Scoring methodology
Every brand score on this site is generated from one weights file. Nothing on this page is typed in by hand.
What the score measures
The score measures the blackjack a brand actually deals. The score does not measure the casino around the blackjack.
Two blocks make up the score. Block A scores the rules of the table. Block B scores the conditions of access.
Block A scores one table per brand. The scored table is the best documented table on the site that is not a multiplier format. Rules are never mixed across two tables, because a mixed rule vector describes a table nobody can sit at.
The weights
The Block A weight of a rule is derived from the published cost of that rule. Cost is measured in percentage points of player return.
Block B carries no published cost, so each Block B weight is assigned. Block B never outweighs Block A. The rules of a table are a direct cost to the player. The conditions of access are friction.
| Criterion | Block | Weight | Published cost |
|---|---|---|---|
| Payout on a natural blackjack | Table rules | 16 | 1.39 pp |
| Deck count and shoe type | Table rules | 10 | 0.48 pp |
| Split limit, resplit and split aces | Table rules | 6 | 0.19 pp |
| Dealer action on soft 17 | Table rules | 6 | 0.22 pp |
| Totals a player can double on | Table rules | 6 | 0.18 pp |
| Double after split | Table rules | 5 | 0.14 pp |
| Dealer peek against no hole card | Table rules | 5 | 0.11 pp |
| Surrender | Table rules | 4 | 0.07 pp |
| Published withdrawal processing time | Access conditions | 10 | n.a. |
| Blackjack tables with published rules | Access conditions | 10 | n.a. |
| Span of published table limits | Access conditions | 8 | n.a. |
| Published verification requirements | Access conditions | 8 | n.a. |
| How a live bonus treats blackjack | Access conditions | 6 | n.a. |
| Total | 100 |
Weights version rev3, effective 2026-08-29.
Why the weights are these numbers
The payout on a natural blackjack is the largest single criterion. The reduced payout costs the player 1.39 percentage points. No other rule on the table costs a quarter as much.
Each Block A weight is the square root of the published cost of that rule. The Block A weights are then scaled to the block total. A straight cost allocation would give the payout almost the whole block. The remaining rules would then score as noise, and a player choosing between two good tables would learn nothing.
The build recomputes every Block A weight from the published costs on each run. A weight edited by hand fails the build.
How a missing field scores
A criterion with no published value scores zero points and keeps its full weight. The denominator is never reduced. A brand is never rewarded for publishing less.
No missing value is ever estimated, inferred to a plausible figure or filled from a comparable brand. A cell with no published value reads n.a. and states what is missing.
Two states are reported separately. A field marked n.a. means the operator does not publish the value. The absence is a finding about the operator.
A field marked pending means the research is not finished. A pending field is a gap in the work of this site and is not a finding about the operator. A score with any pending field is published as provisional.
Confidence classes
Every input carries a confidence class and a source URL. The class multiplies the points that criterion earns.
A claim this site has not verified does not score as a verified claim. An unverified claim is also not treated as false.
| Class | Multiplier | What the class means |
|---|---|---|
| confirmed | 1.00 | The operator publishes the value and this site read it at the source. |
| confirmed_existence_only | 0.50 | The operator publishes that the item exists and publishes no value for it. |
| reported | 0.50 | A source states the value and the operator does not confirm it. |
| inferred | 0.25 | The value follows from another published rule and is not stated directly. |
| assumed | 0.00 | No evidence supports the value, so it earns nothing. |
| n.a. | 0.00 | The value is not published in any source this site checked. |
| pending | 0.00 | This site has not finished researching the field. |
House edge conventions
Every house edge on this site is stored in three conventions. The first is optimal play. The second is basic strategy with a cut card. The third is basic strategy with a continuous shuffle.
The continuous shuffle column is the engine default. The simulator reshuffles after every hand, so the continuous shuffle figure is the figure that describes what the simulator deals.
A cut card figure is never shown beside simulator output. The two describe different games and the gap reaches 0.15 percentage points at one deck. Cut card figures appear only inside the land based comparison module.
Every published figure carries the name of its convention. A figure with no convention named is a build failure.
When the solver and a published chart disagree
This site runs its own basic strategy solver. A solver carries an approximation error, and that error is recorded per rule set.
A published chart wins any cell where the disagreement is smaller than the approximation error of the solver. In that case the solver is not precise enough to overturn the published chart.
Where the disagreement is larger than the error bar, this site publishes its own number. The page says so and marks the cell.
This rule is written down here rather than left to a judgement call at the time.
Retest cadence and who tests
The rules and data desk performs every test. The desk holds the data spine and is accountable for it.
Every brand is retested every 90 days. The retest date is stored beside the inputs and appears on the brand page.
Four events trigger a retest before the 90 days elapse. A brand changes a table rule. A brand changes its published bonus terms. A brand changes licence or registered entity. A reader reports a rule that does not match the site.
A retest can lower a score. A score is not held at its previous value.
How the score is published
Every brand score on this site is published out of 10. A published score is rounded to 1 decimal place. The criterion weights sum to 100. The model total is divided by 10.
The order brands appear in is fixed and is not the score order. A higher score does not move a brand up a list.
What the score does not measure
The list below is published so the exclusions can be checked. Each item is excluded for a stated reason.
| Excluded | Reason |
|---|---|
| Headline value of a welcome bonus | A headline bonus value is not a fact about the game. |
| Total number of games on the site | A total game count counts slots. A slot count says nothing about blackjack. |
| Software provider partnerships | A provider partnership is not a fact about the game. |
| VIP and loyalty schemes | A loyalty scheme is not a fact about the game. |
| Site design and usability | Site design cannot be measured against a published source. |
| Customer support channels | A support channel is not a fact about the game. |
| Multiplier format blackjack tables | This site does not rank multiplier format tables. The pay tables are not published in full. |
| Player advantage rule configurations | Player advantage rule configurations exist in the trainer and the calculator only. No commercial surface carries one. |
| Commercial terms between the brand and this site | Scoring a commercial term would make the score a function of the revenue of this site. |
Affiliate relationship
This site earns a commission when a reader registers with a brand through a tracked link. The reader pays nothing extra.
Commission is not an input to any criterion in the model. A brand cannot pay for a score, a rank or a softer review.
This site tests 5 brands. This site holds a commercial relationship with each of those 5 brands. The limit on coverage is stated here rather than implied.
A brand that scores badly stays on the site with its score. A score is never removed because it is unflattering.
Read the full affiliate disclosure
The files this page is built from
The weights, the inputs and the generated scores are all files in the site repository. A weight change re-scores every surface on the next build.
Where these rules come from
Every rule set on this site is recorded against a published source.
Review and provenance
Where a figure is not published, this page says so. No figure on this page is estimated.