AACSB accreditation
Every score traces to an explicit rule — auditable, mappable to learning outcomes, and reportable as Assurance-of-Learning data.
The evidence behind vLeader
vLeader runs on the KRY engine — a symbolic-AI system of 53 hand-coded causal rules, built in 2002 and deliberately never replaced by a language model. This page sets out why that choice matters, what our own outcome data shows, the sample sizes behind every figure, and the method used to produce it.
The KRY engine
Symbolic AI, not a language model
The simulation is driven by ~53 hand-coded microstrategies organized into five categories — Power, Ideas, Tension, Work, and Relationship. Each is a small, interpretable causal rule. NPC behavior is selected by a transparent utility calculation over those rules, not generated by a model.
Its lineage is classic interactive-AI and agent-based modeling: Mateas & Stern's Façade (2003), Schelling's game theory, Epstein–Axtell agent-based modeling, and the behavior-tree / utility-AI tradition. It is GOFAI — “good old-fashioned AI” — running in production in 2026.
Because there is no language model, the system is deterministic, replayable, debuggable, and hallucination-free. The same choice always produces the same consequence — for every student, in every cohort, every semester.
// the differentiator, in one line
Deterministic causal rules.
0 large language models.
0 hallucinations.
// same input → same output, always
Why determinism matters in 2026
Reproducible, auditable outcomes every time
Government & defense
No external API calls, no probabilistic drift, no hallucination risk — the same stimulus, every cohort. Defense Acquisition University is a current adopter.
Healthcare & legal
Reproducible scenarios that behave identically across runs — the property regulated-domain training and review actually require.
Longitudinal research
Compare cohorts across years under identical conditions — the engine is the controlled variable, not a confound.
Outcome data
What our own data shows
Reflection scores, summer 2026
1,162 graded reflections · 156 students · 15 June – 28 August 2026
Students complete a graded reflection before and after the debrief. Across 1,162 paired reflections, 62% scored higher afterwards, and improvement outnumbered decline better than two to one — 61.9% up against 28.1% down. At a three-point threshold on the ten-point rubric the ratio holds: 27.6% against 13.3%.
The obvious objection is regression to the mean, so here is the test that rules it out. Among students who had already scored 9 out of 10 before the debrief — with almost no headroom left — 63% still improved (n = 383). A pure regression artifact would push that group down, not up.
At student level: 58% show net improvement across their reflections, and about one in eight averages nearly four points of gain. It does not land the same way for everyone, and we would rather publish that than an average that hides it.
Single-group before-and-after with no control group: this shows scores move, not that vLeader caused the movement against an alternative. The rubric is graded by a model, not by human raters, and pre/post scores correlate at r = 0.44. One term, 156 students — 8.5% of students in the period, since only sections whose instructor uses the reflection loop appear.
Where the simulation score falls short
We looked for a replay effect and did not find one. Students replay heavily, but a student's first and last score on the same scenario correlate at r = 0.012 — the simulation score is too noisy to measure learning, so any “students improve on replay” figure drawn from it, including ones favorable to us, would be arithmetic on noise. We examined 27,499 rounds of engine state across 580 games and found that how a student plays does not predict whether their reflection improves.
The practical reading: the debrief, not the simulation score, is where the learning shows up — which is why the reflection loop is what we measure and publish, and why we quote no number we cannot stand behind.
Gurley & Wilson (2011)
Journal of Instructional Pedagogies, vol. 5 · Fayetteville State University (an HBCU)
“Developing leadership skills in a virtual simulation: coaching the affiliative style leader.” A study of the simulation in an MBA class comparing students with an affiliative leadership style against other styles, reporting that affiliative-style students scored lower on developing and using power, and that scores improved after repeated attempts at the first scenario.
This is an observational comparison between style groups, not a controlled trial — there is no control group and no randomization. It is a single study, from 2011, in a practitioner journal without an impact factor. We cite it because it exists and is real, not because it settles the question.
Goosen & Van Tonder (c. 2010)
Gale Academic OneFile
“Leadership development in an electronic frontier: connecting theory to experiential software through supplemental materials.” One of several scholarly treatments of the simulation's use as experiential courseware.
Track record
24 years, deliberately built
-
2002
The KRY symbolic-AI engine is built; Virtual Leader ships — years before the LLM era.
-
2004
Training & Development Journal names it Best Online Training Product of the Year.
-
2011
Gurley & Wilson publish an MBA classroom study at Fayetteville State University.
-
2024
Rebranded as vLeader — same deterministic engine, refined across hundreds of thousands of decisions.
Theoretical foundations
The frameworks your syllabus already cites
Goleman's leadership styles
Directive, visionary, affiliative, democratic, pacesetting, coaching.
Tuckman's team development
Forming, storming, norming, performing — enacted as the meeting unfolds.
Thomas-Kilmann conflict modes
Competing, collaborating, compromising, avoiding, accommodating.
Kolb's experiential cycle
Experience → reflect → conceptualize → experiment — the product's own loop.