Nous holds every Form at once; what emanates are only its instances, each a little wrong. You never see the Form directly — but average enough instances and its shadow sharpens into place. No single instance is the Form, and the mean only approaches it.
A hidden Form (a 2-D point) emanates instances = Form + Gaussian noise. The running sample mean is the maximum-likelihood estimate of the Form and its error falls as σ/√N (the standard error, drawn as a shrinking ring) — the mean provably converges to the Form. Meanwhile the closest single instance stays a finite distance away: no instance equals the Form. Both quantities are measured live.
'Forms', 'Nous holds every Form at once' — Platonic / Plotinian figures. The real content is estimation: a prototype recovered from noisy samples. The Form here is a point we planted, not a truth of the world.
A mind that only ever averages instances cannot know whether the Form it converges on is IN the world or merely the centroid of its own samples — with biased sampling the mean converges confidently to the wrong point. Convergence proves consistency, not correctness.