Trust Integral Explorer — The Model Operator

Live blob editor for T = c·f. Companion to The Trust Functional (Goble 2026).

The picture. Both blobs live in output space Y. The model M's real behavior is what the deployed operator actually does; the observer's predicted behavior M is what the observer thinks M will do. Each blob has a boundary (its behavioral support) and a center of mass (the average prediction Ē[Y], the dot inside). Coverage c is the overlap of the two blobs (the τ-inflated observer must cover the model). Fidelity f is the closeness of the two centers of mass. Trust T = c · f. Both factors must be high.

Quick actions

Align makes the observer match the model exactly — a perfect baseline. Reseed regenerates blob shapes while keeping positions and sizes.

Palette

× =

Click the third swatch to override the overlap colour — decoupled from the input pair, so you can do "indigo + orange = cyan" or whatever lands. The multiply still drives the density gradient underneath.

Parameters

Tolerance τ — inflates observer to form V_τ 12

0 (strict)80 (lax)

Required T* — materiality threshold for trust 0.85

low stakes ≈ 0.70capital-affecting ≈ 0.95

Note on stakes: the stakes weight w_L(x,ω) in the trust functional is currently uniform across Y in this explorer — it only enters by setting where you place T*. Higher stakes ⇒ higher T*.

Presets

Pick a preset, then sculpt the blobs in the editor to explore.

Drag a model or observer blob's edge to stretch it · drag the inside to move the whole blob · drag the center dot to reposition the mean

Live metrics

c (coverage)

—

f (fidelity)

—

T = c · f

—

Δ = max(0, T*−T)

—

area M

—

area M

—

overlap M ∩ M_τ

—

d (centroid)

—

Built from The Trust Functional (Goble 2026). Polygon area via shoelace; overlap via 96×96 grid sampling. CoM moves freely within blob boundary; fidelity decays linearly with d.
HITL regime per Construction 5.1 / §7.6: Δ = 0 → automated · 0 < Δ ≤ 0.25 → sampling · Δ > 0.25 → prior verification.