PSY-67: Ultra‑High‑Rigor Protocol for Photograph‑Based Diagnosis

Protocol PSY-67 implements the unified metrological framework of Protocol 9.0 to test the claimed phenomenon of photograph‑based diagnosis – the alleged ability of a person (the "reader" or "diagnostician") to perceive information about an individual's health, emotional state, or personal characteristics (e.g., current medical conditions, past trauma, psychological traits) solely from a photograph, without any prior knowledge of the person. The primary observable is the accuracy of the reader's diagnostic statements (or the correspondence between the reader's description and the ground truth medical/personal data) as rated by independent blind judges, compared to a baseline (chance) level. The experimental design includes standardised photographic protocols (same lighting, clothing, neutral expression), a pool of target individuals with known medical/psychological profiles (from questionnaires, medical records, or standardised assessments), multiple blind readers (with varying experience), triple blinding (reader, experimenter, and analyst are blind to the target's identity and condition), trap conditions (a photograph of a healthy individual, but the reader is told to "diagnose" a specific condition), sham conditions (a photograph of a non‑human object, or an AI‑generated face, to control for pareidolia), catch conditions (a known, easily detectable condition, e.g., a visible physical characteristic, to validate the reader's sensitivity), positive controls (known physical characteristics that can be inferred from a photo, e.g., age, sex, race – to verify that the reader can extract some information), comprehensive environmental monitoring, and multiple independent judges to evaluate the accuracy of the diagnoses. The protocol mandates a pilot phase to validate the photographic protocol, the diagnostic task, and the judging system. The primary analysis uses a Linear Mixed Model (LMM) for the rating difference (diagnostic accuracy) and a signal detection analysis (d′) for binary outcomes, with α = 0.025 (one‑tailed) and power ≥ 0.90 for a moderate effect (Cohen's d ≥ 0.15). The sample size is N = 100 target individuals (baseline, with adaptive increase to N = 150 if pilot shows high variability). Statistical analyses are preregistered, include Bayesian inference (BF₁₀), permutation tests, and multiverse robustness checks (≥70% of 12 forks). The aim is to detect a reproducible photograph‑based diagnostic ability or to establish an upper bound on the effect size with unprecedented rigor.