The science behind PokerGuard

Poker players obsess over ranges, frequencies, and solver outputs — but ignore the single variable that degrades all of them simultaneously. Cognitive fatigue doesn't make you forget GTO. It makes you execute it worse: slower processing, more automatic decisions, reduced sensitivity to thin value spots, and impaired risk assessment.

The research is unambiguous. Lim and Dinges conducted a meta-analysis of 70 studies and found that sustained attention — the cognitive function most critical for monitoring multiple tables and tracking action — shows the largest impairment under fatigue, with a large effect size of g = −0.78 [12]. Working memory, which you use to track ranges, pot odds, and opponent tendencies across multiple tables, shows a significant but smaller effect at g = −0.37.

Most critically, research on gambling-specific decision-making shows that fatigued individuals simultaneously become more optimistic about gains and less sensitive to losses [18]. Venkatraman and colleagues used fMRI to demonstrate that sleep-deprived subjects show increased reward-area activation and decreased loss-processing activation — the neural signature of a player about to punt off stacks. In a study of 23 online poker players tracked across 588 sessions, Hamel and colleagues found that sleep-deprived sessions produced significantly worse financial results and elevated tilt symptoms [7]. And perhaps most troubling: Van Dongen and colleagues showed that people experiencing cumulative cognitive fatigue are largely unaware of their own impairment, with subjective sleepiness plateauing while objective performance continues to decline [17].

You can't feel yourself getting worse. That's the problem PokerGuard solves.

PokerGuard uses within-subject modelling. When you first use the app, you complete a short calibration — a reaction time test and a working memory test — while rested. This establishes your personal baseline: how fast and accurate you are at your best. All future measurements are expressed as percentage degradation from YOUR baseline, not compared against a population average. A naturally fast player and a naturally slower player are both "green" at their personal best. This eliminates the inter-individual variability that makes one-size-fits-all thresholds useless.

The system then combines multiple fatigue signals. The two core signals are reaction time — measured via a task based on the Psychomotor Vigilance Test, the gold standard in sleep and fatigue research since Dinges and Powell established it in 1985 [6] — and working memory performance, measured through a symbol-matching task scored using the Linear Integrated Speed-Accuracy Score, which Liesefeld and Janczyk showed is more robust than traditional composite metrics when speed-accuracy trade-offs are present [11]. These cognitive signals are weighted most heavily because they directly measure the brain functions that poker demands: vigilance, executive control, and decision speed.

On top of the cognitive core, PokerGuard tracks contextual factors: time on task (because fatigue accumulates whether you test or not), multi-table load (because task-switching imposes measurable cognitive costs [16]), and session financial stress. Each signal is mapped to a normalised 0–1 scale using a dead-zone-to-saturation architecture: small fluctuations within normal daily variability are ignored, meaningful degradation scales proportionally, and severe impairment saturates at the empirically validated threshold for that measure.

After enough labelled sessions, PokerGuard trains a personalised machine learning model that learns YOUR specific fatigue patterns. The model uses logistic regression — chosen deliberately over more complex approaches because it's interpretable (you can see which factors predict your fatigue most strongly), runs instantly on your machine, and is well-suited to the sample sizes a single player generates. The feature set has been reduced to the eight most predictive variables based on statistical best practices for small-sample prediction models [15], ensuring the model learns genuine patterns rather than memorising noise.

The model's predictions are combined with the rule-based fatigue score, creating a hybrid system where the rules provide a scientifically grounded floor and the ML layer personalises on top. Over time, PokerGuard learns how quickly your performance degrades, how effective breaks are for you specifically, and which session lengths are sustainable for your physiology and play style. All processing happens locally on your machine — your data never leaves your device.

While PokerGuard's measurement methods draw from the broader cognitive fatigue literature, the case for fatigue monitoring in poker specifically has direct empirical support. Hamel, Bastien, Jacques, Moreau, and Giroux tracked 23 regular online poker players across 588 sessions over 28 days and found that sleep-deprived sessions produced significantly elevated tilt symptoms and unfavourable financial outcomes [7]. Notably, cognitive tilt was not significantly elevated in experienced players — they maintained their strategic frameworks — but emotional regulation failed, leading to worse execution of those strategies.

The tilt literature reinforces this. Moreau and colleagues found that tilt frequency alone explains nearly half the variance in problem gambling outcomes among online poker players [13]. Palomäki, Laakasuo, and Salmela showed that loss sensitivity is the strongest predictor of tilt severity [14], while Laakasuo and colleagues demonstrated that emotional priming directly reduces mathematical accuracy of poker decisions [10]. Fatigue lowers the threshold at which tilt occurs — you don't need a bigger bad beat to tilt when you're tired, just the same one.

One finding particularly relevant to grinders: Killgore, Grugle, and Balkin found that stimulants like caffeine restored reaction time performance and subjective alertness but completely failed to restore decision-making quality on the Iowa Gambling Task [9]. Decision-making only returned to normal after recovery sleep. Feeling alert is not the same as being sharp. That coffee isn't saving your decision-making — it's just making you confident while impaired.