- §0 hook — opens with the striking number; sit-and-go fix to 6-vs-6
- §3f — added "range floor" synthesis section with the worst-hand-each-seat-opens widget; pairs with §3e's counter-example to bracket how wide the format actually goes NEW
- §3b — added Uri's "ranges adjusted by one position" rule
- §3d — rebuilt around the BB defense curve; the "discounted limp" exploit gets its own treatment UPDATED
- §3e — a counter-example: don't overplay NEW
- Three common mistakes NEW
- §4 — At the table, postflop NEW
- §5 — the takeaway, expanded with Uri's "just conceptualize as money" reframe
- Appendix — Safe vs desperate isn't zero-sum NEW
- §0 — A reverse sit-and-go UPDATED
- §1 — How much is a squid worth?
- §2 — There is an ante in each pot, or there's invisible dead money
- §3 — At the table, preflop UPDATED
- §3a — Limping is correct sometimes UPDATED
- §3b — Raise to the size of the squid UPDATED
- §3c — Effective stack shrinks
- §3d — Defending the big blind: discounted limps and giant opens UPDATED
- §3e — A counter-example: don't overplay NEW
- §3f — Where each seat's range floor lands NEW
- Three common mistakes NEW
- §4 — At the table, postflop NEW
- §5 — The takeaway UPDATED
- Appendix — Safe vs desperate isn't zero-sum NEW
- Methodology and caveats UPDATED
§0 — A reverse sit-and-go UPDATED
In cash NLHE, the button limps 0.3% of dealt hands. In Stand-up Game at val=3, the button limps 30%. At val=10, it limps 94%. Same deck, same blinds, same positions. One rule changed.
You treat Stand-up Game like No-Limit Hold'em (NLHE) with wider ranges. It's not. There's an invisible ante changing your sizing math, your limp decisions, and your effective stack depth — and it's bigger than the blinds at every game state.
Here's the format, if you don't know it. A six-handed sit-and-go is a winner-takes-all tournament: six players sit down, one walks away with everyone's money. Stand-up Game is the mirror image. Six players sit down, the loser pays everyone.
The mechanic is straightforward. Every time you win a main pot, you collect a "squid" — a win token. You can hold at most one. In a 6-handed game there are exactly five squids to win. The game ends the moment five of the six players each hold one. The unlucky sixth — the player who never won a main pot — pays the others.
The penalty has one knob, called val. The loser pays 5 × val big blinds total, split evenly across the five holders. The model trains at five settings: val ∈ {1, 2, 3, 5, 10}.
This article uses val = 3 as its running default. At val = 3, the loser pays 15 BB and each holder receives 3 BB.
Two operational labels run through the rest: - Safe — you already hold a squid; you can no longer be the loser. - Desperate — you don't yet hold a squid; you're still in the race.
That's the entire format change. Five squids to win. One loser pays everyone. Everything else is just Hold'em — until you start counting decisions.
Uri's framing line: "A squid tournament is sort of like a reverse sit-and-go. In a sit-and-go, the winner takes all the money. In the squid format, the loser loses all the money."
1 wins everything
1 (no squid) pays everyone
§1 — How much is a squid worth?
Once you know the rules, the next thing to figure out is the value of the squid button. Not the abstract "what does winning a squid mean for me strategically" — the literal number, in big blinds, that a squid is worth at this moment in this game.
The answer is somewhere between 3 and 18 BB, depending on where in the game you are. Here's the bracket.
The minimum: 3 BB
At the moment the game ends, holders receive their split of the 15 BB penalty. Five holders, 15 BB total → 3 BB each. So at game-end, the squid cashes in for 3 BB. That's its hard floor.
A squid you collect early is worth at least this much, because — barring catastrophe — you'll still be holding it when the game ends. The squid you have is a 3-BB asset on the books.
The maximum: 18 BB
Now skip to the other end. Suppose four players already hold squids and two are still desperate — call this down-to-2 (d2t).
Whoever wins the next main pot becomes safe (+3 BB) and the other player becomes the loser (−15 BB). The squid awarded on this hand is worth the full 18-BB swing — the difference between safe and ruined.
So at d2t, the squid is worth 18 BB. That's its ceiling.
The progression: 3.6 → 18
Between fresh game (six desperate players) and d2t (two desperate players), the squid value always goes up. The math is clean: at any game state, the squid swing equals 6 × val / (N − 1), where N is the number of players still desperate.
Squid value table — val × game stage
What is one squid worth (in BB) at each game state? Hover or tap a cell.
value(val, N) = 6 × val / (N − 1)| Stage (desperate left) | val=1 | val=2 | val=3 | val=5 | val=10 |
|---|
The widget shows the value at every val × stage combination. The fresh-game val=3 cell — 3.60 BB — is the implicit-ante anchor for the rest of this article. The d2t val=10 cell (60 BB) is what's at stake when two desperate players are both still chasing the last squid in a high-stakes game.
Uri's framing line: "Every squid button is worth at least three, because it's worth three at the end. And we also know that the last one, when it's two left, whoever loses is at minus fifteen, whoever wins is at plus three — so there it's worth eighteen. So it's something between three and eighteen."
§2 — There is an ante in each pot, or there's invisible dead money
So far the squid is just a number — a 3.60-to-18-BB asset that floats around the table. The reframe that makes Stand-up Game playable is to stop thinking of it as a number and start thinking of it as an ante in every pot.
Here's the move. The game is fresh and you're about to play a hand. Whichever desperate player wins this main pot collects a squid worth 3.60 BB. Everyone at the table has the same shot. So you can think of this hand as having an invisible 3.60 BB sitting in the pot, payable to the winner — except the chips never actually move; the value moves through squid rules, no chips change hands.
That's the invisible dead money.
Uri's framing line: "By saying how much the squid is worth, we're basically saying there is an ante in each pot, or there's invisible dead money. And then we can talk about how poker works basically with the various sizes of antes."
One caveat: this ante isn't symmetric. When some players are safe and others desperate, the squid pressure falls only on the desperate players. The analogy is cleanest when everyone's still in the race — which is most of the early game and which is what this article's running examples assume. The asymmetric-incentives case is in the appendix.
The size of the ante in this format isn't small. At fresh val = 3, the implicit ante is 3.60 BB — already larger than the 1.5 BB of blinds. At d2t, it's 18 BB. The dead money in front of you is bigger than the dead money in the blinds throughout the game, and gets bigger toward the end.
The first thing that changes: your opens need to be bigger. How much bigger — and why — is §3b's story.
§3 — At the table, preflop UPDATED
The implicit-ante reframe gives a handful of practical shortcuts: when to limp instead of raise, how to size your opens, how to think about your stack depth, how wide to defend, and where to push back — even with the squid huge, the bottom of the range still folds. Five sub-sections walk through each, and a synthesis closer (§3f) shows the cumulative picture — exactly where every seat's range floor lands.
§3a — Limping is correct sometimes UPDATED
Push the ante to an absurd extreme: imagine a 1,000 BB squid sitting in the pot. Limping for 1 BB to play for that pot is obviously fine. At a 1:1000 risk-to-dead-money ratio, almost any two cards make money on the call alone, before any postflop play even happens.
Stand-up Game doesn't push you to that extreme, but it pushes you part of the way. With a 9-BB squid plus 1.5 BB of blinds, the dead money is 7× the limp price. The ratio is what matters: a small absolute investment relative to a large pile of dead money makes the limp call cheap, even on hands that would never complete in cash without antes.
Uri's framing line, paraphrased for clarity: "Ranges get wider if the ante is really big. Suddenly limping becomes a good option. ... So there are a lot of limps in squid."
The data backs the claim hard. In cash NLHE, limping from UTG (under-the-gun, the first to act), MP (middle position), CO (cutoff), or BTN (button, last to act preflop) is essentially never correct — the solver completes <1% of the time.
At val=3, BTN already limps 30% of dealt hands. At val=10, almost everyone limps almost everything: BTN limps 94%, even UTG limps 47%.
Limp frequency by position
% of dealt hands the solver completes (limps) preflop · 100bb 6-max
The progression accelerates through val=5, then flattens as limp% approaches 100%.
(SB is special throughout — it acts last preflop, has the smallest blind invested, and converts wide limps into wide all-ins as effective depth shrinks. Treat its curves in every widget as informative-but-extreme. §3f returns to the full hand-by-hand picture once we've covered the rest of the preflop reframe.)
Limps are half the preflop story. The other half is what to do when you do raise.
§3b — Raise to the size of the squid UPDATED
When you do raise, you have to raise bigger than NLHE would tell you to. The mental model: treat the implicit ante as if it were part of the big blind, then size relative to that bigger blind.
Two ways to do this.
Formula version. Take the total dead money in the pot — squid value plus blinds (plus any real ante) — and divide by 2. That's your effective big blind.
effective_BB = (squid_value + blinds + real_ante) / 2
Worked example. Squid value 9 BB, blinds 1.5 BB combined, no real ante. Total dead money = 10.5 BB. Effective big blind ≈ 5 BB. Every "big blind" in your sizing math should be ~5 actual BB.
Shortcut version. When the squid is large relative to the blinds, ignore the blinds and just raise to the size of the squid. That's an effective min-raise.
Uri's framing line, paraphrased for clarity: "When the squid is big, just raise to the size of the squid and ignore the blinds. And that raising to the size of the squid would be the equivalent of a min raise."
The data confirms the heuristic at the val levels you'll actually play. Cash NLHE opens range from 2.3 BB in early position to 2.9 BB on the small blind. In Stand-up val=3, UTG opens to 3.5 BB — basically right on Uri's predicted 3.6 BB squid value. At val=5, UTG opens to ~5.5 BB while the heuristic predicts 6 BB — also a tight match. The "raise to size of squid" rule is a clean prescription at the val levels live games actually run.
The exception is val=10. Here UTG opens to 6 BB while Uri's heuristic would call for 12 BB. Late-position seats (MP/CO/BTN) sit at 7.2 BB across val=3, 5, and 10 — they don't scale up either. The reason: at val=10 the solver mostly doesn't want to raise at all. BTN limps 94% of hands, CO 75%, even UTG 47% — that's the wide-range, low-SPR response to a giant implicit ante, and it's genuine equilibrium. (The model's RL training also tops out at raise 7.2 BB — exploration past that didn't yield improvement, so larger sizes never made it into the action set. The small slice of hands that do raise are stuck at 7.2.)
For the table, the practical takeaway: at val=3 to val=5, apply Uri's rule as written. At val=10+ the rule under-predicts the headline raise number, but the deeper insight survives: you're playing a wider-and-flatter game with limps everywhere. That's the format.
Open raise size by position
Solver's average open raise (BB) · 100bb 6-max · vs unopened pot
A second, related shortcut: ranges shift one position closer to the button. Uri's framing: at val=3, each seat opens roughly the range of the position one closer to the button in cash. The data validates the rule cleanly at the middle positions:
| Position | Cash VPIP | Squid val=3 VPIP | Cash position one-closer-to-BTN |
|---|---|---|---|
| MP | 22.7% | 29.2% | matches CO cash 29.1% ✓ |
| CO | 29.1% | 42.9% | matches BTN cash 41.8% ✓ |
| UTG | 18.5% | 25.6% | overshoots MP cash 22.7% by ~3pp |
| BTN | 41.8% | 67.1% | overshoots SB cash 57.9% by ~9pp |
| SB | 57.9% | 99.6% | (no closer position) |
The rule is dead-on at MP and CO, slightly aggressive at the table extremes. As a table-side heuristic, it gets you in the ballpark: "If I'm UTG at val=3, play it like cash MP." Take the rule until it fails you and lean on the formulas when it does.
The most useful part of the model is the villain's-open re-translation. When somebody opens for what looks like a 5× or 6× raise in raw chips, run it through the effective-BB lens before you panic-fold. A 10-BB open at squid = 10 is a 2× in effective-BB. Postflop ranges, c-bet sizing (continuation bet — the bet by the previous-street raiser), and your default lines should look like a min-raise pot, not a 5× pot.
Squid raise size translator
Translate a Squid raise into its equivalent NLHE-raise scale (in effective-BB)
The single best widget for feeling this regime is the val-sweep on UTG — the seat where Uri's "raise to size of squid" rule lands cleanest because UTG is min-raise territory in cash. Drag from cash through val=10 and watch open size, limp%, and VPIP (the percentage of hands you voluntarily put chips into the pot) all reshape together.
The progression — UTG open across val sweep
Move the slider to watch UTG's range, raise size, and limp frequency reshape
§3c — Effective stack shrinks
The same logic that shrinks the raise-size scale also shrinks the stack-depth scale.
If your effective big blind is ~5 actual BB (squid = 9, blinds = 1.5), then a 100 BB starting stack isn't really 100 BB deep. It's about 20 effective BB deep. Different math than 100 BB; closer to short-stack play.
Uri's framing line: "If the squid is worth ten and a big blind is five, and now you're a hundred deep — it's actually twenty deep. ... You might stack off super wide. Like maybe someone opens to ten BBs and you open-shove deuces, and that would make sense twenty deep."
The example: villain opens 10 BB at squid = 10. In raw chips, that looks like an absurdly large raise — five times a typical cash open. Re-translate it through the effective-stack lens: that's a 2× open in 20-effective-BB-deep play. At that depth, open-shoving 22 over a 2× stops being reckless and starts looking reasonable.
(Uri's shortcut drops the blinds for speed — the exact formula gives a slightly lower number, around 17 effective BB at squid = 10. Refer to the translator above for the precise math; Uri's "20 deep" rounds for napkin use.)
The pattern works in both directions. When the squid is small and the implicit ante is light, your effective depth is close to your real depth. When the squid grows, your effective depth shrinks a lot even though your real chip stack hasn't moved.
§3d — Defending the big blind: discounted limps and giant opens UPDATED
What does "play almost everything" actually mean for the big blind? The answer depends on what villain opened to.
In cash NLHE, BB defense drops sharply with the open size: 59% defend vs a 2× open, 37% vs 2.5×, 16% vs 5×, 9% vs 10×. Cash never reaches 100% defense — the smallest legal open is a min-raise (2 BB total), so BB always faces a real raise that the bottom of the range can't quite afford. Standard NLHE math.
In Squid the implicit ante sitting in the pot dwarfs the price of small opens, and BB does reach 100% defense. At val=3, BB defends 99.5% vs a 1.5× open — Uri's discounted limp exploit: when villain opens way smaller than the squid value, his bet is barely more than a limp and the dead money in the pot makes calling correct on essentially everything. At val=10, BB defends 100% vs anything 2.5× or smaller — and a quarter of those defenses are direct all-in shoves.
BB defense vs CO open size
How wide does BB defend across villain's open size? Three lines: cash, Squid val=3, Squid val=10.
Three patterns are visible in the curve. All three apply mainly at val=3 and val=10 — at cash, there's no implicit ante and no "discounted limp" to chase, so pot odds alone govern.
- Small open (well below the squid value): discounted limp territory. Defend wide. Most relevant at val=10, where the squid is 12 BB and any open under ~7 BB is functionally a discount.
- Open ≈ squid value: defends like cash 2.5× — the point where Uri's "raise to size of squid = effective min-raise" rule lands. BB defense drops back to near-cash levels because villain has now sized correctly for the regime.
- Open ≫ squid value: tighter than mid-range, but still wider than cash because the implicit ante is still in the pot. At val=3, a 10 BB open gets defended at 21% (vs cash 9%) — 12 percentage points of extra defense come from the 3.6 BB squid sitting in front of the bet.
The takeaway: BB defense width isn't a fixed number in Squid. It scales with how the villain's open compares to the squid value.
Uri's framing line: "It's not that you defend the big blind with everything. It's that if someone opens way too small because he doesn't get how to play, then you defend the big blind with everything."
Promote the discounted-limp pattern to a default response habit: if villain's open looks tiny relative to the squid, defend almost everything. A player bringing cash defense habits (~60% vs a small open) into a Squid table is folding 30–40 percentage points tighter than the solver wants — that's the size of the exploit.
§3e — A counter-example: don't overplay NEW
The one habit the implicit-ante reframe pushes you toward — and that you have to actively resist — is overplaying. The pot is bigger. Open sizes are bigger. Limps are everywhere. It feels like every hand should be a "go" decision. It isn't.
Uri's framing line, paraphrased for clarity: "Imagine the squid is huge, you haven't won a hand all game, and you get dealt 7-2o under the gun. Or middle position opens before you. That hand is still a fold. No amount of dead money in the pot turns 7-2o into a profitable open."
The data is clean. UTG opens 7-2o 0.0% of the time — at cash, val=3, and val=10. Same for MP. Same for CO at cash and val=3. The only seats that play 7-2o at high val are BTN (about 20% of the time at val=10) and SB (which plays it 100% of the time at val=10, because SB plays everything).
| 7-2o opens | Cash | val=3 | val=10 |
|---|---|---|---|
| UTG | 0.0% | 0.0% | 0.0% |
| MP | 0.0% | 0.0% | 0.0% |
| CO | 0.0% | 0.0% | 0.1% |
| BTN | 0.0% | 0.1% | 19.5% |
| SB | 0.1% | 93.2% | 100.0% |
The squid being huge doesn't make every hand profitable. It widens the floor — UTG val=10 plays T4s as its bottom hand, which is an 80-percentile-bad cash holding — but there's still a floor below the floor. The "MTT brain" failure mode (described in §5) is to assume that because the pot is bigger, you have to play every hand. You don't.
You play your cards. Just remember the pot is bigger.
§3f — Where each seat's range floor lands NEW
§3a–§3e walked through five preflop adjustments — limps return, opens scale up, your stack shrinks effectively, BB defense scales with how the open compares to the squid, and the bottom of the range still folds. Stack the five together and a question falls out naturally: given all that, where exactly does each seat's range stop?
The data has a clean answer. In cash NLHE, UTG's worst opened hand is A2s — 39 of 169 distinct hand classes get played, and even A2o folds 100% of the time. At val=3 the range expands to 54 hands; the floor drops to K5s. At val=10, UTG plays 104 hands and opens down to T4s — a hand that's a stone-cold fold from UTG in any cash game regardless of stack depth.
The range floor — worst hand class each seat still opens
Each cell is one of the 169 distinct hand classes (rank pair × suited/offsuit). For each position × setting, this is the lowest-equity class the solver still plays a majority of the time. Tap a cell for detail.
39/169) is how many of the 169 distinct hand classes get played at all. UTG cash plays 39 classes; UTG val=10 plays 104. SB val=10 plays every dealt class — even AA never folds. (A "class" here means the rank pair plus suited/offsuit indicator; each class spans 4–12 individual combos.)
Tap any cell for detail. The pattern is clean across positions: each rightward step (cash → val=3 → val=10) walks the range floor down through suited-junk, then offsuit broadway-junk, then trash. The pattern is symmetric across positions too — every seat's floor drops by roughly the same hand-tier increment as you raise the squid value, even if the absolute counts differ.
This is the synthesis view of §3. Read it as: §3a–§3d explain why the ranges expand; §3e clamps the bottom; this widget shows exactly how far the expansion goes for each seat.
Three common mistakes NEW
Three patterns Uri flags — recognize the shape, avoid the leak.
-
MTT brain. Treating Squid like a tournament where you're the "loser" if you don't win the next pot. Players who do this risk too much to chase pots and force-play marginal hands. The squid is not a tournament position. It's money. Calculate it like money and play your cards accordingly.
-
Blanket-defending vs every villain. The discounted-limp pattern (§3d) is real but specific: it kicks in when villain opens too small relative to the squid. Defending 99% vs a properly-sized 7 BB Stand-up open is a leak — that's not a discounted limp, that's a real raise. Read the open size, then decide.
-
Forgetting the pot includes the squid. A 5 BB bet into a 5 BB visible pot at squid=10 isn't a pot bet — the actual pot is 15 BB and the bet is a 1/3 pot. Add the squid before you compute pot odds. §4 has the widget.
§4 — At the table, postflop NEW
The implicit ante reframe doesn't stop at preflop. It changes how every postflop bet reads.
Here's the move. The visible pot at any postflop street looks like just the chips. In Stand-up Game, the actual pot includes the squid value too. A pot that looks like 5 BB might actually be 15 BB once you add the implicit ante. A bet that looks like a pot-bet is actually a 1/3-pot bet.
Uri's framing line: "Look at the flop and you say, reminder there's actually 10 big blinds or whatever number that money in the pot. ... It looks like the pot is five big blinds and he's betting five big blinds, but actually the pot is fifteen. So it's a third pot bet."
Postflop pot reminder
The visible pot understates the dead money by the squid value. Type in any spot, see what the bet actually costs.
The effect is most pronounced on the flop, where the real pot is small relative to the squid. By the turn and river, real pots have built and the squid is a smaller fraction of the total. So the "pot reminder" is mostly a flop-decision habit: before you click call, ask yourself what the pot odds look like once the squid is in.
Concrete data point. Uri's postflop reframe shows up in solver behavior. On a dry K-high flop (Kc 9d 4s) after CO opens and BB calls: in cash NLHE (CO opens to 2.5 BB, pot ≈ 6 BB) CO c-bets 74% of the time. In Stand-up val=3 (CO opens to 7.2 BB, pot ≈ 17 BB) CO c-bets 60% — a 14-point drop in c-bet frequency. Most of that shift comes from the bigger preflop pot the larger raise builds, which compresses SPR from ~16 to ~5 and makes pot-control checks correct on more medium-strength combos. CO still c-bets the majority of the time at val=3 — it's a tilt of the strategy, not an inversion.
The general rule: in Stand-up Game, postflop strategy looks more like a min-raise pot than a 5× pot — because in effective-BB terms, that's exactly what it is.
The full postflop reframe — c-bet frequencies and sizings across textures (paired, monotone, mid-connected), turn and river adjustments, the bluff-vs-value mix once the pot is already inflated — is its own article. This piece sets the lens; the postflop playbook lands separately.
§5 — The takeaway UPDATED
The squid is a number. That number is an invisible ante. Once you price it, three things follow:
- Your opens get bigger and your limps return — the same way they would in any big ante format.
- Your effective stack shrinks even though your chip stack hasn't moved.
- Your big-blind defense scales not with villain's position but with how their open compares to the squid value.
Uri's framing line: "Once you just conceptualize the squid as money and how much it's worth, all the decisions become very clean, very clear — very devoid of 'I want to win' or 'I don't want to lose.' You play your cards and just remember the pot is bigger."
That's the whole article. The format isn't a tournament you have to survive. It's six players, one extra rule, a number you can compute, and a slightly different table to play poker on.
It's a different game. Go play it.
Appendix — Safe vs desperate isn't zero-sum NEW
The takeaway above is the main thesis. This appendix is for readers who want a layer deeper — how the format actually works once you drop the "everyone desperate" simplification the article ran on.
The article so far has assumed everyone at the table is desperate (no squid yet). The format works very differently when some players are safe and others desperate. Two things change.
1. Different incentives at the same table.
Once 1–2 players have squids, those players' situation is different. Safe players (who hold a squid) can no longer be the loser; their incentive in any given pot is just chip-EV. Desperate players (no squid) are still chasing the squid, and their incentive in any given pot includes the implicit ante on top of chip-EV.
Two seats at the same table can have meaningfully different "right" plays in the same spot. A safe player on the button isn't playing the same hand-class width as a desperate player on the button — even with the same hand. Each plays to their own incentive.
Uri's framing line: "Safe players generally should play to their own incentive when they play."
2. Squid math is less zero-sum than NLHE.
In cash NLHE, my chips won = your chips lost. Direct chip transfer at the table on this hand. That's why aggressive equity-denial — bluffs that fold out hands with reasonable equity — is core strategy.
In Stand-up Game, the squid value isn't paid by another specific player at the table on this hand. It comes from the eventual loser of the entire game.
Practical consequence: a desperate player having a profitable call doesn't mean a safe player can't have a profitable bluff. Both can be true at the same time. You don't have to "deny equity" from the desperate guy. A worth-it bluff for you and a profitable call for him are not in direct conflict.
This is a real structural break from how cash poker normally trains players to think. The dynamics are dramatically different — and recreational players who treat Squid like a tighter cash game miss the asymmetry.
Uri's framing line: "It's less of a zero-sum game than regular poker because the money that the guy who's fighting for the squid is going to get — if you already have the squid — it doesn't come at your expense."
One more thing — open theoretical question.
When payoffs are this asymmetric, it's not even clear the format has a single equilibrium. Different players might play different strategies that all qualify as best-response. We treat that question separately in a forthcoming piece on multi-equilibria in Stand-up Game.
Uri's framing line: "I'm not sure there's even a theoretical solution to how to handle it. Probably multiple stable equilibria for that situation."
Methodology and caveats UPDATED
About this article. Drafted from six WhatsApp voice notes Uri Peleg recorded on 2026-04-30 + 2026-05-01 (~12 minutes total). Transcripts at articles/uri-squid-invisible-ante/transcripts.md. Uri Peleg is a WSOP bracelet winner, elite cash player, and Squid Classic specialist — the verbatim framing devices ("reverse sit-and-go", "invisible dead money", "raise to size of squid", "discounted limp") are his. The math layer (the squid value table, the formula derivations, the data widgets) is solver-cross-checked QuintAI commentary; the structural prose connects the six clips into a single arc.
Data rail. Solver data pulled across two sessions via three pull scripts against the strategy_grid_client preview endpoint (v1 rail, preview.rlserv.aceguardianrl.com). 70+ queries total across 8 batches: per-position open metrics × cash/val=3/val=10 (D1+D2), BB defense vs CO 2.5× (D3), BTN open val-sweep (D4), BB defense vs CO 1.5–10 BB opens (D5), CO c-bet on Kc9d4s (D7), 7-2o opens by position (D8), open frontiers per position (D9), BB defense frontiers (D10). Findings note at findings.md.
Reach weighting. All percentages quoted (limp%, VPIP, all-in%, fold%) are combo-level reach-weighted aggregates from the solver's GTO post-processed strategy. Reach = the probability the equilibrium reaches that decision point with that specific combo, after upstream actions. Hand-class examples in the prose (e.g. "open-shoving 22", "UTG opens T4s as its worst") refer to the aggregate behavior of that hand class, not a single combo.
Action-set ceiling. The trained model's preflop action set has seven discretized actions per position: fold, call, all-in, and four raise sizes maxing at 7.25 BB (or 6.0 BB from SB). Identical across cash and all val levels — verified via diagnose_action_set.py. The ceiling is RL-derived, not a hard engineering cap: larger raise sizes weren't kept in training because exploration past 7.2 BB didn't pay off in the loss function. So avg_raise_bb at val=5+ is a lower bound; the unconstrained equilibrium raise is plausibly higher. VPIP, limp%, all-in%, and BB defense are unaffected. At val=10 only ~1% of mass goes to raise 7.2 anyway, so format-level claims still hold.
Format scope. Stand-up Game (Squid Classic) only. Squid Hunt Regular, Squid Hunt Progressive, and Squid Double are out of scope — book-2 trains only Stand-up Game, and the squid value math (6 × val / (N − 1)) assumes the binary one-per-player cap that defines this variant.
Stack depth. All pulls use 100 BB starting stacks (book-2 v1.8.0 default). Different stack depths would change the §3c effective-depth math materially; that's covered qualitatively but not quantitatively across pulls.
Rake. All pulls include 3% rake with a 3 BB cap. The cash-baseline BB defense numbers (e.g. 37% vs CO 2.5×) may be slightly rake-depressed relative to rakeless benchmarks.
Postflop scope (§4). D7 sampled CO c-bet on a single dry K-high flop (Kc 9d 4s) across cash + val=3 + val=10. The 14-pp check-frequency shift between cash and val=3 is a single-board data point; broader board-class coverage (mid-connected, paired, monotone) is logged as a v0.3 query coverage gap and would benefit from a follow-up batch.
Multi-equilibria caveat. Whether Stand-up Game has a single equilibrium or several stable ones is genuinely an open theoretical question — Uri raised it explicitly. v0.2 surfaces the question in the appendix and defers full treatment to a planned separate piece (slug uri-squid-multi-equilibria). Solver output should be read as "one equilibrium the model converged to," not as proof of uniqueness.
Mental-model framing for §3b. Uri's "raise to size of squid" rule is a mental model. At val=3 / val=5 the solver matches it closely at UTG; at val=10 the action-set ceiling (above) truncates the headline raise. The deeper insight — at high val, opening to ~squid value is roughly a min-raise in effective-BB terms — survives the ceiling.