Writing
Notes, write-ups, and the occasional multi-part series. Some pieces stand alone; others are meant to be read in order.
Origins
before any of it was code. surveying, a skidder, an engine that needed fixing.
Earning Trust in Data
how i get from 'data is hostile until proven otherwise' to actually using it.
I Asked Three Rival Models to Kill My Spec
before writing a line of code for a new system, i ran the spec through four adversarial review passes: my own, then three frontier models from three vendors, cold context, sequentially. 25 recorded fixes later it froze with zero open decisions. what the process caught, what it got wrong, and why the misses were still worth money.
Is What I'm Seeing Real?
how do you earn the right to trust a remote-sensing instrument about ground you have never walked? you check it, relentlessly, against ground you know in your body. calibration, not trophies. the affirmative half of 'data is hostile until proven otherwise.'
The Model Doesn't Get to Decide
This is one argument in four parts, built to be read in order, though each part stands on its own. The running question, every part, is the same: on this axis, does the model get to be the auditor, or the judge? The diagram at the foot of each part grows by one box, because the shape earns itself as you go.
From the Field
field notes and build logs from the work itself, tested against real ground.