The C-Bet You Built for NLHE Is Wrong in Squid

Every NLHE cash reg has the same flop c-bet [KB:cbet] sizing tree in their head. 33% pot on dry boards you want to win small. 50% on standard semi-wet boards. 75% on value-polarized spots. Pot or overbet on the rare extreme-nut-advantage [KB:nut_advantage] textures. The tree has been stable across a decade of solver work. It is the first thing a reg reaches for when the flop comes.

Sit down in a Squid Classic game and fire the same tree, and the solver will flag three of the four sizings in red on the first orbit.

Squid Classic is 6-max No-Limit Hold'em with one rule bolted on: the last player holding no squid at game end pays the penalty. That single change layers a continuous forward-looking equity term onto chip EV. BB now defends 95% of hands instead of 52%. BB's added range is 82–88% offsuit junk — no pair, no draw, nothing. And the shape of c-bet sizing — not just the frequency — shifts texture by texture in ways the NLHE tree does not predict.

This article walks three textures that break the cash tree, shows the solver's replacement, and distills a three-rule heuristic for the flop-sizing decision in Squid.

The NLHE sizing tree, anchored

Before the break, the baseline. Cash CO c-bet frequencies across the six textures we pulled fresh through strategy_grid_client.py run from 34.6% (K94ss monotone) to 90.1% (J72r dry-wide) — averages in the 65–90% range on dry and paired boards; monotone sits in its own lower bracket. Cash avg bet sizes run from 1.8bb on KK5 (condensed-range, small polar) up to 2.6bb on J72r (polarized, wider choice set). Overbets barely register as a labelled action on the flop in Cash across this texture set — the average size stays sub-3bb everywhere.

That tree is what every reg learned. Dry boards get the small bet because the range is condensed and nut advantage is modest. Mid-connected boards get the 50%-pot because CO has a small equity edge and BB has enough draws to need protection. Paired [KB:paired_board] and monotone boards — each for their own reasons — get smaller sizing than the textures suggest because neither player holds extreme nut advantage. Overbets are rare because the defender's range is value-heavy at equilibrium and an extreme size just prices out the bluff catchers.

The whole tree rests on a single assumption: the defender's range is concentrated and value-heavy. Once the defender's range changes shape, the tree changes with it.

The Squid break

At val = 3, BB defends 95.8% of hands against a CO 2.5bb open — up from 51.8% in Cash (+44.0pp). The large majority of the added combos are offsuit junk; the remainder is suited junk and a handful of suited connectors. No new premium hands, no new strong hands, no new suited Ax. The Squid defender is the Cash defender plus a large junk tail.

Source: fresh endpoint pull per METHODOLOGY §4f (BB preflop defense aggregate, cached in ../nick-desperation-polarization/pull_squid_hero_last.out.json). Mechanism spec in book-2 Part 3 §3.1 + Part 4 §4.1. See Methodology & caveats.

That junk tail does two things at once. It raises c-bet frequency almost everywhere — the junk folds to any c-bet, so aggression captures pure fold equity [KB:fold_equity] that the Cash tree didn't have available. And it shifts the shape of the opponent's range from concentrated (Cash) to bimodal (Squid): the Cash-defending combos almost unchanged, plus the junk lobe stuck on the side.

The Cash sizing tree is tuned for concentrated ranges. On a bimodal range, the tree breaks in three different directions depending on what the board does to each of the two modes. Here are the three textures where it breaks cleanest.

Texture 1 — Dry A-high (A94r): from 33% toward 50% and growing into larger sizes

The NLHE standard. A94r is the textbook small c-bet board. Cash CO c-bets 66.4% at an average of 2.3bb — a ~33–40% pot sizing is the modal action. Larger-than-pot bets barely register as a labelled action on this texture in Cash.

The Squid solver. At val = 3, CO c-bets 98.2% of the range at avg 2.6bb. That's a +31.8pp frequency shift — one of the largest positive flop-c-bet frequency deltas on any non-monotone board in the fresh scan. Sizing climbs further with stakes: v5 pulls avg to 2.9bb, v10 to 4.0bb — the range gradually opens up larger sizings as the penalty grows. A line that was essentially absent in Cash becomes a live option on this class of texture.

Source: fresh endpoint pull per METHODOLOGY §4f (A94r cells cached in ../nick-desperation-polarization/pull_squid_hero_last.out.json). Mechanism spec in book-2 Part 4 §4.1 and §4.7. See Methodology & caveats.

Why it shifts. The compositional mechanism drives it. BB's Cash defense on A94r is 51.8% of hands, concentrated on suited Ax, broadways, medium pairs, and suited connectors. In Squid v3, BB defends 95.8% — and the bulk of the additions are offsuit junk. Offsuit junk on A94r has no pair, no ace, no draw. It folds to any size. So CO's range bets at near-saturation.

But frequency isn't the full story. The junk folds to any size; that's the EASY part. What changes at the top of the range is that BB's non-junk defending half now includes more marginal hands that will call a small c-bet and fold to a large one — hands like KQ, QJ, JT. A small c-bet leaves that marginal value on the table; a larger bet extracts from the marginal callers AND still folds the junk. And for the top of CO's value range (AA, AK, some KK combos), larger sizings become live because the combined fold equity against the junk tail plus value extraction against the marginal half justifies a size the Cash tree would never pick — which is exactly what the avg-size creep from 2.3bb Cash to 4.0bb at v10 shows in aggregate.

Replacement play. On dry A-high and dry K-high rainbow at val = 3, c-bet 95%+ of the range. Default size closer to 50% pot than 33%. Mix in larger sizings with your best value combos (AA, AK-of-top-pair, KK with the blocker). The 33% flop c-bet as a default does not exist on these textures in Squid.

Texture 2 — Mid-connected SRP (765 two-tone): 50% gives way to mostly check

The NLHE standard. 765 two-tone is the textbook 50%-pot c-bet spot. CO has a small top-pair-plus equity edge. BB has enough draws that a medium size protects against free cards. Cash c-bet frequency is 69.5% at avg 2.6bb — a 40–50% pot sizing for a contested mid-wet board.

The Squid solver. At val = 1 (the canonical value for this reversal; pattern holds through v10), CO c-bets 57.9% — a −11.6pp drop from Cash. The drop holds across the val ladder: v3 55.5%, v5 56.1%, v10 61.4% — the frequency stays materially below the Cash baseline even at high val while every other texture saturates upward. Avg sizing on the hands that do bet climbs from 2.6bb Cash → 3.7bb v3 → 7.8bb v10 — growing sharply. The structural signal book-2 §4.2 flags — CO's best hands slow-play MORE on this texture, not less — sits inside the frequency data: the range bets less overall AND sizes polarize upward when it does.

Source: fresh endpoint pull per METHODOLOGY §4f (765 two-tone cells in pull_dan_textures.out.json; canonical board 7s6s5d per squid-deltas.md:193). Mechanism spec in book-2 Part 4 §4.2. See Methodology & caveats.

Why it shifts. 765 two-tone is one of three mid-connected textures (alongside 654 and 876r) where BB's Squid-added range is loaded on the board. The hands BB adds to its defense in Squid aren't just junk that folds — they include 54s, 65s, 76s, 87s, small pocket pairs. Every one of those hands smashes 765. CO's high-card opening range (AK, AQ, KQ, broadway) misses 765 entirely — and on the two-tone variant, BB's suited connectors and suited-junk tail also pick up flush draws, amplifying the range edge further. Range advantage [KB:range_advantage] flips to BB.

This is the rare case where the junk-tail doesn't save CO's aggression. The non-junk half of BB's range is stronger than CO's entire range on this specific texture. Betting most of the range becomes dominated by checking. The hands that DO bet are concentrated value (89, 77, 66 for set value) plus the occasional semi-bluff — which is why the average sizing grows: CO's betting range is narrower and more polarized than Cash's. The replacement sizing is counterintuitive: bet LESS often, but bet BIGGER when you do.

Scope note. The reversal is narrow. It applies to exactly three mid-connected boards: 654, 765, 876r (two-tone and rainbow variants both flagged in book-2 §4.2; two-tone is the canonical value tested here). Other connected boards don't reverse — 543, 432r, 987r, T98 all have positive Squid deltas. And in 3-bet pots, the entire pattern inverts: 765 in 3BP runs Cash→v3 +17.5pp, because BB's 3-bet range is AA–TT / AK / AQs and does NOT contain the connectors that made SRP dangerous. With those hands filtered out, 765 reverts to a CO-favorable texture.

Replacement play. On 765 / 654 / 876 mid-connected boards in single-raised pots at val = 3+, check back a meaningful chunk of the range including marginal overpairs. Your c-bet frequency drops from ~70% Cash to ~55% at val 3+. When you DO bet, size 50–60% pot (3.5–4.0bb into a 6bb pot) at val = 3 and larger still at v10 — your betting range is narrower and more polarized than Cash's. In 3-bet pots, revert to aggressive c-betting — the reversal disappears.

Texture 3 — Paired low (772 rainbow): the exception that proves the pattern isn't uniform

The NLHE standard. 772 is a condensed-range c-bet texture in Cash. Both ranges miss; neither has much nut advantage. Cash CO c-bets 68.2% at avg 2.0bb — a modest size that keeps the pot small on a non-contested board.

The Squid solver. At val = 3, CO c-bets 82.8% (+14.6pp, frequency grows as the junk folds). But avg sizing barely budges: Cash 2.0 → v1 2.0 → v3 2.1 → v5 2.1 → v10 2.4. Across the entire val ladder, 772's sizing drifts a total of +0.4bb — essentially flat. Every other non-monotone board in the fresh scan opens meaningfully larger across the same ladder: A94r +1.7bb, K94ss +1.8bb, J72r +3.9bb, KK5 +0.8bb, 765 +5.2bb. 772 is the outlier.

Source: fresh endpoint pull per METHODOLOGY §4f (772r cells in pull_dan_textures.out.json). Mechanism spec in book-2 Part 4 Table 8. See Methodology & caveats.

What 772 tells us about the pattern. The easy read of Textures 1 and 2 is that every texture sees its sizing lever move in Squid — A-high grows, mid-connected grows harder and polarizes upward, paired-high grows, monotone-adjacent grows. The Squid overlay, on that read, pulls sizing up pretty much everywhere. 772 is the counterexample that forces a tighter mechanism. The pattern is not "Squid sizes up uniformly." It is narrower: sizing grows where a top-of-range nut-advantage mode exists to carry the larger size. On 772, that mode is missing on both sides, and the sizing lever simply doesn't move — even as the frequency lever fires hard (+14.6pp).

Why it stays flat. On 772 the Squid overlay adds fold equity at the BOTTOM of the range — junk folds — without creating any nut advantage at the TOP. Neither side flopped two-pair-plus much; neither side has a polar line to open. The junk-fold gain is absorbed entirely as frequency (the c-bet% jumps from 68% Cash to 94% at v10) while the sizing distribution stays anchored in the 1.8–2.8bb buckets. Compare A94r, where the same junk-fold gain combines with CO's top-of-range (AA, AK, top-pair-top-kicker) to justify larger sizings against the marginal callers on BB's non-junk side. No top-of-range mode on 772, no sizing movement.

In shorthand: frequency scales with fold equity; sizing scales with nut advantage. On 772 those two levers fire independently — frequency jumps, sizing holds.

Replacement play. On paired low rainbows (772 and close analogs — 332, 443, 554 pending confirmation in a broader sweep), c-bet 80%+ of the range at val = 3+ but keep the size SMALL — 2.0–2.5bb into a 6bb pot, roughly 30–40% pot. This is the texture where the NLHE reflex ("small c-bet on no-nut-advantage board") transfers intact. Squid amplifies the frequency but preserves the sizing. The broader lesson: the Squid overlay doesn't pull sizing up on every texture — it pulls sizing up on textures that already had a nut-advantage mode to carry it. Paired-low flops don't, so they don't.

Mechanism in one frame

The three textures distill to one question the cash tree doesn't ask: what SHAPE is the defender's range, not just its density?

NLHE tree: the defender's range is concentrated and value-heavy. C-bet sizing scales with nut advantage.

Squid defender: the range is bimodal — the Cash-defending combos (roughly unchanged) plus a large offsuit-junk tail that folds to anything. That bimodality does three different things to the sizing decision:

• Dry / A-high / paired-high / monotone — the junk tail amplifies nut advantage. Large bets and overbets win against the fold-heavy tail AND extract better from the marginal portion of the non-junk half. Sizing grows, frequency saturates. (Texture 1)
• Mid-connected (the 654 / 765 / 876 trio, two-tone canonical) — the junk tail is still junk, but the non-junk half of BB's range is stronger than CO's opening range on this specific texture. CO's range advantage inverts; check back most hands, bet the polarized top bigger. (Texture 2)
• Paired low — no nut advantage exists for either side. The junk tail folds cheaply — but no polar line is open on this texture, so sizing stays condensed even as frequency climbs. (Texture 3)

The NLHE sizing tree was built for one distribution. Squid's defender is a different distribution. Three textures, three different failure modes of the tree, one common cause.

The three-rule heuristic

Rule 1 — Range advantage decides the BRANCH. If NLHE gives CO a range advantage (dry rainbow, A-high, paired, monotone), Squid amplifies it: c-bet 95%+ at val = 3, size up from Cash, mix 5% overbets with the top of the value range. If NLHE gives BB a range advantage — specifically on 654 / 765 / 876 mid-connected in SRP — Squid amplifies the reversal: check back your premiums, let BB's non-junk half lead into you. The Cash range-advantage test doesn't change; the consequence on each side is bigger.

Rule 2 — Nut advantage decides the SIZING. Frequency scales with fold equity (junk folds to any size). Sizing scales with nut advantage (polar lines require it). On dry A-high the nut advantage is real and growing — size up. On paired low the nut advantage is absent — size down, even as frequency climbs. Don't let the penalty-driven frequency jump seduce you into sizing up on textures where neither side has nut advantage.

Rule 3 — Check the range SHAPE, not the board wetness. The NLHE shorthand "wet = small c-bet, dry = small c-bet, monotone = check" assumes a value-heavy concentrated defender. Squid's defender is bimodal. Ask first what BB's added junk looks like on this exact texture: does it fold to a standard c-bet (most textures), does it connect hard (654 / 765 / 876 in SRP), or does it fold but with no polar line to replace it (paired low)? The answer to THAT question sets the sizing, not the wetness label.

What doesn't actually change

Five guardrails where the cash tree still applies:

• 3-bet pots. BB's 3-bet range is AA–TT / AK / AQs — no connectors, no junk. The mid-connected SRP reversal (Texture 2) disappears completely in 3BP; 765 becomes one of the best c-bet textures for CO. Inside a 3-bet pot, you can default closer to Cash theory.
• Multiway. All three textures' sizing data is heads-up. In multiway pots, additional defenders dilute the junk-fold effect because somebody likely does have a draw or a piece. The monotone and A-high conclusions attenuate with each extra player; the reversal on 654/765/876 persists but less sharply.
• Val < 3. The sizing story matures at val = 3; below that, textures are closer to Cash. The v10 extremes in the data (sizing climbing to 8bb on 765) are the asymptotic limit, not the canonical play.
• Later streets. Turn and river sizing revert toward Cash-like logic on blank runouts — the squid-equity overlay captures most of its value at the flop c-bet call. The one exception is the river on dry blank runouts, where polarization intensifies sharply. Turn/river resolution sits outside the Postflop.cbet rail surface; cross-reference the mechanism spec in book-2 Part 5 §5.6.
• MP postflop. The source data concentrates on CO. MP-specific postflop behavior at Squid val = 3+ is thin in book-2. Apply these rules conservatively from MP until the data matures.

Squid Classic is NLHE with one new equity term. The cash sizing tree isn't wrong; it's tuned for a narrower distribution of opponent ranges than Squid produces. Rebuild the tree around range SHAPE and you have a cash-adjacent framework that survives the penalty.

Where to go deeper

This article is the cash reg's entry door. The deep reference lives in book-2 — Squid Classic, co-authored with the QuintAce research team:

• Part 4 — Flop C-Bet — the full seven-finding walkthrough across textures. Dry rainbow, A-high, paired, monotone, the mid-connected exception, slow-play theory splits, overbet usage. Source for §3 Textures 1–3 above.
• Part 5 — Later Streets — turn barrel decomposition, delayed c-bet, probe frequency, limped-pot aggression, check-raise response, river polarization.
• Parts 1–3, 6–9 — preflop, BB defense, 3-bet pots, hero-last state, open questions, and the 25-point actionables cheat sheet.

Read book-2 on internals.quintace.ai.

Methodology and caveats

Review-stage verbose methodology per article pipeline convention. Compressed for final publish.

Data source. All numeric claims in this article are fresh endpoint pulls through strategy_grid_client.py against the Gameplay-AI range-viewer preview endpoint (https://preview.rlserv.aceguardianrl.com/api/strategy_grid), driven by Postflop.cbet with SquidConfig.hero_last("BB", val=V, num_players=6). Pull scripts: pull_dan_textures.py in this folder for 765 (two-tone, 7s6s5d, canonical per squid-deltas.md:193) / 772r / J72r / KK5; A94r / K72r / K94ss and the BB preflop-defense aggregate are reused from ../nick-desperation-polarization/pull_squid_hero_last.out.json and ../uri-nlhe-reflexes-to-unlearn/pull_k94ss_cash.out.json (same rail, same date, same endpoint — shared c:/tmp/sg_cache_articles cache dedupes). Per METHODOLOGY §4c (rail discipline) + §4f (rail-divergence policy — fresh endpoint always wins), book-2 v1.8.0 squid-deltas.md locked tables are cross-references for mechanism context, not primary sources for numeric values. The mechanism spec for each texture lives in book-2 Part 3 §3.1, Part 4 §4.1 / §4.2 / §4.7, and Part 5 §5.6; the numbers ship from the fresh pulls.

Base conditions. Unless stated otherwise:

• Classic mode (the only Squid variant the research covers)
• 6-max, 100bb effective
• Fresh state — all players start with zero squids at the time of the decision
• Heads-up postflop (CO vs BB SRP)
• val = 3 as the canonical anchor; val = 1 used in Texture 2 because the reversal data is published at that value
• Standard 2.5bb open

Per-category breakdowns not re-pulled. The Postflop.cbet rail surfaces aggregate reach-weighted bet_pct / check_pct / avg_bet_bb via Metrics.postflop — not per-hand-category breakdowns (AA–JJ premium-bet frequency, combo-level overbet share, or AA-specific slow-play rates). Texture 2's "premium slow-play" and Texture 1's "overbet share" mechanism claims are cross-referenced to book-2 Part 4 Research notes as qualitative mechanism, not quoted as fresh-pull numbers. Per §4f, this article ships only what the fresh rail reports.

Checkpoint drift on Texture 2. Book-2 Part 4 Research notes flag that 654 / 765 / 876 Cash→v1 magnitudes vary across training runs, while the direction (negative c-bet delta in Squid SRP on these three boards only) is stable. Our fresh endpoint pull on the canonical 765 two-tone (7s6s5d) returns Cash 69.5% → v1 57.9% (−11.6pp), which matches the book-2 published magnitude exactly. Note: KI-6 disambiguation batches run 2026-04-16 against the current checkpoint recorded the same Cash→v1 delta at −28.9pp for this board, so the published and current-checkpoint numbers disagree at a −17pp magnitude. The fresh rail surfaced here agrees with the published book-2 value; the structural finding ("these three mid-connected boards invert — and only these three") is stable across all three readings. Fresh endpoint always wins per §4f; direction and relative ordering are robust, per-cell magnitude remains the unstable dimension. Mechanism in book-2 Part 4 §4.2 + KI-6 notes.

No EV values reported. Book-2 known-issues register flags three issues (KI-1, KI-2, KI-4) that affect EV output reliability in Squid. This article cites policy/frequency outputs only (c-bet %, bet frequency, average bet size in bb). The Squid regression smoothing artifact (KI-2) doesn't propagate to frequency/size outputs.

Reach verification (§6c compliance). All claims are aggregate / class-level (by texture, by hand category, by val) from book-2's §6a aggregate-scan data pattern. Per METHODOLOGY §6c, hero_reach is a reliability lens for hero-specific mixed-strategy outputs (§6b pattern) — it doesn't apply to class-level aggregates. No combo-specific mixed-strategy claims appear in the article.

Scope limits. This article stays inside book-2's tested scope:

• SRP / single-raised pots for all three textures. 3BP scope callout in Texture 2.
• Heads-up postflop. Multiway dilutes the mechanism; noted in "What doesn't change."
• Val ≥ 3 as the canonical zone. Below val = 3 the shifts attenuate.
• CO postflop. MP postflop data is thin in book-2; flagged in the closer.
• Classic mode only. Blood Battle and Double mode are declared but not covered.

Widget selection (§8b compliance). Cycle 1 scan of docs/poco-widget-inventory.md completed 2026-04-21. Four widgets shipped:

1. Texture Sizing Picker (W1) — ad-hoc build; no library widget compares sizing DISTRIBUTIONS side-by-side per texture. Tab-selector + stacked bar.
2. Pick Your Size Quiz (W2) — adapted from CbetBoardTextureV2 / BetSizingQuizV2 shape; MC-with-reveal across three rounds.
3. Range Shape Explorer (W3) — ad-hoc build; no library widget compares two defender-range SHAPES (concentrated vs bimodal) side-by-side. Tap-to-reveal spatial visual.
4. Three-Rule Cards (W4) — adapted from ExampleCard/Framework primitives; flippable-card reveal.

Four distinct interaction shapes (tab-compare / MC-with-reveal / tap-spatial-reveal / flip-card). No widget duplicates a static chart or table. Count = 4, inside METHODOLOGY §8b [2, 6] range. Cycle 2 + Cycle 4 screenshot-review logs in deviation-log.md entries D-2 and D-3.

Voice and author attribution. This v1 runs in third-person narration. Daniel Dvoress is this article's author (Tier 1 coach, GTO Lab co-founder, cash-game specialist); his first-person voice enters in the final.md pass after he reviews the deployed page. Per METHODOLOGY §1b, author review runs against the live URL, not a doc draft. Coach relationship mode: technical reviewer. Register: peer validation — solver "shows" or "draws," Dan "explains," QuintAce "grounds the numbers." No "we grade Dan" framing.

Numbers flagged for follow-up before publish. (Dan may edit any of these during review.)

• The article cites the 50% pot replacement on dry A-high as a default in §3 Texture 1 based on the avg sizing climbing from 2.3bb Cash → 2.6bb v3 → 4.0bb v10 on A94r (fresh-pull values). Dan may want to tighten this to a specific pot-percentage range based on his GTO Lab cash-reg experience with Squid.
• The article cites the 60–75% pot sizing on 765 when CO does bet based on avg 4.2bb at v3 into a ~6bb pot. Dan may want to refine this.
• The 3-rule heuristic wording is derived from the data; Dan should stress-test against his own GTO Lab framing and reword if he'd describe the decision differently.