Every number we publish is computed, not asserted.

Defense concept work fails in a predictable way. A number gets written into a slide, the slide gets copied into a document, the document gets quoted in a proposal, and by the time anyone checks it nobody can say where it came from. Two documents in the same package quietly assume different values for the same quantity, and the first competent reviewer finds it.

So our engineering package is built the other way round. Every analytical figure lives in a runnable script, and nothing appears in prose that a script does not produce. There are eight of them, covering human exposure limits, the DRIFT effector and airframe, the SIROCCO propagation chain and the gyrotron source.

The cross-checks matter more than the models. Any quantity that appears in more than one place is asserted in a single harness that fails loudly. The tube output has to actually feed the aperture power the link budget assumes. The mass budget has to close inside the Group 2 ceiling. The modulation band has to clear the photosensitive-epilepsy band through the eighth harmonic. Ten of them run in about a second, and a change that breaks consistency fails rather than ships.

The drawings are generated from the same figures. The sheets are drawn to ISO 128 with a title block, tolerance block, section hatching by material and a parts list, but they are produced by a script rather than drafted by hand, so a dimension change propagates to the sheet instead of being re-drawn and re-checked. When the DRIFT rotor grew from 0.90 to 1.00 metres to meet an acoustic requirement, the span, the disc loading, the hover power, the battery mass and the drawing all moved together.

It also means we find our own errors. The tip-to-tip span on this site read 2.2 metres for a while because the first version used an arbitrary multiplier rather than the rotor layout. Computed properly from a quadrotor with 50 millimetre tip clearance it is 2.05 metres across the flats and 2.48 on the diagonal. That correction propagated through four documents and a drawing in one commit.

None of this makes the systems work. It makes the claims checkable, which is a different and smaller thing. Nothing in the package is verified by test, no hardware has been built, and we say so on every sheet and in every document.