The same engine the field tool uses — refrigerant data computed with CoolProp and validated against NIST REFPROP. What is different here is that you commit to a number before it shows you one.
This sizes one round duct at a friction rate you choose. It is the same relation your duct wheel plots, so it should agree with the wheel in your hand — check it. It is not Manual D. Manual D sizes a system: total effective length, available static pressure, fittings, and the balance between trunk and branch. This is one input to that.
Two rectangles with the same area are not the same duct. The aspect ratio costs you, and this is where you see it.
This does not give you a tonnage, and that is deliberate. A number here would be a student sizing a system, and a wrong number there gets installed and lives in somebody's house for twenty years. What it does instead is hold every input fixed except one, move that one across its plausible range, and show you how far the answer moves. The output is “this input is worth 40% of your load and you guessed at it”.
Line-set sizing is not reimplemented here, on purpose. The field tool's table is stamped “starting point only; manufacturer tables and long-line guidance govern”, and it stops you outright when the equivalent length exceeds the manufacturer's maximum. A student-grade version that softened either of those would teach the wrong lesson about the one part of this job where the manufacturer, not the rule of thumb, is the authority.