Lux is a classical test theory, fixed-form assessment in its current phase. That is a deliberate boundary. Lux-specific adaptive calibration is gated until at least 500 eligible Lux responses exist. That required dataset is not yet available. The threshold applies to the adaptive-calibration stage; it is not a promise that each cohort will be calibrated once it reaches 500.
What the gate actually protects against
Calibration work depends on having enough eligible Lux data to estimate item behaviour. FND has set 500 eligible Lux responses as the minimum gate before that work begins. Crossing the gate would make a dataset eligible for calibration analysis; it would not, by itself, establish validity, fairness, population norms or fitness for use.
Until the gate is met, Lux remains fixed-form. Any future calibration study would still need a defined population, analysis plan, quality checks and conclusions limited to the evidence it produces.
Producing output isn’t the bar. Producing output that means what we say it means is the bar.
What happens before the gate is reached
The current Lux codebase contains a fixed-form, 19-layer scoring engine with versioned fields, committed fixtures and regression tests. The inherited internal-consistency figures of α .800 to .898 come from complete ten-item Big Five banks in a 603,322-response source/reference dataset. Those responses are not Lux-participant data, and the figures do not establish Lux-wide validity or unconditional production readiness.
Lux’s active evidence-development programme combines technical testing now with staged empirical evaluation as suitable data becomes available. Test–retest work will assess score stability as repeat-response data becomes available. Other pre-defined studies will examine Lux-specific validity, norms, intended-population fairness and clinical utility.
The 500-response gate is a prerequisite for one adaptive-calibration stage, not evidence that calibration or Lux-specific validation is complete. Eligible data makes a study possible; participant volume alone does not establish reliability, validity, norms, fairness or utility.