The Department is building a four-star Autonomous Warfare Command to buy and fight autonomous systems at scale. The primes will build the drones. Aztech builds what the command still needs: proof of which systems work, what each defeat costs, and a record of who decided what, on what data.
Project Agincourt's stated goal is a repeatable procurement system whose output is “not a particular drone, but the ability to adapt in combat, on the right side of the cost exchange.” That takes measured results, a cost-per-kill view and an auditable decision trail — software problems, not airframes.
Built and tested in October 2026, each one a bolt-on to the primes’ autonomy stacks rather than a second operating picture.
Government test events, field use, exercises, simulations and vendor claims in one scorecard. Every result is graded by source and sample size, systems ranked on vendor claims alone are flagged, and plain-English questions are answered in the page with the reading shown.
Cost per kill and exchange ratio against the threat’s own cost, with attritable-only and exquisite-only comparisons, played out in a three-wave one-way-attack swarm whose red cell adapts. One click sends the results to DRAGONFLY.
Every engagement decision becomes a dated record: the named decider, the authority they acted under, the data relied on and how old it was, and contrary reporting kept beside it — hash-chained, so an edited record fails verification.
Hostile and suspect detections packaged as structured records — technical data and source grade from OCELOT, hazard, mitigation and outcome from the analyst — for counter-drone data repositories. Word, PDF and JSON.
A draft spectrum request package built from RAVEN’s deconfliction engine: planned emissions, every conflict with friendly assignments and protected receivers, and a proposed mitigation for each.
Enter or click a route and each leg is scored against live GPS-interference areas and documented hotspots, scaled by the platform’s navigation dependency. Word, PDF and GeoJSON for planning tools.
The frequency fight in one picture: friendly coverage, jamming footprints, emitter bearings and reported interference on a map with a plan-time slider, a victim-by-interferer matrix, side-by-side COA comparison, EA control from planned to ceased, and an SFAF-style draft for the spectrum manager.
Items from different platforms linked by place and time, a shared identifier, a common frequency or an explicit reference, with the reason for every link; MONTEZUMA’s signed decisions re-verified across the portfolio; plain-English questions answered with the item behind every claim.
Every drone vendor scored on delivery against award and schedule, component origin against the covered-country restrictions, and the shared sub-suppliers that could stop several vendors at once — with a projected completion date from the actual run-rate.
UUV, USV, seabed-crawler and anchor-drag threats run against a chosen cable segment and the protection laid around it: chance of detection, chance of detection in time to respond, and the stretches of cable each threat could reach unseen.
One click on any platform reads what it can prove about itself — deployed-file hash, sign-in, every outside connection, browser storage, headers and third-party scripts — maps it to NIST SP 800-53 and 800-171 controls and writes the POA&M.
One form opens a full case: the drone and its Remote ID, the operator’s estimated launch point, the flight path as a timeline and a hashed incident report. OCELOT threat records import directly.
Six courses — DRAGONFLY, MONTEZUMA, OCELOT, RAVEN and GOLD operator training — for the new autonomous-warfare career fields, ending in a path certificate.
An Operator course and a Technical course for every platform, with hands-on drag exercises and labs in the live platform, built from the real code.
Per-learner accounts, certificates with an ID, hours and skills that anyone can verify online, a training record in Word or PDF, and a roster for training managers.
Scenario prices and quantities are notional inputs. Capabilities are described as built; none is represented as fielded or government-accredited.