Take ten databases of mid-stakes regulars and sort the findings by price, and in seven or eight of them the same thing will be in first place: folds on the river with hands that beat the opponent's bluffs. Not because regs cannot call, but because the river is the street where the price of one decision is at its maximum, and there are thousands of such decisions per database. Each of those folds on its own looked like discipline. Together they look like minus bb/100.
Why the river
The pot on the river is the largest of the hand, and a fold gives it away whole. A mistake on the flop costs part of a small pot, a mistake on the river costs the whole big one. And the decision on the river comes up more often than it seems: in a single-raised pot against a late-position aggressor, the defending BB faces a river bet in hundreds of hands per ten thousand.
So a deviation of a few percent in the river call frequency, invisible in the stats, turns into bb/100 in money.
Where the "pool doesn't bluff" myth comes from
At NL25–NL200 the pool really does bluff the river less than equilibrium requires. Hence the conclusion many regs take as a rule: on the river, against a big bet, bluffcatchers should be folded.
The rule is half right. It works against large bets in big pots: there a break-even call needs a high share of bluffs, and the pool does not give it. But it gets transferred to every bet in a row, including those where a call pays off even with a small share of bluffs. And that is where the money goes. The "he never bluffs" rule is also convenient because it is never tested: you fold and never find out.
Where folds cost the most
Across database reviews the typical picture looks like this.
| River bet | Pot | Typical price of folding bluffcatchers |
|---|---|---|
| 45–90% of the pot | small, up to 12 bb | the most expensive category |
| 45–90% of the pot | medium, 12–30 bb | expensive |
| 25–45% of the pot | any | expensive, almost always a call |
| over 100% of the pot | large | the folds are mostly correct |
Small pots are deceptive. A two-thirds bet into an 8 bb pot looks like a meaningless 5 bb. But a call against it pays off with a bluff share of about 29%, and the pool at micro stakes bluffs more readily in small pots: the bet is cheap, the risk is small. The fold here is systematic and systematically losing. Five bb is nothing. Five bb a thousand times over is something.
Large bets in big pots, by contrast, are usually folded correctly. The pool makes them for value, and a fold with a bluffcatcher is right. That is why the overall river fold frequency shows almost nothing: mistakes and correct decisions are mixed in it.
How to check yours
Export the hands where you faced a river bet and held a hand that beats bluffs: second pair and better on dry boards, top pair with a weak kicker, a low flush and so on. Split by three features:
- Bet size relative to the pot.
- Pot size in bb.
- Pot type: single-raised or 3bet pot, in position or out.
In each cell, count what a call would have brought relative to a fold, based on what opponents showed down in similar situations. Where the call is in the plus on a sufficient sample and you mostly fold, the price lies.
The sample matters most here. One cell should have at least a few dozen hands, otherwise the number is random. And before changing your play, read with your own eyes the ten most expensive and the ten most typical hands in the cell: sometimes the "bluffcatcher" turns out to be a hand the opponent beats with his entire value range, and then the fold was correct. Second pair on a board with four to a straight is not a bluffcatcher, it is a souvenir.
What not to do
Do not switch into "call everything" mode. Fixing the fold bias does not by itself make calls profitable: against a large bet in a big pot a bluffcatcher remains a fold. What needs fixing is specific cells, not the frequency. It is easy to go from an overfolder to a calling station, and both times the tracker will say "river: below the norm".
And do not wait for quick confirmation from the winrate. River calls bring results with high variance, and a month of play will prove nothing. What proves it is the calculation per cell over the whole database, and then a comparison of the same cells three or four months later.