Holder TrustScore and review ratings methodology

Inputs to the score

The first source is the base exchange score supplied by a data provider. Holder divides it by two: 8 out of 10 becomes 4 out of 5. This is a provider metric, not ten actual reviews.

The second source is numerical signals from published reviews on the relevant language site. The calculation uses top-level reviews with a valid value from 1 to 5. Thread replies, unpublished entries and text without a numerical signal do not add a vote to this calculation.

Each review contributes once. An explicitly submitted author rating takes priority. If there is no author rating, text sentiment analysis may be used; otherwise a retained historical score of unknown origin may contribute. Only the first group represents a numerical choice made by the review's author.

Formula and rounding

TrustScore = (A × N + G × W) / (N + W).

• A is the mean of the review signals, first rounded to one decimal place.
• N is the number of included signals.
• G is the base score converted to a 5-point scale.
• W is the weight of the base score. The default is 10; the actual weight is shown in the score explanation on the page.

The result is also rounded to one decimal place. If the base input is missing or zero, its weight becomes zero. With no review signals, only the base score remains. With neither a base score of positive weight nor review signals, no TrustScore is calculated.

Illustrative example. An exchange has a base score of 8 out of 10, so G = 4. Two signals of 5 and 1 give A = 3. With W = 10, the calculation is (3 × 2 + 4 × 10) / 12 = 3.83, displayed as 3.8. This is a worked example, not the result for a particular exchange.

How an exchanger score is calculated

For an active exchanger, Holder calculates the base score G from stored profile data. This is an editorial model, not a CoinGecko score or a user vote. It starts at 3.8.

• Years in operation: +0.05 from 3 years, +0.10 from 6 years, or +0.15 from 9 years.
• Exchange directions: +0.05 from 300, +0.10 from 3,000, or +0.15 from 10,000.
• A rate feed URL: +0.10; a feed fetch less than one day ago: another +0.05.
• Cash directions in at least three cities: +0.05.
• A populated reserve field: +0.05. A manual-check flag for reserves or the starting year: another +0.05.
• No locally stored logo: −0.10.
• An editorially verified representative reply on an external platform: +0.3. The reply URL and company identity must be recorded; both the reply and its verification must be no older than 90 days. A promotional post without a reply or an unverified claim earns no bonus. This confirms public communication, not response speed or successful resolution.

Only one increment applies in each tiered group. The result is bounded to 3.8–4.4 and rounded to one decimal place. Its weight in the shared formula is W = 5; each numerical review signal has a weight of 1. An inactive exchanger has no base score: only the average of available review signals remains, if any.

Reserves and directions may be reported by the exchanger itself. A feed, logo or reserve entry does not establish solvency. Affiliate compensation is not an input.

What AI analysis means

Text without author-selected stars may be assessed by a language model. The current analyzer maps sentiment onto a 2–5 scale in 0.5 increments; standard processing uses a text excerpt of up to 800 characters. It interprets the wording; it does not verify transactions, identity or events. Its output is therefore labeled as text analysis rather than a rating submitted by the user.

This data has a limitation: the analyzer's minimum is 2, while an author can select 1. Irony, ambiguity and a small number of reviews can also distort the picture. Historical scores of unknown origin do not become confirmed user votes merely because a number was retained.

Author stars and TrustScore measure different things

The author-rating average includes only entries identified as user-submitted scores. It excludes the base score, AI analysis and scores of unknown origin. It may therefore differ from the analytical TrustScore. The number of published review texts can also exceed the number of author ratings.

Telegram polls, subscriber counts, affiliate compensation and registration bonuses are not inputs to this formula. Licenses, fees and trading volume are not separate direct terms in it either; TrustScore alone cannot establish these properties. Some factors may be reflected in the provider's base score under its own methodology.

Using the result

Read the composition and number of signals alongside the score, review dates and the venue's primary documents. A high score does not guarantee solvency, permission to operate in your country or a successful withdrawal. Comparing final numbers without their underlying evidence is insufficient.

If a review source or metric appears incorrect, send the page URL through feedback. See the review policy for publication and challenge guidelines.