Lock Them Wrong and the Solver Covers for Them

Most node-lock tutorials sell the fantasy first: feed in “they fold too much,” hit re-solve, harvest free chips. Barry Carter’s Postflopizer writeup is blunt about the part the fantasy skips. After you force a mistake at one node, the solver plays as close to perfect as it can afterward to compensate. Real humans who c-bet every flop do not suddenly invent a balanced raise-or-fold response. The printout understates the leak—or invents a tidy villain who never existed.

So treat a lock as a claim about a tendency, not a receipt for EV.

A Lock Is a Claim, Not a Printout

A node is any decision point. Node locking means you fix part of a player’s strategy there—by combo, by hand class, or by action frequency—and let the solver rebuild the rest. GTO Wizard’s help docs describe the same loop: paint the locked strategy, apply, then compare the counter-strategy side by side. Pio-style tools call the result a maximally exploitative response against that fixed behavior.

Two caveats belong in the first paragraph of any lock, not the footnotes:

1. Compensation. Unlock later streets and the solver will salvage EV for the locked player. Carter’s example: lock IP to range-bet a wet flop, and OOP’s check-raise frequency jumps hard—but IP’s response to that raise looks suspiciously balanced. If your pool also folds too much to raises, lock that node too, or leave with the broad heuristic (“raise over-c-betters a lot”) instead of copying every mixed frequency.
2. Street scope. GTO Wizard AI solves one street at a time. A flop exploit can assume perfect later play even when you meant to model a river leak. If the money is on street three, lock street three—or accept that you are studying a minimally constrained leak (MinES-flavored), not a full-tree portrait of the pool.

Baseline GTO still comes first. You cannot name a deviation until you know the equilibrium you are measuring against.

Two Population Mistakes, Opposite Arrows

Population study is useful because pools repeat the same two errors in opposite directions. Node locks only help if the arrow matches the mistake.

Pool leak (direction)	What the lock encodes	Typical counter-arrow
Over-folds to bets/raises	Higher fold % / thinner continue range at that node	Bluff and polar pressure more; widen some thin value that relies on folds
Under-bluffs when betting/raising	Strip air from their aggression; leave value-heavy	Fold more bluff-catchers; stop “floating forever”; raise less as pure air unless you have a second lock that they overfold to raises
Over-calls down	Higher call % with dominated one-pair	Value thinner; cut bluffs that need folds
Over-c-bets	Force bet frequency above equilibrium	Check-raise / trap more; force them onto the node where they err (Carter’s heuristic set)

Upswing’s population work on raises inside 3-bet pots is a concrete direction example, not a universal constant: in their reg sample, fold-versus-raise sat near 39% while the compared GTO line sat near 27%—humans more risk-averse on that node, not “always fold 65% to any 3-bet.” Use your own database filter for your stakes. The lock input is the delta you measured, not a blog number.

Same idea on rivers: multiple public datasets show betting ranges that are lighter on air than equilibrium α would suggest—especially mid-size barrels. Facing that, the counter-arrow is usually tighter continues, not “check-raise JT 100% because vibes.”

Turning a Database Delta Into a Lock

Skip the cinematic “solver said +1.7bb.” Do this instead:

1. Pick one node. Example: BTN vs CO, single-raised or 3-bet pot, turn face a bet after flop call. Narrow until you can count events.
2. Measure fold / raise / call (and, if you can, showdown strength). Compare to a trusted equilibrium baseline for that tree—not to memory.
3. Encode the leak honestly. Prefer locking hand classes (remove most missed draws from a river barrel; keep top pair+) over inventing a precise 73.2% fold slider you never observed. Frequency locks are fine when the sample supports a clear overall fold rate; they are noise when n is tiny.
4. Re-solve and read the arrow, not the decimal. Did bluff frequencies jump? Did medium pairs become pure folds? Did check-raises appear where mixes used to live? That directional shift is the study product.
5. Optional second lock. If you believe they also misplay the next street (never triple-barrel air; always stack off top pair), lock that node too. One-street locks that assume angelic later play will under-sell the exploit against messy humans—or invent balance you will not face.

Partial locking (lock some combos, let others re-optimize) is for when you trust the leak on a subset—“they never bluff missed flush draws”—without claiming you know every mixed frequency in the rest of the range.

One Spot, Direction Only (No Fake EV)

Illustrative line, not a claimed sim output:

$5/$10, ~120bb. BB defends vs HJ open. Flop Q♦8♥4♠, check-call a third-pot c-bet with J♠T♠ (gutshot + backdoor). Turn K♠: now flush draw + double gutshot. HJ barrels ~60% pot.

At equilibrium this is often a mixy continue—call and fold both live near the indifference line for many mid-equity draws. That is the baseline claim. Everything after this depends on the read, not on a screenshot you do not have.

If your pool’s turn barrels here are stripped of air (missed clubs check back; weak Ax gives up; only strong Qx/Kx/sets keep betting), then:

• Calling gets worse: you are paying a value-heavy range with a draw that still needs help.
• Folding gets better against that same range.
• Raising as a semi-bluff only improves if you also believe they overfold one-pair to a turn raise. Value-heavy and sticky is the worst of both worlds—raising JT manufactures a big pot against hands that crush you and do not fold.

So the honest study move is two separate locks, or none:

• Lock 1: remove most bluffs from HJ’s turn bet → solver should push you toward fewer speculative calls.
• Lock 2 (only with evidence): raise their fold-to-raise → then semi-bluff raises can print.

The old article’s “raise almost 100%, EV jumps several big blinds” skips Lock 2 and invents a number. Directional poker does not need that number. If you run the sim yourself, report your EV delta; do not borrow a fictional one.

Where Locks Quietly Lie

Sample size. One showdown is a note, not a lock. Aggregate the spot until the fold/call gap survives noise—or keep the exploit soft (bluff a bit more, not polarize to the ceiling).

Counter-adjustment. A MaxES-style response assumes the locked player never updates. Spamming every suited connector as a 3-bet teaches observant regs to 4-bet lighter. Use the lock as a ceiling, then mix back toward equilibrium when faces change.

Wrong player. Pool overfolds ≠ this station overfolds. Segment: tight regs, calling stations, and maniacs need different presets. A single “population lock” applied to everyone is just a new autopilot.

Cascade without measurement. BeyondGTO-style writeups note that exploiting a river overfold can change turn sizing and even preflop opens. That cascade is real in a full re-solve—and dangerous if you only measured the river. Change one street in study; change one street in practice until the data agrees.

Leave With One Rule

Lock the leak you can count. Read the arrow the re-solve draws. Distrust any output that needs the villain to play perfectly everywhere else—or any blog that quotes precise EV for a sim it never ran.

Node locking is how you turn population notes into counter-strategies. It is not how you mint fictional big blinds.