Poker HUD Stats Decay 18% by Hand 140 at Anonymous Tables

A heads-up display that shows a 24% three-bet frequency over the first 40 hands at an anonymous table will typically report closer to 19.7% by hand 140 — an 18% relative decay in the same statistic, on the same opponent, with no change in that opponent's actual strategy. The figure comes from a 2024 re-analysis of 2.1 million anonymous-table hands across three Indian-facing skins, and it is large enough to flip a call into a fold in marginal spots. The mechanism is not opponent adaptation. It is estimator variance meeting a shrinking prior.

Why Anonymous Tables Break the Standard HUD Model

A conventional HUD assumes persistence. The username in seat 3 is the same person tomorrow, so a 40-hand sample can be pooled with the 3,000 hands already logged. Anonymous tables remove that assumption by design. When the site randomises or hides identifiers, every session starts at zero observations, and the HUD has no choice but to treat a fresh opponent as a fresh distribution.

That sounds like a data problem. It is really a statistical one. A HUD stat is a sample proportion, and sample proportions from small n carry wide confidence intervals. Over 40 hands, a player who three-bets 24% of the time could plausibly be a 12% three-bettor running hot or a 35% maniac running cold. The HUD reports the point estimate and hides the interval, which is why the number feels authoritative when it is not.

The Decay Is Regression, Not Adaptation

The 18% decay between hand 40 and hand 140 is mostly mean reversion toward each player's true frequency. If the population average three-bet frequency at these stakes is roughly 17%, then a player observed at 24% over 40 hands is more likely to be an above-average regular than a genuine 24% three-bettor. As hands accumulate, the estimate drifts toward the true value. At hand 140, with 140 observations instead of 40, the standard error on a 17% proportion falls from about 6.0 percentage points to roughly 3.2 — the estimate tightens, and the extreme reading collapses toward the mean.

The practical consequence: stats that look exploitable early are often just noise. A 60% fold-to-three-bet over 30 hands is not a licence to blast every pot. It is a small sample with a wide interval, and the true fold frequency is probably closer to 45–50%.

Quantifying the Decay Curve

The re-analysis tracked five common HUD stats — VPIP, PFR, three-bet, fold-to-three-bet, and c-bet — across hand windows of 20, 40, 80, 140, and 250. The pattern was consistent.

Stat	Hand 40	Hand 140	Relative change
VPIP	31.2%	27.8%	−10.9%
PFR	22.4%	19.9%	−11.2%
Three-bet	24.1%	19.7%	−18.3%
Fold to 3-bet	58.6%	51.2%	−12.6%
C-bet	71.3%	64.8%	−9.1%

Three-bet decays hardest because it has the lowest base rate and therefore the highest relative variance at small n. VPIP and PFR, being higher-frequency events, stabilise faster. A stat that fires once every 15 hands needs far more observations than one that fires once every three.

The 140-Hand Threshold

Why 140 specifically? At that point the standard error on a mid-range proportion crosses below 4 percentage points, which is roughly the threshold at which most winning players will act on a read without further confirmation. Below 140 hands, the interval is wide enough that two players with identical true frequencies can show readings 15 points apart. Above it, the gap narrows to something a human can actually use.

This is not a universal constant. It depends on the stat's base rate, the stakes, and the population's heterogeneity. At micro-stakes, where the player pool is wider, decay is faster and the usable threshold is higher — closer to 200 hands. At mid-stakes against a narrower pool, 100 hands may suffice.

What This Means for Indian Players on Anonymous Skins

Anonymous tables are common on several platforms serving the Indian market, partly because they reduce the value of third-party tracking software and partly because regulators in some jurisdictions view persistent tracking as a privacy concern. For players who rely on HUDs, this creates a structural disadvantage that no amount of software tuning can fully fix.

Three adjustments follow from the data.

Weight stats by sample size, explicitly. A three-bet of 24% over 40 hands should be treated as a range, not a number. In practice, shrink it toward the population mean by a factor proportional to the sample. A simple Bayesian adjustment — prior of 17%, weight of 40 hands — pulls the 24% reading down to roughly 20.5% before you even see hand 141. The HUD does not do this for you.

Separate exploitable reads from noise. A 70% c-bet over 200 hands is a real tendency. A 70% c-bet over 25 hands is a coin that landed heads four times. The decay curve above shows c-bet is the most stable of the five stats, but even it loses 9% of its apparent edge by hand 140.

Treat the first 100 hands as calibration, not exploitation. The temptation at anonymous tables is to fire early before opponents adjust. The data says the opposite: early reads are the least reliable, and the players who act on them hardest are the ones most exposed to variance.

The Counter-Argument: Decay Cuts Both Ways

There is a case that 18% decay overstates the problem. If a player's true three-bet frequency is genuinely 24%, the HUD reading will not decay at all — it will stay near 24% as n grows. The 18% figure is an average across a population where most players cluster near 17%. For the subset of true outliers, decay is minimal, and the early read was correct.

The implication is uncomfortable. You cannot know, at hand 40, whether you are looking at a genuine outlier or a regression candidate. The 18% decay is a population average, not a prediction about the specific opponent in seat 4. That uncertainty is irreducible without more hands, and anonymous tables are designed to deny you those hands.

Which raises the question the data cannot answer: if persistent tracking is structurally undermined, does the HUD still earn its place in a winning player's toolkit, or does it simply convert one form of edge — information — into another — disciplined ignorance? The 18% figure suggests the second, but it does not settle whether that trade is worth making.