How the measurement is constructed
The structure is what earns the trust, not the language.
Every hypothesis, exclusion rule, and statistical test is registered publicly (Open Science Framework) before the data-collection window opens — including the exact criteria for declaring a result supportive, inconclusive, or inconsistent with the hypothesis.
Participants are blind to condition; administrators do not see running aggregates; analysts receive data stripped of condition labels until a pre-specified unblinding date.
Objective instruments, not just self-report: before-and-after voice-biomarker readings (physiological stress markers from a voice sample) and on-device response measures. No clinical or diagnostic claims — these are research measures of change, published openly. See the modalities →
Results are reported with effect sizes, 95% confidence intervals, achieved power, and pre-registered correction for multiple comparisons. Pre-registration plus frozen analysis code structurally prevents fishing for a result.
At each milestone the de-identified dataset and the exact analysis code are released under an open licence. You do not have to trust us — you, or any ' 'statistician, can re-run the numbers.
Protocols are reviewed by independent methodologists whose role is to find the flaws. Their reports are published. No reviewer is compensated on outcomes.
