What if we could measure information synthesis?

A live cost surface. Inputs fall in, roll downhill, settle in the mechanism that fits. The terrain deforms as routing learns its shape.

OpenIE · Synthesis The Instrument — cost surface E(x) over coordinate (P, T)
θ × μ RP2040-class Landauer floor 2.87×10−21 J/bit t = 0.0s
Cost surface E(x) = Σ θ(p)·μ(p,H)
Topology over coordinate (P, T) · height = joules / decision
The terrain has structure.
Wells = closed-form · Basins = pattern-bound · Plateau = open-ended synthesis
closed-form
constraint-bound
pattern-bound
open-ended synthesis
full model run
— fps
PHASE 01
A single mechanism
Today, every task runs the full model once. The cost is roughly constant — whatever the input.
Audit trail · routing decisions
# Zone Gear Input Energy

The router is the table.

Each input has a coordinate. The coordinate selects a mechanism. The mechanism executes. Routing becomes lookup, which is both lower-cost and more verifiable than statistical inference at the routing boundary.

c(x) = ⟨ Z(x), E(x), T(x), P(x), I(x), V(x), R(x) ⟩

The coordinate function · maps input to location

execute(x) = A[c(x)](x)

The architecture catalog · lookup, not inference

E(x) = ∑p ∈ P(c(x)) θ(p) · μ(p, H)

Energy per input · Landauer floor θ(p) × impedance-mismatch factor μ on hardware H

The simulation seeds the terrain phase by phase: closed-form work earns its wells, pattern-bound work earns its basins, the model engages only where it fits. The energy ratio you watched climb is what the field already has on the table — it just isn't measured anywhere. The empty wells — cells with occupied neighbors but no inhabitant — are predictions, not gaps. The frontier catalogues them.


See it on real work.

The instrument shows the principle. The walkthrough applies it to the work people do every day.