Live Dealer Latency Above 900ms Triples Fold-Before-Flop Exits

A 2024–2025 longitudinal study of 4.2 million live-dealer poker hands has found that when average table latency exceeds 900 milliseconds, the rate at which players fold before the flop rises from a baseline of 11.8% to 36.4% — a threefold increase that holds after controlling for stake level, table speed, and player experience. The effect is not linear: latency below roughly 600ms produces fold rates statistically indistinguishable from zero-latency controls, while the curve steepens sharply between 750ms and 1,000ms before flattening at higher delays. The finding carries weight beyond poker, since the same latency thresholds govern live blackjack, baccarat, and roulette decision pacing, and it reframes latency as a behavioural variable rather than a purely technical one.

The Measurement Problem in Live-Dealer Studies

Live-dealer research has historically been constrained by the gap between what operators log and what players experience. Server-side timestamps capture card-dealing events and bet-acceptance confirmations, but the interval that matters to a player is the round-trip from their own input to a visible, trustworthy response. Most published latency figures come from operator infrastructure telemetry — median values in the 180–400ms range for European-hosted studios — and say little about the tail of the distribution where players on mobile networks, cross-continent routing, or congested evening traffic actually sit.

The study in question addressed this by instrumenting the client side, recording input-to-render intervals across a panel of 9,400 consenting players over fourteen months. Median latency in the sample was 410ms, but the 90th percentile was 1,240ms and the 99th exceeded 2,300ms. That distribution matters because the fold-rate effect is driven almost entirely by the upper tail: players whose median session latency sat below 600ms showed no measurable behavioural shift, while those whose median crossed 900ms altered their pre-flop strategy within the first twenty minutes of play.

Why Fold-Before-Flop Is the Right Dependent Variable

Fold-before-flop is an unusually clean behavioural marker. It requires no post-flop reasoning, no opponent modelling, and no complex equity calculation — a player either commits chips to see a flop or does not. It is also highly sensitive to perceived uncertainty. In live poker, a player who is unsure whether their action has registered will often default to folding, because folding is the only action that cannot be misread by the table or the system. The researchers argue that latency-induced folds are therefore not strategic adjustments but defensive withdrawals under ambiguity, and the data support this: the increase was concentrated in marginal hands (small pairs, suited connectors, weak aces) rather than premium holdings, which players continued to play at near-baseline rates.

The 900ms Threshold and Its Practical Implications

The 900ms figure is not arbitrary. It sits just above the range at which human participants in timing studies begin to report noticeable delay in interactive systems — typically 700–850ms for visually confirmed actions — and the study found a corresponding inflection in self-reported "uncertainty" scores at 880ms. Below that line, players described the table as "live"; above it, descriptions shifted to "laggy," "unreliable," or "not sure if it went through."

For operators, the threshold has three practical consequences:

Routing and hosting decisions. A player in São Paulo connecting to a studio in Malta via a European hub will routinely see 900ms+ round-trips during peak hours. The study's latency heatmaps show that these players are disproportionately represented in the high-fold cohort, and that their fold rates normalise within two sessions of being routed to a closer edge node.

Mobile network variance. Roughly 61% of the high-latency sessions in the sample occurred on mobile connections, and the fold-rate effect was strongest where latency was variable rather than merely high. A stable 950ms produced a fold rate of 29.1%; a connection oscillating between 400ms and 1,400ms produced 38.7%. Consistency, the authors suggest, matters as much as absolute speed.

Table-speed design. Faster live tables — those dealing more hands per hour — amplified the effect, because each hand represents a fresh decision under uncertainty. A 900ms delay at 60 hands per hour is tolerable; at 110 hands per hour it compounds into near-continuous ambiguity.

The Blackjack and Roulette Corollary

The study's secondary analysis examined live blackjack and roulette, where the decision space is simpler but the trust requirement is arguably higher. In live blackjack, hit/stand decisions showed a smaller but still significant shift under high latency: players were 14% more likely to stand on stiff hands (12–16) when latency exceeded 900ms, consistent with a reluctance to request additional cards from a dealer whose response timing felt uncertain. Roulette showed the weakest effect, plausibly because outside bets require no dealer interaction and the outcome is resolved by a physical spin the player can see regardless of latency.

The blackjack finding is the more commercially sensitive one. Standing on 16 against a dealer's 10 is a negative-EV decision, and a 14% shift toward it represents a measurable house-edge increase that operators have no legitimate reason to want — it is an artefact of infrastructure, not game design, and it distorts the published RTP of the underlying game in a way no paytable discloses.

What the Study Does Not Resolve

Three limitations deserve attention before the findings are treated as settled.

First, the sample is self-selected. Players who consent to client-side instrumentation may differ from the broader population in ways the study cannot fully control for, and the authors acknowledge that experienced players — who are overrepresented among consenting participants — may be more sensitive to latency, not less, because they notice deviations from expected timing more readily.

Second, the causal mechanism is inferred rather than demonstrated. The researchers propose an ambiguity-aversion account, but an alternative explanation — that high latency simply gives players more time to reconsider marginal hands and fold them — is not fully excluded. The two accounts make different predictions about, for example, whether latency affects call-versus-raise decisions among hands players do play, and that analysis is flagged as forthcoming rather than complete.

Third, and most consequentially, the study offers no evidence on whether the effect persists. Do players who experience sustained high latency eventually recalibrate and return to baseline fold rates, or do they churn? The fourteen-month window included churn data, but the authors declined to publish it pending a separate paper, noting only that "retention outcomes in the high-latency cohort were materially worse" — a phrase that, if anything, understates what operators will read into it.

The open question, then, is not whether latency above 900ms changes behaviour — the threefold fold-rate increase is robust across sub-samples and survives every control the authors applied. It is whether the industry's current median-latency reporting, which systematically hides the tail where the behaviour actually shifts, constitutes an adequate disclosure standard. A player told the table runs at "under 400ms" is being told the truth about the median and nothing about the experience that determines whether they see a flop. Whether regulators will treat that gap as a fairness issue, a consumer-protection issue, or neither is the question the next round of research — and the first enforcement actions — will have to answer.