The 7 Pillars of Poker Strategy
and what our solver says about each

Bet sizing is where most of the money lives.

You can get your frequencies roughly right and still hemorrhage EV if the sizes are wrong — overbetting a board where small bets print, or sizing down on a texture that demands a polar range. The six theories in this pillar describe when the solver bets big, when it bets small, and the specific conditions that tip one direction or the other.

Four of the six are fully confirmed by our solver verification. One — stack depth and sizing — is confirmed on the sizing axis but required a refinement pass when the frequency data turned out to be more nuanced than the theory predicted. And one — river sizing splits by blockers — is directionally confirmed but the specific blocker mechanism couldn't be cleanly isolated from indifference balancing with our current query infrastructure. Where we're sure, we say so. Where we aren't, we say that too.

All data below comes from our own solver, queried across 12 board textures, 6 stack depths, and multi-street runouts. Every number traces to a specific source file and line range.

The wetness parabola: why sizing is not monotonic in board texture

Dry boards get small bets. Wet boards get bigger bets. Very wet boards go back to small bets — or get checked entirely.

That non-monotonic shape is the wetness parabola, and it shows up clearly in our data. Here are the solver's c-bet strategies across 12 flop textures, same position (CO vs BB), same stack depth (100bb), same preflop action — only the board changes:

CO c-bet strategy across 12 flop textures. Single-raised pot, 100bb. Bet frequency drops from 86.4% on J72 rainbow to 32.2% on K94 monotone, but the sizing tells its own story.

Source: cash-baselines.md, Flop C-bet (CO vs BB, SRP, 100bb)

Look at the sizing column. Dry boards (K72r, J72r) use 2.4–2.6bb bets. Connected wet boards (T98, 765, 654) use 2.5–2.6bb — roughly the same or slightly larger. But monotone boards (K94ss, 652ss) drop back down to 1.9bb, with frequencies collapsing to 32–48%.

You cannot bet big unless two things are true

Big bets need two conditions — not one. You need nut advantage and fold equity. Miss either one and the large sizing stops working.

When we tested this across board textures, the pattern was visible everywhere. Here's the solver's sizing breakdown by texture category from our test set:

• KK5 (paired, no clean nut advantage) — 99% small bets. CO doesn't have a clear nut edge — BB defends trips and full houses from their preflop range. With no nut advantage, the sizing stays tiny.
• K94ss (monotone, nut advantage but no fold equity) — 90% small bets. CO has the overpair advantage, but BB's flush draws won't fold — fold equity is gone, so the solver downsizes.
• T98 (wet connected, split equity) — 71% medium bets. Some nut advantage and some fold equity, but neither condition is dominant.
• K72r (dry K-high, both conditions met) — 62% medium + 38% small bets. CO has the strong hands (KK, AK, K-high), and BB's range is full of weak hands that fold. Both conditions are satisfied.

Source: cash-baselines.md, R9 D3 verdict notes (KK5, K94ss, T98, K72r — all CO vs BB, 100bb SRP)

Flop overbets are almost nonexistent

Overbetting requires extreme nut advantage — the kind where your opponent's range is completely capped and cannot have hands strong enough to raise. On the flop, that condition almost never holds.

When we tested overbet frequency across all seven boards in our Round 9 analysis, the solver used overbets (bets greater than 100% pot) on 0.0–0.3% of bets. Across every board texture — dry, wet, paired, monotone — the flop overbet is essentially zero.

This doesn't mean overbets never appear in poker — they do, on later streets. The turn and river are where capping actions (checking, calling) have narrowed one player's range enough that the other player can credibly overbet. But on the flop, with both ranges still wide, the solver says: save your money.

Bets grow geometrically across streets

The solver doesn't pick bet sizes at random on each street. It plans ahead: the flop bet sets up the turn bet, which sets up the river bet, and the whole sequence grows the pot toward the stack size.

When we ran multi-street action sequences (c-bet → call → turn bet → call → river bet) on two different board runouts, the geometric growth was clearly visible:

Multi-street bet sizing progression, CO vs BB, 100bb. Both runouts use blank turn and river cards.

Source: cash-baselines.md, R10 D1 verdict notes (K72r blank-blank, A94r blank-blank)

Each street's bet is roughly 2.5–3.5 times the prior street's bet. That's not arbitrary — it's the solver planning a multi-street pot-building sequence. On the K72r runout, the pot grows from a ~2.4bb flop bet to a ~16.5bb river bet, a nearly 7× increase across two streets.

Notice the frequency side: the flop is bet at 84%, the turn at 22%, the river at 6.2%. The range polarizes dramatically — by the river, only the very top of CO's value range (and corresponding bluffs) is still betting. (This cross-confirms the turn polarity theory from Pillar F: lower frequency, bigger sizing.)

Shallower stacks compress sizes but the frequency story is more nuanced

At shallower stack depths, the solver uses smaller bets. That much is straightforward and confirmed cleanly across all six depths we tested on K72r. But the frequency side — whether the solver bets more often at shallow depths — turned out to be more nuanced than the original theory predicted.

K72r flop c-bet: CO vs BB, single-raised pot. Six stack depths tested in our Round 12 analysis.

Source: cash-baselines.md, R12 D5 verdict notes (K72r, six stack depths). Frequency data not available for 75bb and 150bb.

The sizing column is monotonic: 1.82bb at 20bb, growing steadily to 2.50bb at 200bb. Every additional 50bb of stack depth adds roughly 0.1–0.5bb to the solver's average flop bet. That's the compression the theory predicts — with less room for multi-street maneuvering, the solver keeps each bet smaller.

The frequency column, though, doesn't follow a simple "shallower = more frequent" pattern. The peak is at 50bb (89.6%), not at 20bb. At 200bb, frequency drops to 75.7% — the solver checks more at deeper stacks, presumably because the multi-street risk of building a big pot grows when there's more money behind. But at 20bb the frequency is 84.4%, not the highest point. The theory's "shallower stacks increase frequency" prediction holds for the 50bb–200bb range but oversimplifies what happens at the very shallow end.

River sizing splits across the same hand class

On the river, the same strong hand class uses multiple bet sizes. The theory says blockers drive the split — a set that blocks the opponent's calling range sizes smaller, while one that doesn't block sizes bigger. Our data confirms the multi-size pattern but couldn't cleanly isolate the blocker mechanism.

Here's what we found on the K72r → 2d → Kc runout (a K-paired river). At the hand-class level:

• KK splits between bet12.8 (18%) and bet17.1 (81%) — the strongest hand uses two distinct sizes.
• QQ splits between bet8.6 (74%) and bet12.8 (25%) — a weaker overpair also mixes sizes, but skews toward the smaller option.

Source: cash-baselines.md, R10 D6 verdict notes (K72r → 2d → Kc river)

When we went deeper into the combo level on a different runout (K72r → 2d → 5h, a blank river), the picture got more interesting:

• KK has 3 surviving combos (K♣K♥, K♣K♠, K♥K♠), and all three use ~95% bet12.8 + ~5% bet8.6 — very similar strategies across the combos.
• AKs has 1 surviving combo (K♥A♥), which uses 4 different sizes: bet8.6 (9%), bet12.8 (38%), bet17.1 (38%), bet25.6 (13%).

Source: cash-baselines.md, R13 D6 verdict notes (K72r → 2d → 5h river)

The multi-size splitting is real. But the reason KK's three combos play nearly identically while AKs spreads across four sizes could be blocker effects, or it could be the solver expressing indifference across sizes that have similar EV (the same mechanism that drives mixed strategies elsewhere). We couldn't separate the two explanations with the query infrastructure available. (See the Research notes for more on why this stays partially verified.)