Validation status / Axiom Cortex
A documented method is meaningful. A working paper is useful. A synthetic study can test model behavior. None of those, by itself, proves predictive validity, fairness, or a hiring outcome.
We use science to align tomorrow’s IT talent with the work ahead, and we show the boundary around that science.
Current public boundary
The public materials document 44 governed methods across six mathematical families, their purpose, required inputs, limits, and human-review boundary. A listed method is not proof that it ran in a specific evaluation.
The 2025 report documents the evaluation framework. The 2026 paper documents a human-task-agent alignment model. SSRN publication makes the work inspectable; it does not make the papers peer-reviewed validation.
The six-domain alignment study tests its model on synthetic profiles. That can test equations, sensitivity, and coefficient recovery under stated assumptions. It does not establish future job performance.
TeamStation AI reports related methods and processes used across 30+ US companies, empirical source material from 13,000+ technical interviews, and more than eight years of research history. These figures are company-reported.
The public sources do not establish predictive validity, quantified accuracy, fairness across relevant populations, comparative superiority, or customer hiring outcomes for the current product version.
Read the label first
Tests can verify that a version follows its stated software contract. They do not prove a scientific or employment outcome.
A working paper can explain the model, equations, assumptions, and limits. Publication alone does not prove external validity.
A defined population, frozen version, qualified study design, held-out cases, and measured outcomes are needed for predictive or fairness claims.
Repeatable calculation, useful product output, and validated hiring outcome are related questions. They are not the same claim.
Claims that need more evidence
The current public record does not establish:
Those claims need their own version, sample, method, comparison, date, limitations, and qualified review.
For buyers and reviewers
Request the exact evidence for the claim you care about. Confirm the product version, population, interview design, comparison baseline, error handling, reviewer role, and limits. If that evidence is unavailable, treat the claim as unknown.
