Bugs That Don't Crash: What an Audit of Our System Simulator Found
By DigiEntropy Engineering · 2026-09-06 · 10 min read
Pipes nobody sized, two pipes that cost nothing, and a COP the canvas could not keep. What five kinds of testing found in the System Simulator, and why none of it ever crashed.
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.
A user wrote to us: "I deleted a component from the circuit on the canvas, then the next rebuild failed somehow." It took four rounds to fix, and every round turned up something new. What those bugs had in common was more interesting than any one of them: none of them crashed. Each one produced a plausible number, a tidy canvas or a green tick, and was wrong.
So we stopped fixing one report at a time and audited the System Simulator, the engine behind Step 6 of Efficiency 360, where a chosen compressor becomes a complete circuit you can price. We looked specifically for failures that do not announce themselves. This post is what we found, and what we changed so the same kind of bug gets caught next time.
5.3 % - COP overstated, every pipe unsized - on the reference circuit
On a reference circuit (the one Auto Design built for our test fixture) leaving every pipe unsized overstates the cycle COP by 5.3 %.
1.86 - COP the optimizer promised - the canvas then showed 1.97
Tested at a load the circuit could meet comfortably, the optimizer promised a COP of 1.86. After Self Adjust the canvas showed 1.97, better than promised, right next to a warning that the condenser was undersized.
0.2 % - evaporator vs compressor - was 5.6 % at a point called balanced
On the reference circuit the evaporator and the compressor now agree to 0.2 % instead of 5.6 %, and the evaporating temperature settles about a quarter of a kelvin warmer.
5,620 - fuzzing checks - on 1,500 random circuits
Fuzzing catches graph edits no hand-written case covers: 5,620 checks on 1,500 random circuits.
1,458 - characters in a pipe id - grew from 170 in four cycles
Every delete and rebuild built the new pipe's name from the old ones, so names grew from 170 to 1,458 characters in four cycles, heading for the browser's storage limit.
44 coils - in the physics sweep - 8 brands, evaporators and condensers
12 - browser scenarios - desktop and phone
The browser scenarios replay every path in this post, desktop and phone, in about three minutes.
Why these bugs survive
A crash gets fixed the day it ships. A quiet failure can live for months, because it looks exactly like a correct answer:
…