I built a web ballistics engine and checked it against published firing data - where's my physics wrong?

I've been building a free web ballistics calculator, and I care more about the engine being correct than pretty, so this seems like the right place to have the physics picked apart.

It's a fixed-step point-mass trajectory integrator, Mach-indexed Cd off standard drag tables. Two things I'd like eyes on.

G7: for loads with a measured G7 BC, I run the standard G7 table with a scale constant derived straight from physics (air density, cross-section, the usual), no curve-fitting. It reproduces Applied Ballistics' published firing-test numbers for the Berger 155 VLD at 1000 yd to about a third of an inch of drop, exact retained velocity, and about 0.4 inches of wind. Honestly surprised me - I expected to need a fudge factor and didn't.

G1: here's the ugly part I'd rather you find than a stranger. The G1 Cd table I'm using runs below the standard BRL G1 curve, with a scale constant compensating for it. It's internally consistent and matches manufacturer drop tables to about 0.15 MOA out to 500, but it's a fitted system, not first-principles like the G7 path. I want to know if that's going to bite me somewhere I haven't tested.

Full method, the load-by-load comparison, and the known weak spots (transonic, rimfire, advertised-BC optimism) are on a public page. Not selling anything - it's free, no ads - I just want the physics wrong-ness found before anyone trusts the numbers. Link in the comments.

Where would you attack this?

Here's the accuracy + methodology page with the load-by-load breakdown against manufacturer tables and the Applied Ballistics firing-test case: https://caliberatlas.com/accuracy

The tool itself is at caliberatlas.com if you want to run one of your own loads through it. It's mine and it's free — no ads, nothing to sell — I just want the physics poked at. Happy to get into the weeds on the integrator, the atmosphere model, or how I derived the G7 drag constant if anyone wants to dig in.