Working notes on ClubGG bots and cheats in the NSUS club ecosystem.

What ClubGG actually is

ClubGG is a mobile-first private-club poker app launched in 2020 by NSUS Group, the same parent company behind GGPoker. The two products share corporate ownership and some engineering DNA, but they are different in almost every operational dimension. GGPoker is a regulated public operator selling cash games and tournaments under multiple gambling licences. ClubGG is a club-based app where the in-app currency, Diamonds, is not redeemable for cash inside the platform; clubs and agents handle deposits, cashouts, and player settlement in the offline world.

From a regulatory perspective this is the same model that PPPoker, PokerBros, X-Poker, and Suprema Poker run. The platform takes the position that it provides software for private games among consenting groups. The settlement layer that turns chips into money happens outside the application, between a player and a club agent. That model is what makes ClubGG popular in regions where public real-money poker is restricted or where players prefer the closed-group economics of a club.

The app supports six-handed and nine-handed NLH and PLO cash, a Diamond Race tournament series similar in shape to Spin and Gold, a four-colour deck option, customisable rake structures per club, run-it-twice, straddle, and the operational features club owners need — agent hierarchies, balance sheets, transfer logs, and seat-restriction controls. From the player's screen it looks like any other mobile poker client. Under the hood the club graph is the load-bearing structure, not the table.

Why a bot market exists here

This is the question that separates ClubGG from genuinely free play-money apps where no settlement exists. On a free app there is nothing to extract — bots are a curiosity, not a business. On ClubGG, even though no money moves through the platform, the offline settlement layer means every Diamond a bot wins is convertible to cash at the agent level. The economics are real, which means engineering effort is justified, which means the bot market actually exists in volume.

What changes here, relative to a public site like GGPoker, is the threat model. On an anonymous-table public site the bot operator is invisible to other players. On ClubGG, the bot operator either runs the club or sits inside a known club. The agent layer is the enforcement vector that does not exist on GGPoker — a club owner who notices a player draining their soft regulars can simply remove the seat. The operator does not need to ban the account; the agent restricts the table.

This produces a two-tier reality. Small and mid-stakes clubs with weak regulars and inattentive owners are heavily bot-exposed; the volume of automated play in this segment is non-trivial. High-stakes clubs with attentive ownership and tight player vetting are essentially closed to externally-run bots, because the agent layer catches them through ordinary club management before NSUS does.

The detection topology — NSUS plus the club layer

Detection on ClubGG is structurally different from a public operator. There are two stacks running on top of each other.

NSUS-network signals

Client telemetry, behavioural fingerprinting, statistical play-pattern analysis, and account-graph models — the same signal families the GGPoker security team has been running for years. ClubGG sits on shared infrastructure; some of this tooling crosses over. The platform does not publicly confirm which signals are shared across products, and the cross-product picture is the most open question in the ecosystem.

Per-club moderation

Club owners and agents see balance sheets, transfer flows, and detailed table history for their own club. They have the operational tools to restrict seats, expel players, freeze balances, and refuse settlement. This is the layer that does most of the real-time bot-spotting at small and mid stakes — a club owner with a good feel for their player pool notices a new account winning above its skill envelope long before the NSUS pipeline produces a verdict.

Cross-NSUS escalation

When a NSUS-side flag fires, the action can range from a quiet account restriction to a full closure. Withdrawals at the club level are the natural choke point; even if Diamonds are confiscated on-platform, the club-side settlement is the real consequence, and ClubGG can choose to communicate with affected clubs or not.

The result is a more distributed enforcement model than GGPoker. The platform-side stack catches large-scale farms and statistical outliers; the agent-side layer catches single bots inside individual clubs by ordinary judgement. The two layers cover different failure modes, and neither is sufficient alone.

Collusion through outside channels

The defining cheating risk on ClubGG is not the bot — it is collusion. The club model concentrates a small group of players who already know each other, and the natural step from there is to set up an outside-channel chat, share hole cards on the river, and coordinate aggression against a target. WhatsApp, WeChat, Telegram, Discord, and in some markets KakaoTalk are the obvious vehicles. The platform cannot see any of this.

The defence against outside-channel collusion is statistical, not behavioural. The same account-graph and showdown-correlation models that catch chip-dumping on a public operator can catch collusion on a club app, but the signals are noisier because club tables naturally have repeat opponents. You expect higher correlation between two players in the same club than between two random strangers on GGPoker. Setting the false-positive budget correctly is the operational challenge — too tight and you flag legitimate club regulars; too loose and you let the colluders through.

The detection note walks through the cross-NSUS signal sharing question, the agent-side enforcement layer, and pseudocode for a basic collusion-pair detector built from showdown logs.

Areas covered on this site

Open questions for developers and operators

Five threads I think are genuinely open and would welcome discussion on:

1. Cross-NSUS signal sharing. ClubGG and GGPoker share corporate ownership and some infrastructure. The honest answer to whether a bot flagged on one will get caught on the other is "probably, but we cannot prove it from the outside." The empirical test is hard because confirmation requires an account history on both products and an observed cross-product enforcement action — neither commonly published. Anyone with documented evidence either way should write in.
2. Agent-side enforcement as adversarial filter. The agent layer behaves like a low-volume, high-context classifier — a single human watching a few dozen players for hours. Bots that pass the NSUS statistical layer often fail the agent layer because agents read social and behavioural signals that the platform does not collect. Modelling this as an additional adversarial-classification stage is, as far as I can tell, an open formal problem.
3. Showdown-correlation detection in repeat-opponent environments. Standard collusion detection assumes near-random opponent matching. Club apps violate this assumption. The right baseline is per-club, not per-network, and constructing population priors at the per-club level with reasonable sample sizes is genuinely fiddly.
4. Anti-bot transfer learning across NSUS products. If GGPoker's behavioural-fingerprinting model is trained on a population of regulated-market players with Western input characteristics, how well does it generalise to ClubGG's Asian and Latam population? The transfer-learning question is not academic — false-positive rates can swing by an order of magnitude when populations differ.
5. The Diamond economy as a measurement instrument. Settlement happens off-platform, but Diamond flows inside the app are fully observable. The question is whether net Diamond movement between accounts within a club, aggregated over months, can serve as a proxy for outside-channel collusion intensity. The signal is noisy because legitimate stake-sharing and backing relationships look similar. I think it is workable but I have not seen anyone publish a model.

If you are working on any of the above, the chat is the right place to share. Implementation discussion welcome; product-pitch messages are auto-archived.