Poker HUD Stats Misread 12% of River Aggression at Table 4

A 2024 audit of 1.2 million tracked river decisions across four mid-stakes tables on a single India-facing skin found that HUD-derived aggression frequency (AF) diverged from ground-truth aggression in 12.4% of river spots at Table 4, against a 6.1% divergence at the other three tables combined. The gap was not random: it clustered around multi-way pots above three players and around opponents whose river leads came from a small, positionally skewed sample. The headline figure is a measurement artefact as much as a behavioural finding, and the distinction matters for anyone using a HUD to make river calls in Indian rupees rather than in theory.

What the 12.4% actually measures

The audit defined "misread" narrowly. For each river decision, it compared the HUD's displayed aggression frequency for a specific opponent against a recalculated frequency built from the full hand history, with the same filters the HUD claims to apply. A spot counted as misread if the displayed value sat more than one standard deviation from the recalculated value, or if the two values fell on opposite sides of a decision threshold the user had configured (commonly 30% for "station" versus "aggressor" tags).

Table 4's 12.4% breaks down unevenly:

• Heads-up, single-raised pots: 4.8% divergence
• Three-way pots: 11.2%
• Four-way or more: 19.7%
• Pots where the tracked opponent had fewer than 40 river opportunities: 27.3%

That last line is the mechanism. A HUD that reports aggression frequency over, say, 22 river hands is reporting a number with a confidence interval wide enough to swallow most of the practical difference between a nit and a maniac. At Table 4, the player pool was small and self-selecting — the same 30-odd regulars cycling through, many of them multi-tabling — so hand histories accumulated slowly and asymmetrically. The HUD was not lying. It was answering a question nobody had asked: "What is this player's river aggression across a sample that happens to be dominated by the two seats to their left?"

The positional skew problem

River aggression is not a stationary property. A player who fires 45% of rivers when they hold the button and 12% when they are in the small blind will display a blended AF that describes neither situation. In the audit, 61% of the flagged Table 4 spots involved opponents whose river leads came disproportionately from late position — specifically, more than 70% of their tracked river aggression originated from the last three seats to act. The HUD aggregated these into a single figure, and users read that figure as if it were context-free.

This is a known limitation, documented in the poker tracking software ecosystem for over a decade. It is also routinely ignored, because the alternative — filtering every stat by position, stack depth, and pot type before acting — is slow enough that it defeats the purpose of running a HUD at four tables.

Why Table 4 and not Tables 1 through 3

The obvious question is what made Table 4 different. Three factors, in order of measured effect:

1. Pool size. Table 4 averaged 34 unique opponents over the sample window, against 71, 68, and 59 at the other tables. Fewer opponents means fewer hands per opponent means noisier per-opponent stats.
2. Stack depth variance. Table 4 had the widest spread of effective stacks (40bb to 210bb), and HUDs rarely filter aggression by stack depth. A short stack's river shove and a deep stack's river bet are not the same action, but they land in the same counter.
3. Rake structure. Table 4 ran a 4.5% capped rake against 5% uncapped elsewhere, which shifted the pool toward more multi-way limped pots — precisely the configuration where HUD aggression stats degrade fastest.

The 12.4% is therefore not a universal constant. It is a property of a specific table ecology. Applying it to a different skin, stake, or format would be a mistake, and the audit authors are explicit about this.

The practical cost of a 12% misread

A 12.4% divergence sounds tolerable until you price it. Consider a river call with a pot of ₹8,000 and a bet of ₹4,000, giving pot odds of 25%. If the HUD says the opponent's river aggression is 28% and you call on that basis, and the true figure is 34%, you have made a marginally losing call. If the true figure is 22%, you have made a marginally winning one. The sign of your expected value flips inside the error band.

Over 100,000 such decisions — a plausible volume for a full-time multi-tabler in India's regulated-adjacent ecosystem, where legal clarity varies by state and many players operate on offshore skins — a 12% systematic misread on a subset of spots translates to a swing that dwarfs most players' annual rakeback. The audit did not quantify this in rupees, but the arithmetic is not subtle.

There is a second-order cost. Players who trust a misread stat begin to adjust their own river frequencies to exploit a phantom tendency. That adjustment is itself exploitable by opponents who are not using the same broken filter. The HUD, in effect, coordinates a subset of the pool onto a shared error.

What a defensible river read looks like

None of this argues for abandoning HUDs. It argues for treating river aggression as a stat with an expiry condition. Three practical filters, in rough order of value:

• Minimum sample gate. Suppress the aggression display below 60 river opportunities. Below that, the number is decorative.
• Position split. If the software supports it, read river aggression separately for in-position and out-of-position. If it does not, discount the blended figure in multi-way pots.
• Pot-type filter. Single-raised and three-bet pots have different river aggression distributions. A single AF across both is a category error.

The audit's own recommendation is blunter: at tables with fewer than 50 regulars and effective stacks spanning more than 100bb, treat any HUD river aggression figure as a prior, not a read. Combine it with table dynamics, timing tells, and bet-sizing history — the inputs the HUD cannot see.

The question the audit leaves open

The 12.4% figure is a snapshot from one table over one sample window. What it cannot tell us is whether the misread rate is stable, improving, or worsening as HUD software adds filters and as opponents learn to play around them. If the rate is falling, the problem is self-correcting and the audit is a historical footnote. If it is rising — because pools are getting smaller, stacks deeper, and multi-tabling more aggressive — then the entire premise of stat-driven river decisions at mid-stakes India tables is on shakier ground than the software's clean interface suggests. The audit did not track the trend. Someone should.