Gordon OlsonAction AI. Real world systems. Trustworthy AI.

Live

AI Psychometrics Lab

Live research platform at aipsychometricslab.com for structured AI model assessment using validated psychometric inventories, Big Five (IPIP NEO 120), MBTI (OEJTS 1.2), DISC, Dark Triad, run against any OpenRouter model through a Stateless Independent Context Window Approach (SICWA).

Live

Research platform / Model evaluation and psychometrics

AI Psychometrics Lab cover

What Gordon built and maintains

Built and shipped aipsychometricslab.com as a live public platform with model profiling, a Model Explorer, Recent Runs log, and a Publications section.

Designed the SICWA (Stateless Independent Context Window Approach) methodology to eliminate context carry over and isolate genuine model behavioral tendencies.

Integrated validated psychometric inventories, Big Five (IPIP NEO 120), MBTI (OEJTS 1.2), DISC, Dark Triad, against any OpenRouter model ID.

What was hard

Psychometrics applied to AI is genuinely novel and requires adapting frameworks designed for human subjects to stateless model evaluation.

SICWA had to be designed carefully to eliminate context contamination without losing the ability to capture meaningful cross item behavioral variance.

The line between research rigor and product usefulness is easy to lose when building evaluation systems, keeping both required real product discipline.

Featured research

Synthetic Personality Signatures of Large Language Models Under Stateless Repeated Administration

AI Psychometrics Lab, Lab Report / April 14, 2026 (revised April 18, 2026)

An exploratory evaluation of contemporary large language models using validated psychometric inventories administered under the Stateless Independent Context Window Approach (SICWA). Repeated stateless runs surface a shared baseline personality profile across modern chat models, while still preserving interpretable, family specific behavioral differences.

Key findings

  • Across repeated stateless runs, baseline chat models converge on a shared high Conscientiousness, high Agreeableness profile.
  • Interpretable family specific differences remain visible in assertiveness, response variability, and emotional tone.
  • SICWA isolates genuine model tendencies from context carry over, producing repeatable signatures that standard benchmarks do not surface.
  • Suggests psychometric profiling is a practical layer in model selection and AI product evaluation, not just an academic exercise.
Read the full lab report →

Proof at a glance

Live

Public platform

aipsychometricslab.com is live with Model Explorer, Recent Runs, and Publications.

View evidence →

SICWA

Novel methodology

Stateless Independent Context Window Approach isolates genuine model behavioral tendencies.

4 inventories

Validated instruments

Big Five (IPIP NEO 120), MBTI (OEJTS 1.2), DISC Assessment, Dark Triad.

  • Live public platform at aipsychometricslab.com with a Model Explorer, Recent Runs, and Publications.
  • SICWA methodology isolates model behavioral tendencies in a way that standard benchmarks cannot.
  • Connects structured psychometric measurement to product decisions in a way that most AI teams treat as an afterthought.

Sources and verification

Where the public evidence lives.

Project record reviewed

Live product / AI Psychometrics Lab

AI Psychometrics Lab

The live model explorer, supported inventories, SICWA label, and public run interface.

Link verified

Open evidence →

Research publication / AI Psychometrics Lab

Synthetic Personality Signatures of Large Language Models Under Stateless Repeated Administration

The published SICWA methodology, reported model measurements, authorship, and publication date.

Link verified

Open evidence →