The Simulator Is the Easy Part: Flow, Unit Readiness and the Human Baseline for Autonomous Decision Tools

11 May 2027
Human Performance Theatre
The Human advantage and cognitive edge

Anyone can build a simulator today. Game engines, synthetic terrain and AI tooling are available to any competent team. What remains hard, and where much synthetic training quietly fails, is making a complete unit rehearse as if it matters: realistic enough to be trusted, structured enough to learn from, and paced so the unit stays engaged from first contact to final debrief.
This paper is built on wartime iteration. For fifteen months, a 24-station mobile classroom ran with IDF Central Command as design partner, iterating scenarios and session structure throughout. Across IDF, Israeli Police and community emergency-response programmes, we have trained more than 4,200 participants from 210 units and organisations, producing over 12,600 recorded person-hours. At ITEC 2026 we presented failure-driven rehearsal. Here we examine why it works, and what it produces.
Our design rests on two conditions. First, flow: challenge matched to capability. A unit exits the channel upward through overload (flooded nets, missing acknowledgements, parallel information-pulling, traffic concentrating on one commander) or downward through under-challenge. Second, belief. Disbelief shows as shortcuts the unit would never take for real. Our debriefs record overload and disbelief: a commander who lost track of his force as the net flooded, and trainees entering buildings alone, which the instructor put down to "it's the computer".
This is our central argument. Flow and belief are not only teaching conditions; they are what make behaviour data valid. Data from an overwhelmed, under-challenged or unconvinced unit describes the simulator, not the unit. Training design sets the ceiling on data quality, and no downstream analytics can recover what poor session design failed to capture.
It also changes how progress should be read. A volunteer emergency-response team completed four sessions over eleven weeks, with difficulty raised each time. Across sessions two to four, its average trainer-rated command score went 85, 77, 85. Read naively, that is no progress. Read against difficulty, the team restored its standard in a far harder problem. Meanwhile, a command bottleneck around its senior commander remained unresolved. Readiness measures that ignore difficulty will misread adaptation as stagnation.
Radio traffic is transcribed and each transmission automatically classified by speech act, such as command, report or acknowledgement, alongside movement and drone-employment logs. We propose this as a human performance baseline against which autonomous and decision-support tools can be tested. Whether it can fill that role is an open question; a first sample is now with an external evaluator.
Limits are stated plainly. Capture across all streams is not yet reliable. Decisions are not yet linked to outcomes. The scores above are indicative, and transfer to operations is unproven. Consent for secondary use is being formalised, and training data must never become punitive, or units hide mistakes and the data stops being real.
Attendees will leave knowing what makes unit rehearsal work, how to see when it stops working, and what human behaviour data can and cannot yet support.

Chairperson
Paddy Little
Paddy Little, Founder and Executive Director - Cervus Defence and Security Ltd
Speakers
Or Ben Shabat
Or Ben Shabat, Founder & CEO - DCA - Digital Combat Academy