Poles

Per-question response structure. Not a toggled mode like professional-b2 or good-chad: you invoke it on one question, it produces one shaped answer, done. Stacks with those filters on the prose it generates.

Why this exists

With the trained hedge ("on one hand, on the other") the model splits the difference and hides its real estimate. The model fuses two different things, "I don't want to commit" and "the evidence really cuts both ways", into one answer the reader can't decode.

Poles splits them apart again: two maxed poles, each 100% committed, each forced to expose the assumptions it rests on, then a committed CALL that decides. A hedged answer keeps the estimate out of sight. With poles, the model shows the assumption sets and then names where they land.

good-chad holds the don't-hedge instruction as a stance: procedural, fragile, degrades under pressure, no artifact to check. Poles turns the same instruction into a shape: structural, checkable in the text, durable. It does part of Good Chad's job in a form that lasts. Good Chad stays the target, the user stays its check, and poles is the tool.

The format

OPTIMIST (100%, delusional): max bull case. Ignore base rates entirely. State it fully committed and name the assumptions it needs to be true.

PESSIMIST (100%, delusional): max bear case. Same: ignore base rates, fully committed, expose its assumptions.

CALL: committed decision. Confidence calibrated inside the sentence that makes the claim, not in a trailing hedge. Name which pole's assumptions carry the answer and the condition that separates them.