Frequently asked questions

Is any of the data real?#

No. Every receiver, camera, image, telemetry stream and score in the prototype and on this site is synthetic. Nothing is derived from operational data or real aircraft geometry. Validation on Government-provided KC-46 video is the next phase.

Does it need a GPU or the internet?#

Neither. It runs on CPUs and makes no outbound connections. A 4-core box processes a 90-second engagement in about ten minutes.

Does it connect to the aircraft?#

No. AARCV works on recordings after the flight. The edge unit is a ground box the recording is plugged into after landing; it has no connection to aircraft systems. Everything it produces is advisory debrief material.

How accurate is it?#

On 13 held-out synthetic sorties, 95% of category scores fell within 10 points of the score computed from simulator truth and all 12 grades matched. Inside 15 m, median position error is 0.21 m. See Evidence for the method and the misses.

Can instructors disagree with a score?#

Yes, and they should when it is wrong. Overrides are recorded, counted in the agreement statistic and used to improve the rubric and models. See instructor review.

Which receiver aircraft does it support?#

Whatever is on the allow-list. Each receiver type needs a 3D template and a qualification run before it is added; until then its sorties are refused and purged.

Who decides the classification markings?#

Your security classification guide, loaded as policy by a security officer. The prototype ships with notional placeholders only.

What about night and infrared?#

The prototype is daytime visible imagery. Night, infrared and stereo (RVS 2.0) evaluation is planned research work with our university partner.

Can we integrate it with our systems?#

Yes. Everything in the web app is available through a documented REST API; training records export as xAPI and CSV; release packages verify offline.