A verified output is a coordinate.

Every machine-checked artifact — a tactic-line closure, an SMT certificate, an Olympiad lemma, a CompCert recompilation, a Mathlib promotion — occupies a place on the same map. Seven axes describe the shape. Position predicts what produced it, what checked it, and how many joules it cost.

c = ⟨ S, V, E, D, C, R, τ ⟩

The cell coordinate · seven axes, one tuple

The cheapest sufficient solver wins.

A proof artifact carries its production history. What tactic emitted the step, what oracle checked it, what cache the lemma came from, what difficulty the goal sat at, whether the same attempt would repeat — these are not metadata. They are the coordinate. Read the coordinate and the joule bill explains itself.

Most subgoals in any real proof close at the cheap end of the ladder — kernel automation, then a decision procedure, then a small in-domain model, escalating to the frontier only on failure. A system that always reaches for the largest model spends twelve orders of magnitude more than the goal required. The waste is not in the substrate. The waste is in the abstraction layer that decides what to invoke.

The map is a way of seeing, not a verdict. It classifies what is provable, makes the structure of the design space legible, and lets a verification stack be reasoned about before it is built — the same move the sibling stacks make for computation, synthesis, data, state, memory, and communication.

Seven axes describe a verified output.

Three are shared with the sibling stacks — E inherits the Landauer floor from compute and memory, V is shared with synthesis, data, and communication, and τ with data, state, memory, and communication. The other four are proof-specific.

The addressable map.

The product of the axis cardinalities is the discrete coordinate space. Structural constraints prune it to an admissible remainder on which every verified artifact has a place.

5 · 6 · 6 · 5 · 4 · 4 · 4 = 57,600

Discrete cell count · the addressable map of proof-space

57,600

discrete cells

7

orthogonal axes

5

solver tiers on the ladder

Joules per verified output, bounded below.

E(x) ≥ θ(D) · μ(S, V)

The Kepler step · Landauer-class θ for difficulty D, scaled by impedance mismatch μ

The joules to verify an output x are bounded below by a Landauer-class θ for difficulty D, multiplied by an impedance-mismatch μ between the solver tier S and the verifier kind V. Cheap solvers paired with their natural verifiers sit at the floor. Mismatch is multiplicative: a frontier model invoked to close a decidable arithmetic goal is twelve orders above the floor not because the goal is hard but because the solver was wrong for it.

This is the direct child of the relation on stack.openie.dev: E(x) = Σ θ(p) · μ(p, H). It is testable as the table fills with measured (E, S, V, D) tuples reported alongside pass-rate. Empties at well-attested intersections become predictions.

One method, applied across fields.

The same axes recur. E inherits from the compute and memory stacks. V is shared with synthesis, data, and communication. τ is shared with data, state, memory, and communication. A primitive carries consistent coordinates across all of them.