Built by engineers who wanted the range to come to them.

The simulation core produces the trajectory. Engines — the browser viewer today, Unreal next — are renderers and sensor rigs that consume it. Nothing ever steps physics inside a renderer, so a replay and a batch result can never disagree.

Every parameter extracted from a datasheet carries a verbatim quote, the page it came from and a confidence. A gap becomes an open question for a human, or a flagged estimate — never a silent guess. The Test Director agent we are building is held to the same rule: it proposes, people approve.

Batches run headless and cheap. Only the runs you select pay for photoreal rendering or sensor capture, and every product — replay, live flight, dataset — reads the same trajectory files.

Results are unvalidated design-iteration estimates until a validation programme says otherwise. The disclaimer is on every report, in every export format, and cannot be removed. Calibration against a reference airframe is a CI gate. We do not use accreditation language until it is earned.

MIT, BSD and Apache in product code. LGPL only as an external process or an unmodified library. No GPL or AGPL, ever. Every borrowed component is recorded in a notices file in the same change that introduces it, so the whole dependency chain is auditable.

No real defence CAD, customer technical data or real threat-system parameters enter our repository or any cloud service. Fixtures and demos use fictional or public airframes. Threat models ship as generic, parameterised capability classes; customers load their own data only inside their own enclave deployment.