Evidence & worked examplesDisclosure coverage and game probability are different questions
This report asks whether reviewed sources disclose a field. The new probability study asks whether a defined event can be calculated from complete rules. A game can disclose RTP without disclosing hit or bonus frequency. The fixed Dataset v1.0 sample below is unchanged.
Read the 2026 Game Probability Transparency Study (Traditional Chinese) · Browse all research topics
How to interpret and reproduce the disclosure audit
This is a methodological expansion, not a new sampling round. Dataset v1.0 retains its 2026-08-13 cut-off, inclusion decisions and original statistics. The following examples are deliberately hypothetical and are excluded from the dataset.
Six checks before interpreting a disclosure percentage| Check | Record | Interpretation boundary |
|---|
| Unit of analysis | Source record → screened title → included title | 143 records, 95 screened titles and 82 included titles describe different stages, not three independent samples. |
| Identity | Provider, stable game ID, release and configuration | A title spelling match is not enough to merge products; a configuration is not automatically a new title. |
| Disclosure | Field present, absent, ambiguous or version-mismatched | Absence in reviewed sources is not proof that the operator lacks the data. |
| RTP variants | Store each quoted value with its source and configuration | Do not average four RTP values into a fictitious playable setting. |
| Probability coverage | Hit event, bonus trigger, maximum-win event and denominator | An RTP field alone identifies none of these event frequencies. |
| Evidence preservation | Capture date, quoted field, source type and review decision | A changing web page should not silently overwrite the fixed Dataset v1.0 result. |
Why game categories need different evidence
For slots, record configuration and mode before comparing RTP. For table games, record deck composition, drawing rules, payouts and strategy assumptions. For lottery models, record the draw space and exact winning event. For sports, preserve time-stamped market definitions and distinguish price-implied probability from a calibrated prediction. For fishing games, a score multiplier does not disclose capture weights. These are proposed reading criteria; the original slot-focused sample does not become a completed cross-category audit.
Game Probability Transparency Study — rules, models and worked examples (Traditional Chinese)
V65-RESEARCH-EN-01｜One title, two RTPs, zero justified hit-rate estimates
Rules / assumptions:Hypothetical matched title; distinct confirmed configurations
Known facts:Configuration A states 96%; B states 94%. Neither discloses a hit definition or frequency.
- Store two configuration-level values under one title identity.
- Keep hit frequency as not disclosed in the reviewed evidence; do not fill 4%, 6% or 95%.
- If the user’s loaded configuration is unknown, neither value is verified for that session.
Result and reason:Disclosure of two RTP values increases documented configuration detail, not the number of measured products or known hit probabilities.
Change one condition:Change only the evidence: a matching help panel identifies configuration B. Its 94% can now be attributed to that configuration, while hit frequency remains unknown.
Original worked example under the stated assumptions; not an undisclosed product statistic or observed play record.
V65-RESEARCH-EN-02｜A missing field and a failed fetch are different
Rules / assumptions:Illustrative three-source review, not Dataset v1.0 additions
Known facts:A readable page omits RTP; a second returns an access error; a third lists RTP for an unmatched release.
- A supports “not found on the reviewed page”.
- B supports “not assessed due to access failure”, not confirmed absence.
- C supports “version mismatch”; do not transfer its value.
Result and reason:Use distinct evidence states and a predeclared denominator; collapsing them creates a misleading coverage rate.
Change one condition:Change only B to a readable page with matching RTP. Update that evidence state, record the date and issue a new dataset version if the fixed sample is reanalysed.
Original worked example under the stated assumptions; not an undisclosed product statistic or observed play record.
This addition explains the site’s own methodology or explicitly hypothetical examples. It does not certify any commercial implementation.
2026-09-07 Change log
- Added the probability-study cross-reference and clarified the difference between disclosure coverage and event probability.
- Kept the original sample, cut-off date and results unchanged.
- Expanded six-field audit interpretation, cross-category evidence boundaries and two worked examples; fixed original dataset unchanged.
This is an editorial addition, not a new audit of the original fixed dataset.