Every other page in this course teaches GTO: a strategy built to be unexploitable — the strongest baseline against an opponent about whom you know nothing. But "unexploitable" and "maximally profitable" are different goals. Once you've actually watched an opponent long enough to know a real, repeated leak — not a guess, a pattern — playing straight GTO against that leak leaves profit on the table. Exploitative play means deliberately deviating from the chart to attack a specific known weakness. The trade is real: the moment you deviate, your own strategy stops being unexploitable, and a sharp opponent who notices the pattern can counter it right back. Exploiting is a bet that the read is worth more than the robustness you give up chasing it.

This app's own Play-mode bots are built from exactly the small set of named leaks a real read usually boils down to — the Nit / Station / TAG / LAG / Balanced archetypes the bot system already models. TAG and Balanced sit close enough to a straight GTO baseline that there's no single deviation worth naming; the other three are concrete, nameable leaks with a concrete counter-adjustment:

• Vs a Nit (over-folds): bluff more — a Nit gives up hands a balanced range wouldn't. Value-bet thinner, but carefully, since a Nit's own continuing range is genuinely strong. And fold more readily to THEIR aggression — a Nit's bets and raises are close to always real, not the mix of value and air a balanced range would carry.
• Vs a Station (over-calls): bluff less — a Station pays off bets that would fold out a balanced opponent, so a bluff here just burns chips into a range that isn't going away. Value-bet wider and bigger instead: hands too thin to bet for value against a balanced calling range are profitable bets against one that calls too much.
• Vs a LAG (over-aggressive): bluff-catch wider — a LAG's betting range carries more bluffs than a balanced range's would, so hands that would be a marginal fold against a tighter bettor turn into calls. Let their own aggression build the pot and trap with strong hands instead of leading out yourself.

This app's own postflop model already encodes part of that Nit/Station gap quantitatively, not just as a label. The same ARCHETYPE_NARROW table this app's grader uses to estimate how much of a villain's range survives a bet or a call sets a Nit's callKeepOffset to -0.2 — narrower than the model's own neutral baseline — and a Station's to +0.2 — wider (the matching betKeepOffset shifts how wide their own betting ranges are read; check reads are archetype-independent since v7 — a check caps the range's top the same way for everyone). That's this app's own shipped model quietly agreeing with the two adjustments above: a Nit really does fold tighter than the neutral assumption, and a Station really does keep calling wider. One scope note: that table governs the grading estimate of what villains likely hold; since the bot-realism update, Play-mode bots themselves read opponents neutrally — their personality lives in how they act, not in a skewed read.

Honesty note — read this before trying any of the above. Every grade this app produces — Trainer, Play-mode hand review, and every walkthrough checkpoint — measures exactly one thing: how close your action sits to this app's GTO chart or postflop heuristic baseline. Villain responses are always modeled neutrally, and a verdict never credits exploitative intent. One precise scope note: in Play mode the grader's estimate of what a villain holds does use that villain's true archetype (the ARCHETYPE_NARROW offsets above) — Trainer and the walkthroughs are fully neutral — but knowing what a Station likely holds is not the same as rewarding you for attacking the leak. That means a genuine exploit — the extra bluff that prints against a real Nit, the thin value bet that only makes sense against a real Station — can grade as an Inaccuracy or worse here, even on a hand where it's the more profitable play against the specific opponent it's aimed at. That isn't the grader making a mistake; it's answering a different question than "what beats this villain." The reverse holds too: a play that ignores a known leak entirely can still land on this app's own Best, because Best here means closest to the chart, not closest to maximum profit against whoever's across the table.

And the read has to be real before the deviation pays. Treating every wide-calling opponent you're unsure about as a proven Station, or bluffing into a random unknown as if they were a proven Nit, throws away exactly the edge these adjustments are built on — a deviation is only profitable because the specific leak is real and repeated, not because deviating itself is inherently good.