = broad coverage. Even a full bar does not establish representativeness, absence of manipulation or broker safety.
The displayed indicator is calculated for the page’s publication snapshot. A freshness notice appears if the observation is older than seven days; the observation date always remains visible. Republishing without a new observation never resets that date.
03 · Small samples and missing data
We show a valid published source rating even when its sample is small, alongside its count and limited evidence indicator. If a provider does not publish a valid rating, we display “Not available”; we never fabricate one. Zero evidence is not a zero broker score.
For a future dataset with exact favourable count k and total n, an observed favourable share may be calculated as k/n with a Wilson interval, provided its assumptions and sampling limitations are stated. This estimator is not applied to today’s weighted aggregate ratings or rounded star distributions.
04 · Provenance and initial coverage
IG, OANDA and Pepperstone form a small, non-sponsored editorial starting set with identifiable public review profiles. They are not a complete market universe, “best broker” list or representative sample. Default ordering is alphabetical. User-selected sorting applies only within this set.
IG on Trustpilot — 3.7/5; 10,023 published reviews; observed 12 Sep 2026.
Only the observed numeric aggregates and source references are stored. No review text or reviewer identity is retained. Each source profile reports a paid Trustpilot subscription; that is not sponsorship of FXNumbers. Public brand profiles may combine entities, services and jurisdictions.
Current profile selections and homepage placements are not sponsored. Inclusion does not imply affiliation, approval or endorsement by a broker. Future paid visibility must be clearly labelled and must not alter source values, the evidence formula or organic sorting. Corrections are assessed against evidence, regardless of commercial status.
06 · Updates, corrections and automation
The launch observations were checked editorially on 12 September 2026. Database export, evidence calculation and page generation are automated commands. Recurring collection from external platforms is not yet enabled; it requires an approved feed or reuse arrangement. We do not describe these values as live.
Failed acquisition must retain the last valid observation and its original date. New observations are stored as separate snapshots. Changes to formulas or thresholds require a new methodology version; recalculated results must be labelled accordingly.
Send source links and the disputed field to [email protected]. We review corrections before updating the published snapshot.
07 · Website privacy
Public browsing requires no account and uses no advertising trackers or analytics scripts. The editorial admin uses Cloudflare Access. Private broker invitations store the named contact email, proposal, review status and activity history in Cloudflare D1; uploaded images are stored in R2. Invitation links expire and can be revoked. Proposals remain private until editorial approval and publication. Contact FXNumbers to request correction or deletion of your submitted information. Broker and source links take you to external websites with their own policies. The hosting provider processes ordinary request information to deliver and protect the website. Email links open your email application; sending an email shares the information you choose to include.
