From evidence
to a rating.

Provisional editorial calibration · effective 12 September 2026.

60% data, 40% editorial · blend-pilot-1.0

Combined percentage = 60 × (data score / 5) + 40 × (editorial score / 100). Divide by 20 to display /5. The original data score is unchanged. Missing data or editorial input yields no combined score: weights are not silently redistributed.

The initial test used an explicitly simulated editorial input of 50% for each broker, equivalent to 2.5/5 and contributing 20 percentage points. The admin can replace it with a documented criterion assessment. The public page labels test and completed-review inputs separately. Test records are stored separately with their status and review deadline; when they expire the combined score is unavailable until replaced.

The proposed editorial rubric is: public review evidence 30%, navigation and usability 25%, clarity of pricing and information 30%, company and document visibility 15%. These are the currently published weights (revision 0). A real review must record reviewer, date, sources, scope and reasoning. Brand-wide sentiment must not be mislabelled as jurisdiction-specific, nor counted in both components.

Market and regulatory jurisdiction

For traders selects a residence, not the location of the broker. Only explicitly verified residence-to-entity mappings with a current review date are eligible. By regulatory jurisdiction selects the contracting entity’s regulatory context. A missing mapping is under review, never presumed unavailable. Both perspectives retain the common-parameter comparison. Overall is available only in the regulatory view; duplicate legal entities are included once, using the first common context in the published jurisdiction order (UK, USA, Singapore, Hong Kong, Japan, Germany, Australia, Canada, UAE, South Africa, Cyprus). This is not a client-weighted global average.

Country coverage policy · 12 September 2026

For a single country, brokers with no current usable observations remain unrated; they do not suppress the documented scores of other brokers. All rated brokers in that country use identical shared parameters and weights. If only one broker has evidence, its result is explicitly a standalone partial rating, not a peer comparison. Parameters may differ across countries, so country scores are not directly interchangeable. Overall retains the full displayed-brand common-jurisdiction requirement and does not average different country sets for different brands.

1. Conversion to /5

Advertised EUR/USD minimum spread: A = 0.5 pips receives 5/5; B = 2 pips receives 1/5. Score = 1 + 4 × (B − value) / (B − A), clipped to [1,5]. These are editorial pilot anchors, not statistically validated market percentiles. They stay fixed within this version and do not depend on the three displayed brands.

Inactivity policy: 5/5 for no stated charge, 1/5 for a stated charge. This narrow policy indicator does not assess fee size, exceptions or total account cost, which remain visible beside the source.

2. Weights and partial ratings

Within the 60% data component, five dimensions each have 20% planned weight: Trading costs, Client experience, Funding experience, Service response and Digital experience. Cost weights: minimum spread 10%, realised average spread 40%, commissions 20%, overnight cost 20%, inactivity policy 10%. Weights are editorial choices, not empirical estimates.

Available common parameters are normalised to sum to 100% for the partial score. Minimum spread and inactivity therefore each contribute 50% of that partial score, but together represent only 20% of Trading costs and 4% of the planned data framework. With minimum spread alone, data-framework coverage is 2%. The high effective weight of these narrow indicators is a limitation of this pilot, not a claim about their overall importance.

3. Identical comparison set

Scope: retail FX CFDs, standard spread-only pricing. Within the geographical selection, first select parameters with valid observations for every displayed brand in at least one jurisdiction. Then intersect those parameters’ common jurisdictions. The resulting brands × jurisdictions × parameters rectangle is identical for every score. If empty, show no comparable rating. Jurisdiction score = weighted mean of converted parameters; Overall = equal-weight mean of included jurisdiction scores. Convert before averaging.

Changing geography can change available parameters. The interface shows the included set and effective weights; scores with different sets are not a like-for-like trend. Missing is never zero.

4. Evidence

Each observation has an entity, jurisdiction, source URL, conditions, date and review deadline in the database. Only published, current, matching-scope observations are eligible. Completeness measures available planned weight, not statistical confidence. Published terms are not execution measurements. Initial source collection is manually reviewed; calculation and expiry are automatic. Observations expire after 30 days unless reviewed.

5. Calibration still to complete

A wider homogeneous benchmark is required before replacing pilot anchors with empirical references. Initial source checks encountered dynamic tables, differing account types and historical periods; these were not combined into a purported statistical sample. No empirical calibration is claimed. Future threshold changes require a new version and published benchmark derivation. Commercial placement never changes scores.

Trustpilot brand-level data remains separate and is never assigned to jurisdictions. This partial score is not an overall endorsement, safety assessment or typical trading-cost estimate.