Ten numbers, 8,315 machines - the God Theory of Compressors on trial
By DigiEntropy Engineering · 2026-09-05 · 21 min read
One physics model fitted to 8,315 compressors from 16 brands. The lab opens on a finding, designs a machine that beats the field on paper, then puts its own theory on trial - including the verdict where it is reported as refuted.
About this post — Authored by an AI assistant using DigiEntropy's polynomial corpus, the universal compressor predictor, and the project's chart-generation tooling. Charts are produced by Python scripts that read the same database the live site queries; tables and formulas are pulled from the same engineering modules. Findings reflect the project's current dataset and methodology — send corrections or deeper questions to admin@digientropy.com.
Every positive-displacement compressor ever built is about ten numbers: swept volume, clearance, a polytropic (re-expansion) index, a leakage coefficient, peak isentropic efficiency, the pressure ratio the valves were tuned for, the width of the efficiency curve, a mechanical-loss slope, motor efficiency, and how that motor behaves at part load. The motor's rated power makes eleven dials in the lab, but it is read from the machine, not fitted. Give those numbers an operating point and a refrigerant, and one forward physics model returns mass flow, capacity, power and COP — for a Bitzer recip, a Frick ammonia screw or a Copeland scroll, with no per-brand code anywhere.
The lab at /god-theory fits that model to 8,315 machines from 16 brands, each one to its own maker's EN 12900 / AHRI 540 polynomials. This post walks through it in the order a visitor meets it: an opening that states a finding, a Design Lab that starts from the market's median machine and a guided journey that makes it beat the field on paper, and then the Research Lab, where the same physics goes on trial — including the bench we built to try to prove it wrong.
That trial is not decoration. A model with ten free parameters will fit almost anything over a bounded temperature box, so a low error against the surface it was fitted to proves the optimiser worked, not that the theory is right. The lab scores the theory on data the fit never saw, and says so when it fails.
8,315 - machines fitted - from 16 brands
The lab at /god-theory fits that model to 8,315 machines from 16 brands, each one to its own maker's EN 12900 / AHRI 540 polynomials.
6,147 - trusted fits - median error at most 8 %
A machine counts only if its fit is trusted: median error at most 8 %, and neither isentropic efficiency nor displacement sitting on a solver limit. 6,147 of the 8,315 pass.
12 % - more electricity, bottom tenth - same size, same job
Same size, same job, and the bottom tenth needs 12 % more electricity than the top tenth.
2.64 - design COP, on paper - from 2.28, R404A at 20 kW
On paper, on R404A: COP 2.28 → 2.64 (+16.0 %) at 20 kW by the physics.
…