MindMetrics

• Procedural generation with solver-verified unique answers — unlimited retakes without item memorization.
• Percentiles only against named published norms; everything else explicitly within-person.
• Practice effects modelled, plus Reliable Change Index badges instead of raw score deltas.
• Forecasts gated at n≥5, horizon-capped, with widening confidence bands.
• Per-instrument licensing stated up front; no proprietary instrument used.

• 

  Ten instruments, each with its licence

  Every card carries what the test is, what it is not, and who owns it: CC BY-NC-SA for OEJTS, public domain for the IPIP inventories, original procedural content for the matrices. The caveats are on the card, not buried — the ICAR sample is called a spot-check rather than an IQ estimate, and the IPIP percentiles are labelled an internet convenience sample.

• 

  Cadence advice that never blocks

  The right column says, per instrument, whether retaking now would mean anything — “outside the advisory window, ready when you are” against “not taken yet”. It advises and never refuses, which is the point: procedurally generated tests can be retaken freely, and the static questionnaires get guidance instead of a lockout.

Psychometric tests are built to be taken once. Retake one and you are measuring memory of the items, not the trait — so the consumer apps either block retakes or quietly sell you your own practice effect as growth. What I wanted was the opposite: test whenever I like, as often as I like, and get numbers that stay honest under repetition.

Ten instruments on a plugin engine behind one accessible test player. The cognitive tests are procedurally generated from seeded RNG — matrices from a Carpenter-taxonomy rule engine with solver-verified unique answers, plus series, digit span, reaction time and fluency — so no item ever repeats while parallel forms hold difficulty composition constant. Norms appear only where real published ones exist, always with the population named; generated tests are labelled within-person. Practice is modelled rather than ignored, via a log(sessionIndex) term, a practice-adjusted trend view, and Reliable Change Index badges that separate real movement from measurement noise. Forecasts unlock only at five sessions, are horizon-capped, and are drawn dashed with widening 95% bands.

Code-complete beta on Next.js 16 with Drizzle over SQLite, multi-user data model ready for Postgres. Proprietary instruments (MBTI®, CliftonStrengths®, WAIS) are deliberately absent — open equivalents are substituted and every instrument carries its licence in writing. Next up is psychometric hardening: simulated-respondent checks and IRT calibration of the matrix difficulty model from accumulated responses.