The pattern the rule reads
A protection rule cannot read a person, so it reads the account. Five families of signal carry almost all of the weight in a sampled policy, and each one is a comparison the account already supports rather than a judgement about the player. What separates a good rule set from a bad one is not the list of signals; it is whether the thresholds are relative to the player own history or fixed for everybody.
- signal families
- 5
- flags raised
- 6,000
- long sessions
- 1,480
- deposits after a loss
- 2,640
- self-reported flags
- 240
- flags per flagged account
- 1.5
Read the signals
Five comparisons the account already supports, run continuously rather than after a complaint. 6,000 flags in the sample, 96.0% of them relative to the account own history.
Choose a response
Nothing, a message, a limit or a break, or a restriction - one of four, graded by how much it takes from the player. 960 of 4,000 reviews in the sample changed access.
Write it down
Signals, crossing dates, rule version, path, response and outcome - one record per review, including the 1,600 that ended with nothing done.
Take the appeal
A route outside the desk that decided. 480 appeals in the sample, 176 of them upheld, which is 36.7% - high enough that the route does real work.
A protection rule reads the account, not the person. The five families it counts are session length, a session spend above the account own historical mean, the interval between a loss and the next deposit, a request to raise a limit soon after a loss, and a self-reported problem flag. Four of the five are relative to the account own history, which is why they produce most of the flags and also most of the false positives.
Why the signals are comparisons rather than verdicts
A fixed number cannot describe harm. Two hundred pounds spent in a month is ordinary for one account and ruinous for another, and the same account can spend both amounts in different months for reasons that have nothing to do with gambling. So the rule set leans on comparisons that the account itself supplies: this session against this account own sessions, this spend against this account own 90-day mean, this interval against this account own intervals.
That design has a cost that is worth stating plainly. A relative rule has no way to know that the change was caused by something good - a holiday, a bonus, a pay rise - or something bad. It knows only that a line moved. In the sampled desk, 6,000 flags were raised across 40,000 accounts, and 1,600 reviews closed with no action because the movement turned out to be explained. The rule set did its job; the threshold was simply reading behaviour, and behaviour is ambiguous.
The five families, and what each one costs
| Signal family | What is compared | Nature | Flags |
|---|---|---|---|
| A session longer than six hours | the session against the account own sessions | relative | 1,480 |
| A session staked at three times the 90-day mean | the session spend against the account own mean | relative | 1,120 |
| A deposit within ten minutes of a loss | the interval between a loss and the next deposit | relative | 2,640 |
| A limit-lift request soon after a loss | the request against a recent loss | relative | 520 |
| A self-reported problem flag | nothing; it is a statement by the player | absolute | 240 |
The 2,640 deposit-after-a-loss flags are 44.0% of all flags and, on their own, they are the weakest evidence in the table - a player who deposits ten minutes after losing is doing something ordinary. Their value is as a multiplier: they rarely open a review alone, but they are what turns a long session into the second flag that meets the two-flag rule. That is why the two-flag design exists at all.
Worked example / sample B
- flags raised across the sample: 6,000
- accounts that raised at least one: 4,000
- 6,000 / 4,000 = 1.5 flags per flagged account
- deposit-after-a-loss share: 2,640 / 6,000 = 44.0%
- self-reported share: 240 / 6,000 = 4.0%
- relative signals: 5,760 / 6,000 = 96.0%
What the rule cannot see
Every one of the five families is a number the account already produces, and that shapes what can be missed. The rule cannot read the reason for a change, the source of the money, or whether the player is enjoying themselves. It also cannot see play that happens on another operator, which is exactly the gap that appears in sample H: 72 of the 360 missed cases were patterns split across two accounts, neither of which reached the rule on its own.
- The rule reads session length; it cannot read whether the session was one player or a shared device.
- The rule reads the interval between a loss and a deposit; it cannot read that the deposit was a scheduled standing payment.
- The rule reads a spend against a declared band; it cannot read that the band was declared a year earlier and never revised.
- The rule reads one account; it cannot read a pattern that is spread across two.
- The rule reads a bar being cleared; it cannot read a player who never clears a bar and needs help anyway.
That last line is the honest limit of this whole approach, and no desk should pretend otherwise: a signal-based duty is good at catching a change that leaves a trace in the account, and silent about a player whose situation is unchanged but already unsafe.