Don't Name the Hand—Reweight the Range

"They have KK" is not a read. It is a freeze. RiverOdds' hand-reading notes put the same failure at the top of the list: lock a single combo early, then ignore every street that contradicts it. Deep Poker and VIP-Grinders describe the fix without the jargon theater—maintain a weighted set of hands and let each action filter it. Bayesian updating is just the name for that filter when you admit the weights are probabilities.

You do not need decimals at the table. You need three objects, honest priors, and the discipline to change your mind when the line says so.

Three Objects, Not a Lecture

Bayes' rule in poker shrinks to a multiply-and-renormalize loop. Academic opponent-modeling notes (and older Bayesian poker agents such as Korb et al.'s BPP) write it as: posterior range ∝ prior range × likelihood of the observed action. In English:

1. Prior mass — how much of the range sat in this bucket before the action (position + previous streets + dead cards).
2. Likelihood — how often that bucket actually takes the action you just saw (bet, check, raise, size).
3. Normalized weight — after multiplying, divide so the surviving buckets sum to 100%.

Hands that rarely print the action lose mass. Hands that print it often gain mass. That is the entire engine. The formula on a whiteboard does not make you better; refusing to skip the renormalize step does.

Naming Is the Leak

Walk the classic trap once, then stop doing it.

Cutoff opens. You defend big blind. You decide, preflop, that they "have KK." Flop A♠7♦2♣. They c-bet small. KK is still possible, but Ax now eats a huge share of a credible continuing range—exactly the reweight RiverOdds illustrates. Turn pairs the seven; they barrel. KK is a minority story, not "the hand." If you still hear "KK" in your head, you are no longer updating. You are defending an ego attachment.

Say the range out loud instead: weighted toward Ax, some overpairs that hate the ace, some air that keeps firing. Naming one combo collapses the distribution. Reweighting keeps decision-relevant proportions: how often you are ahead, how often you are crushed.

Caps Delete; Soft Lines Only Reweight

Not every action is the same kind of evidence.

Hard deletes (caps). A check-back in position on a dry flop after raising preflop usually strips most nutted value that wanted to build the pot. A turn check after a flop c-bet does the same job on the next street: strong value that would keep betting leaves the range at high frequency. Treat those as near-deletes for the monsters your opponent is "representing" later—not as mild tips. VIP-Grinders' range notes hammer this: ranges are not static after the open; checks on later streets are where money hides because the value bin thinned.

Soft reweights. A small flop c-bet on a dry king-high board keeps a wide middle. Value bets often; medium hands (pairs, draws) bet some; pure air bets less. Mass shifts toward the middle without deleting air entirely. That is a reweight, not a purge. Confusing the two is how people invent "they always have it" after one small barrel.

Rule of thumb: if the action is one that strong value almost never takes, delete first. If strong value still takes it routinely, only reweight.

Base Rates Beat Beautiful Stories

Base-rate neglect is the Bayesian failure mode that feels smart. "This line is perfect for a set" can be true as a likelihood statement and still false as a posterior statement, because sets start as a tiny prior slice of most opening ranges. Air and missed draws that can print the same three-street story often begin with far more combos.

Before you buy the story, ask two questions:

• How many combos of the nutted hand class were in the prior after board and blocker removal?
• How many combos of medium/air could take the same line at non-zero frequency?

If the nutted class had eight combos and the storytelling air had forty, even a high likelihood on sets does not automatically make sets the majority of the betting range. Korb-style Bayesian poker programs learned the same lesson the hard way: action likelihoods without honest priors produce confident, wrong posteriors.

Likelihoods Are Guesses—Label Them That Way

The old version of articles like this invents tables: value c-bets 90%, medium 50%, air 15%, then prints posteriors to one decimal place. That is cosplay precision. Population pools under-bluff rivers; some regs over-barrel; solvers mix. Your likelihoods are estimates, not measurements.

Calibrate in buckets, not fake decimals:

Evidence quality	What to use
Showdown database / HUD	Fold-to-c-bet, turn barrel, river aggression—adjust the air likelihood first
Population default	Most pools under-bluff multi-street; shrink air's river likelihood vs a solver mix
Unknown opponent	Use wide, conservative bands ("rarely / sometimes / often") and widen your error bar

When the river price demands ~30% equity as a bluff-catcher, you do not need "24.1% bluffs." You need a directional call: does this player's line, in this pool, clear about a third bluffs—or is the betting range still value-heavy? If your honest answer is "probably under," fold. If showdowns and aggression say they spray, call. Inventing 76% value to three significant figures does not make the fold more rigorous; it just hides the guess.

One Worked Loop (Qualitative, Then Price)

6-max cash, ~100bb. Cutoff opens; you defend T♠9♠ in the big blind. Flop K♥T♦5♣ rainbow. You check; they c-bet small (~⅓ pot); you call. Turn 2♦ blanks; you check; they bet larger (~⅔–¾ pot); you call. River 7♣ bricks; you check; they bet ~¾–1× pot.

Prior after the open (rough, not sacred): cutoff has a wide but not button-wide range—pairs, broadways, suited connectors. On K-T-5, bucket into value (sets, KT, strong Kx), medium (Tx, weak Kx, draws), air (nothing + weak overs).

Flop small bet: soft reweight. Value and medium both bet often on this dry texture; air bets less. Posterior leans medium+value, air thinned but not gone. Second pair is fine continuing—you are not "hoping they have KK"; you are priced against a still-wide betting range.

Turn larger bet: harder filter. Many weak medium hands give up; persistent air shrinks again; value that wants to tax calls keeps betting. Posterior shifts toward value relative to flop, but missed draws that still "have a story" remain in the medium/air mix. Do not leap to "they're nutted."

River bet after the brick: now ask the base-rate question. Which value hands happily fire three streets here—sets, two pair, strong Kx that never slowed down? Which bluffs remain—missed straight draws like QJ/J9/98-type hands that called twice and now have zero showdown value? Your T9 is a pure bluff-catcher: beats the missers, loses to the value bin.

Price it. If they bet 18 into ~22, you need roughly 18 / (22 + 36) ≈ 31% equity to break even on a call. That number is exact from pot odds. The bluff share is not. In a pool that almost never triple-barrels air, your Bayesian air likelihood is near floor → posterior bluff share likely fails 31% → fold. Against a maniac who shows up with missed broadway often, raise the air likelihood → call. Same cards, different likelihood table, different posterior. That is the point of the framework.

Review Drill That Forces Updates

Away from the table, pick five hands to showdown. Before the river action resolves, write:

1. Prior buckets after preflop (one line).
2. What the flop action deleted vs reweighted.
3. What the turn action deleted vs reweighted.
4. Your estimated bluff share vs the pot-odds number—call or fold, with one sentence why.

Then reveal. Misses teach calibration: you overweighted the story, ignored a cap, or used a GTO air frequency on a nit. Academic work on belief updating under strategic uncertainty keeps finding the same human (and model) failure—updates that do not scale with evidence strength. The drill exists to punish that inertia.

Hand reading done this way stays boring on purpose: priors, likelihoods, renormalize, respect caps, distrust pretty lines without combo mass. Name fewer hands. Reweight more ranges.