Raul Moriarty

Background

I am Communications Lead at Poker Bot AI, the umbrella project behind this site. For the past fifteen years I have built software products, worked with online poker operators and their developers, and explained the differences between how the software works and what end-users are told about it. Most of what gets published publicly about "poker bots" and "poker hacks" falls into one of two distortions: marketing copy claiming that solver-anchored decision support is some form of operator exploit, or forum-thread cynicism that treats any kind of automation as deck-prediction snake oil. Both are wrong. They keep readers from seeing what real poker software does, where the engineering problems sit, and how a room like BetOnline handles detection.

BetOnline is a particularly useful case study because it sits in a different segment from the rooms that get most of the public technical attention. It is US-facing in the post-Black-Friday market, lighter-touch on regulation than European licensed operators, runs on an older PartyPoker-derived codebase, and has a documented enforcement history (the 2014 and 2018 bot cleanups) that provides evidence of how detection works at this scale. The notes here are what I want a developer or researcher to read before forming an opinion on this room.

Areas of focus

The threads I keep returning to in poker software:

Modern poker software architecture

Solver-anchored baselines from CFR-derived outputs (PioSolver, GTO+ for heads-up and 6-max, MonkerSolver for multiway), compressed to support real-time querying, paired with opponent models that converge online during a session. In environments like BetOnline, these models are pre-loaded from long-horizon HUD data rather than converged from a cold start.

The BetOnline / Chico Network ecosystem

Stable screen names, tolerated HUDs, a PartyPoker-derived client with years of accumulated patches, a US-facing player pool that has stayed soft since the major operators left in 2011. These features make the engineering problems different from those at GGPoker or PokerStars.

Detection from the operator side

The four-layer model (behavioural fingerprinting, statistical play-pattern, anti-collusion graph, human review) on a smaller operator budget. Where naive implementations get caught, where the enforcement cadence becomes bursty, and how the 2014 and 2018 cleanups illustrate the empirical pattern. Framed as an adversarial-classification problem with an asymmetric cost matrix, not a checklist of "human-looking" tricks.

Business and product

After fifteen years in software engineering and business development, I can distinguish poker-AI claims with engineering substance from sales copy. Most of what gets sold as a "BetOnline hack" is sales copy. Saying so directly has been more useful to readers than another neutral review.

Game theory in practice

Where the math says "stop." Heads-up no-limit hold'em is effectively solved well enough that further automation is rounding error at most stakes; deep-stacked multiway turn play and ICM-heavy MTT endgames are still meaningfully open.

About this site

Three long-form notes (hacks, detection, FAQ) plus the homepage cover what I think is worth saying publicly about BetOnline right now. Pages are revised when the field changes; the date at the top of each piece is the last revision, not the original publication. Companion notes on adjacent rooms (GGPoker, the Chico Network siblings, ACR, Ignition) live on sibling sites that share authorship.

There is a chat link at the bottom of each page. The Poker Bot AI team reads all implementation questions, dataset offers, corrections, and disagreements.