The part of the picture that is never shown

Nothing on a wins display is false. The defect is subtraction - of the losing rounds, of the median case, of the attempts behind a row and of the population the wins were pooled over - and subtraction is a design choice rather than a lie.

Open the partner account Read a feed in six minutes

Feed stub

Losses shown: 0
Median qualifying win: 328x
Median feed row: 1,640x
Inflation of the typical row: 5.0x

the barThe multiple a win must reach to be eligible for the display - on the sample, 250 times the stake

the poolThe players and spins the rows were drawn from - on the sample, 4,750 players placing 1,900,000 spins in the hour

the intervalThe hours between one player's own qualifying wins - on the sample, 10.4 hours at 400 spins an hour

Direct answerA wins display omits four things: the losing rounds, the median win rather than the largest, the number of attempts behind a row, and the population the wins were pooled over. On the sample the median qualifying win is 328 times the stake while the median row displayed is 1,640 times - so even the typical row is inflated by 5.0 times before the losses are considered.

The four subtractions

The losing rounds

On the sample the feed carries no losing round at all, which is definitional: a feed of wins excludes losses. The category error to avoid is reading a feed’s absence of losses as evidence about their frequency.

The median, in favour of the largest

A display is drawn to its top band, so the typical row is not the typical win. Across the sample the median row is 1,640 times the stake against a median qualifying win of 328 times, a factor of 5.0.

The attempts

A row has no denominator of its own. The 457 qualifying wins in the hour sat behind 1,900,000 spins, and a display of twelve rows carries neither number.

The population

Wins are pooled across every player, so a feed’s pace is set by 4,750 people rather than by one. The reader’s own rate is the pooled rate divided by the pool, which is the arithmetic of the denominator page.

True rows, wrong inference

The characteristic failure of a wins display is not that it states something false but that it invites a conclusion it cannot support. The invitation is arithmetic, and it can be refused with arithmetic.

What a reader might infer, and what the sample supportsinference  "wins like these are common"
  supports   12 rows an hour, true
  missing    457 qualifying wins behind them, 2.6% shown

inference  "a win like this could be mine soon"
  supports   every row is a real player's real round
  missing    1 in 4,155 spins, or one per 10.4 hours per player

inference  "the site is paying out"
  supports   payouts happen, constantly
  missing    the return rate, and every player's net result

inference  "the game is different from the way it played for me"
  supports   nothing in the display
  missing    the display is a sample, not a second opinion

Each inference is a step from a numerator to a rate. The display supplies the numerator and the reader has to supply the rate, which is the whole of this desk’s method.

The state of the sample book

Forty invented sites, counted for what their displays publish rather than for what they claim.

What 40 sites' wins displays publishsites examined                                   40
  with a recent-wins feed                        12
  of those, stating the threshold                 0
  of those, stating the pooled volume             0
  of those, showing a losing round anywhere       0
  of those, showing only a multiple, no stake     9
  of those, offering a verify link                3
  of those, verifying a named round               1
  of those 40, showing a spend counter anywhere   2
median row across the feeds                    1,640x
median qualifying win behind them                328x
the ratio                               1,640 / 328 = 5.0x