Methodology
Honest, thoughtful, and as rigorous as a self-knowledge tool can be. Made in Melbourne 🇦🇺.
Our assessments are inspired by and structured around validated psychological frameworks. They are not clinical diagnostic tools and do not replace professional psychological evaluation.
The 12 modules and their frameworks
Internal consistency
Our question banks are designed to achieve internal consistency reliability (Cronbach’s α ≥ 0.70) across all dimension subscales. Reverse-scored items are flagged explicitly and inverted before aggregation. A validation script at packages/shared/scripts/validate-consistency.js runs against live response data and flags any subscale below threshold for review.
Question formats
Twelve distinct formats: visual matrix grids (Raven’s style), drag-and-drop sequences, image-based emotion recognition, timed reaction rounds, memory recall, scenario illustration cards, colour-and-shape patterns, word association, spatial rotation, Likert sliders, priority ranking card sorts, and yes/no binary taps. Each module uses at least four formats. Difficulty adapts in real time: three consecutive fast-correct answers step the tier up; two consecutive errors step it down. A one-line micro-insight appears under the progress bar after every answer explaining what the format measures \u2014 not whether you got it right.
How scores are derived
Each module is scored deterministically: same answers, same score. Right/wrong items use a binary correct map; preference/Likert items use weighted options (0–3 by default, reverse-coded where appropriate). Composite scores are explicitly named: IQ Proxy maps to a 75–145 scale (Cognitive Power Index); other modules report 0–100 with a named band.
How percentile is calculated
Each module starts with a seeded benchmark distribution (mean and standard deviation chosen to match published norms where applicable). As real test takers accumulate, percentiles become increasingly empirical. We compute both a global percentile and a cohort percentile (age band × role band) so you’re compared against peers, not the universe.
The Combined Cognitive AI Index
Your Mind Map is the average of normalised module scores. The composite is shown as a 0–1000 index when at least 2 modules are completed. The visual radar shows raw normalised scores per dimension, not the average — that’s the whole point.
How the AI Coach uses your data
Every Coach message is sent server-side to Claude Sonnet 4.5 via Anthropic’s API (through Emergent’s gateway). The system prompt includes only the data you have produced on this product: your scores, your deltas, your badges, your prior chats. We don’t share with third parties or train external models on it. If Claude is unavailable the LocalFallbackService delivers the same JSON shape from curated templates.
Daily training adaptive difficulty
Daily training tracks accuracy per dimension. After 3 sessions in a row at >75% accuracy on a dimension, difficulty steps up. After 3 sessions at <40%, difficulty steps down. The streak counter tracks consecutive days, not consecutive correct answers.
Data privacy
We store: email, intake answers, module answers, derived scores, Coach chats, badges, streaks, and unlock events. We anonymise leaderboard and cohort aggregates. You can request full export or permanent deletion at any time — see Privacy.
What this is not
Not a clinical assessment. Not a diagnosis. Not a hiring tool by itself. Not a guarantee of anything. It is a cognitive self-knowledge instrument for the AI era — designed to be sharp, useful, and re-takable.