We graded 16 studios on what they publish before reading most of their pages. Two of the first five we actually read contradict the grade, in opposite directions.
Published 2026-08-18Not changed since publicationNo review date set yetSources not re-read since publication
Our source policy grades a studio on what it publishes at title level. A means the configuration set is stated. B means a single return figure plus a full specification. C means one or the other, patchy. D means a marketing page with no usable numbers.
The file then states in bold that no provider anywhere scored A. That sentence was written from a sample of seven studios, and it is the kind of claim that stands until somebody opens a page nobody in the sample had opened.
16 provider rows now carry a grade. Five have had a game page read end to end, which is the only reading that can confirm or break one, and two of those five broke it in opposite directions on the same afternoon. That is a poor strike rate for a scale.
Hacksaw Gaming was graded B. Its own page for Max Win Machine publishes three returns side by side in a labelled data block: 92.20% to 96.22%◦. Its page for Wanted Dead or a Wild publishes four. The same page states the ceiling as a multiple of the stake rather than a cash figure: 10,000×◦.
That is grade A on the file's own definition, twice over, from the studio the file did not predict it from. The provider row was created at A with the reading attached, and it is the first row in this database at that grade.
None of the three builds is marked as the default, because the studio names none. Naming a default the studio never named would put our guess in the one field a reader treats as settled, so the record carries a warning about the missing default instead. That is the correct state, not a gap in the work.
Habanero is the older case, and it is the one that should have moved the scale first. Its own game page publishes five return configurations for one title: 86.00% to 98.00%◦. 12 percentage points separate the lowest build from the highest.
That is grade A behaviour on a row still graded C. Nobody noticed. The research file names Habanero as the counter-example to its own universal claim, in the same section that assigns the grade, and nobody reconciled the two.
So the pattern is not that two readings were unlucky. It is that a grade assigned from a sample survives contact with the first page read from it about half the time.
Fa Chai looks like a B on paper. Its page for Muscle Fortune Cat publishes a return figure, a volatility label, a maximum multiplier, a ways count, a release date and a named feature list. That is a full specification and it is not patchy.
Then we read a dozen more of its titles, from a game with 4,096 ways to one with 27, and every slot in the catalogue carries the identical return figure. Its fishing games carry a different one and are a different format entirely.
A number that never varies across a catalogue is either a house target the studio applies everywhere or a template value nobody has filled in since the page was built, and nothing anywhere on the site distinguishes the two for a reader trying to use it. So our record for that title has an empty configuration set on a page whose studio published a number, and the figure sits on the provider row at provider grain instead.
The deeper problem is not that two grades were wrong. It is that studios do not publish more and fewer of the same fields. They publish different ones.
JILI's page for Super Ace is a server-rendered specification table: a ceiling, a volatility word, a line count, a launch month. It carries no return figure anywhere. PG Soft inverts that exactly.
Its page gives us the 1 configuration we hold for Fortune Tiger, next to a session volatility printed as a bare 2. Neither the bounds nor the direction of that scale appears anywhere. We keep the return and drop the 2, because turning an unscaled integer into a word would be our guess in the studio's voice.
One studio discloses the volatility and not the return. The other does the reverse. A single letter puts them in the same band and describes neither, which is a problem with the instrument rather than with either studio.
The argument against all of this is practical, and it is a good one. A reader deciding between two games does not want a paragraph on disclosure methodology, and a facet cannot filter on a paragraph. A sortable single value is what makes a studio comparison exist at all.
So we keep it. We keep the grade for exactly that reason. What changed is what the grade is allowed to be: a research estimate in a document, and a reading on a row, are two different objects, and they are permitted to disagree in public.
So the row wins where a page has been read, the document keeps its estimate where none has, and the disagreement is written down in both places rather than settled by deferring to whichever was written first. A grade that quietly updated itself would leave nobody able to see that the estimate had been wrong.
The practical version is short, and it is three sentences long. A studio that publishes a set tells you where your build sits inside it. A studio that publishes 1 figure tells you a set exists and is not for you. A studio that publishes the same figure on everything has told you about its house target and nothing about the game in front of you.
What none of them tells you is which build your casino runs. That is the join our bridge table was built for, and it holds 0 rows today. Reading five studio pages, at roughly one page per title and three external reads per run, moved two of the sixteen grades this database holds. Reading one information screen inside a real account would move something bigger, and that is the next thing worth doing rather than a sixth studio page.
It measures what a studio publishes about its own games at title level, not whether the games are fair or well tested. A grade of A means the configuration set is stated on the studio's own page. A grade of D means the page carries marketing copy and no usable numbers.
Both were assigned from a research estimate before any of those studios' pages had been read from this machine. One studio publishes far more than the estimate predicted, and one publishes a figure that does not function as a title-level disclosure at all. The readings replaced the estimates on the rows, and both disagreements are recorded.
Because a number that never varies is a house target or a template value, and nothing on the page distinguishes them. Storing it against a single title would put a figure in the one column a reader takes as a statement about that game's maths. We store it on the studio instead, at studio grain.
No. It means the studio told you more about what you are playing. A studio can publish its whole configuration set and still ship a low build to your casino, and a studio that publishes nothing can be running a generous one.
Five of the 16 provider rows have had a game page read end to end. The rest carry a research estimate, and that estimate is labelled as one. Two of the first five broke the scale, which suggests the remaining eleven will move too.
Not a bibliography — these are the addresses behind the numbers this site stores about the 5 records this piece is about, with the date we read each one. Anything the article argues from that we have not stored as a figure is not in this list, and we would rather say that than pad it.
Nothing on this page has been corrected since it was published on 2026-08-18. When something is, it is listed here with the date and what changed — not quietly edited. How we handle a correction.
A wagering requirement times a house edge gives a real cost in pesos or dollars. The build that sets that edge is stored on this site for dozens of slots, and most bonus terms never name it.
Two of PG Soft's own game-information sheets name a selectable range of RTP builds. Four others state flatly that no such range exists for that title. TopSlotsAsia checked all thirty-two titles it has published from this studio.
Two of the seven brands this site tracks were checked this week for a no-deposit bonus's wagering terms. Neither page would open cleanly.
MelBet's own terms cap wagering-period stakes at €5, and a separate clause names bets over that figure as evidence of bonus abuse.
