A mind doesn’t just pick the best option — it decides how hard to bet on it. That “how sharply do I commit” is a quantity it sets and tunes, reaching from decisiveness to doubt. It’s called precision, it’s the last rung of Part II, and it’s the bridge to how a mind grows.
Five doors; the plain line runs straight through — take whichever fit: ▶ Picture it, ✎ Try it (commit before the reveal), ✓ Check it yourself, ▸ Go deeper, ⬡ See it in UNI. Runs locally — no account, no network — adapts to light and dark; honours reduced motion.
The one fence that never moves — and in Part II it is stricter. “Precision,” “confidence,” “resolve” name a functional quantity: a gain on a model’s own evaluations. For you we use felt words — the confidence or doubt you feel turning a decision over. For UNI we use functional ones: a number that sharpens a value landscape. A mind’s precision signal has a known functional home in the brain’s neuromodulatory systems, so the design may carry a neuromodulator label — but only ever as a functional analogue: a gain that behaves like that system, never the specific brain chemical and never a feeling of being sure. And precision over policies rides on expected free energy, which is not yet built or sealed in DDNA, so it stays held open. “Built to tune,” never “does.”
No peeking. A chess player, short on time, stares at two candidate moves, scoring each by expected free energy.
Reconstruct two things from Module 14 before you check: the two halves of the score, and whether scoring a move also tells you how hard to commit to it. Commit in your own words first.
The score ranks the options; how sharply to bet on the winner is a separate quantity — and that quantity is this module.
Picture two people facing the same choice, with the same G scores for every option — identical evaluations. One commits instantly and fully to the best. The other dithers, half-takes it, keeps glancing at the runner-up. Same scores, opposite behaviour. So something beyond the scores is driving them — a setting for how decisively the best score gets turned into action.
Precision is a single quantity — call it γ — that sets how strongly the best-scoring policy wins over the rest. Crank γ up and a mind commits hard to the lowest-G option (decisive, focused, even rigid). Turn it down and it spreads its bets, stays flexible, explores by default (open, hesitant, even scattered). It’s the gain on your own evaluations — the volume knob on how loudly the winner is heard over the field.
Four policies with FIXED G scores — the same landscape every time. One control: precision γ. Raise it and the choice snaps to the best (decisive); lower it and the weight spreads out (hedging). The reveal: the scores never change — only how sharply the best one wins. A hesitant mind isn't scoring differently; it's running lower precision.
A mind can’t leave γ fixed, because the right amount of commitment changes with the situation.
So precision is a mind’s live estimate of its own reliability, and tuning it well is wisdom: commit when you should be sure, hesitate when you shouldn’t. Set it wrong and the failures are recognisable — too high: rigid, unable to let go of a plan the world has stopped rewarding; too low: unable to commit to anything, every option forever open. Decisiveness and doubt are not character; they’re a dial, and a mind that tunes it to the world is a mind that acts well in it.
Precision wears a vivid felt face — confidence, resolve, focus when it’s high; doubt, hesitation, overwhelm when it’s low. In a brain, the signal that sets it has a known functional home in the neuromodulatory systems. The design will let a precision value carry a neuromodulator label — but only ever as a functional analogue: a gain on evaluations that behaves like that system, never the specific brain chemical and never a claim that the organism feels anything. The quantity is a number that sharpens a landscape. The felt words are for you.
That completes the engine of behaviour. A mind scores its paths by expected free energy (risk + ambiguity), lets explore and exploit auto-balance inside that one number, acts in a living plan-act-learn loop, and sets precision to say how hard to commit — decisive when the world is trustworthy, open when it isn’t. Perception and action, want and wonder, resolve and doubt — all one free energy, pushed downhill.
But we’ve assumed the whole time that a mind’s model — its priors, its likelihoods, the very shape of its landscape — is fixed. It isn’t. A mind grows: it adds concepts, rewires, changes the model it minimises against — and precision, you’ll find, is exactly the signal that governs when to trust the old model and when to rebuild it. How a model grows — earned, never scripted — is Part III.