The Null Institute / Publications
Verification Pilots
Focused, independently checkable studies of public claims, published research, datasets and alternative explanations. Each report describes what was tested, what the evidence supports, and what remains unresolved. A pilot is not a claim of peer review.
13 publications
Dunning–Kruger Reanalysis: The Scale, the Baseline, and the Error Curve
Independent 151-person reanalysis confirms the skill–self-assessment error gradient. The sample-wide downward offset, unusual response scale, baseline-to-test change, and rank reference limit claims about the mechanism.
Small SBA Loans Fell 35%. Outside Five Lenders, Their Share Rose
We independently analysed 403,644 SBA 7(a) loan records. Approvals under $50,000 fell 34.7% in FY2026, but five lender names account for 88.2% of the net decline; the small-loan share actually rose among other lenders.
AI Tutors and the Human Work Behind 94,385 Checks
An independent audit of Google and Eedi's 1,179-pupil trial: immediate maths gains, no detected Star Maths advantage, and 94,385 mandatory human checks. The trial cannot yet establish staffing savings.
Friday the 13th: We Rechecked the Study Behind the Accident Scare
We reanalysed all 61 observations behind a famous Friday-the-13th accident study. Hospital admissions rose in six matched date pairs, but the statistical conclusion changes with the test. Motorway traffic was lower, while shopping was not.
A Sixfold Birthday-Week Disease Signal Under Audit: Statistical Inconsistencies and Automatically Generated Birth Dates
An independent audit of a reported sixfold HFMD birthday-week effect finds a nonreproducible chi-square statistic, regression-model inconsistencies, and historical evidence that China's disease-reporting system generated birth dates from ages. The mechanism's role in Yunnan remains unproven.
Can We Rank Countries by Smiling? An Audit of 125 Geographies and 24 Source Families
A secondary synthesis of 24 smiling-data source families across 125 geographies. We audit cross-study concordance, model weighting and the limits of claims about happiness.
Inconsistent Denominators and Image-Coordinate Frames in a 28-Country Visual Communication Dataset
An independent, source-verifiable reanalysis of 79,500 political images finds a numerator–denominator mismatch in published country percentages and incompatible facial coordinates and image frames. Includes reproduced logistic models, methodological limitations and open code.
A Likely Double Count in a Published Water–Mortality Study
An independent audit of NHANES 1999–2014 finds that the water-source distributions reported by Zhou et al. (2022) closely match a reconstruction with likely double counting of drinking water after 2005. The mortality implications remain unresolved.
How Often Does One Blood Pressure Reading Change the Category? NHANES 2021–2023
In a survey-weighted NHANES 2021–2023 reanalysis, 16.4% of adults changed US blood-pressure category between the first reading and the average of three. These are within-visit category changes, not diagnostic errors.
What Does the Beverage Hydration Index Actually Measure?
The Beverage Hydration Index is reproducible. But does less urine after a drink really mean better hydration? A construct-validity, outcome, and claim-drift audit.
Does Killing Oral Bacteria Actually Make the Mouth Healthier?
Chlorhexidine kills oral bacteria and reduces plaque. But does suppressing the microbiome actually improve health? A falsification-oriented evidence review.
Repeatability Is Not Transportability: What Makes a Standardized Glucose Challenge Informative?
A harmonized reanalysis of CGMacros and Stanford/Hall suggests that a standardized glucose challenge can be repeatable without being useful for cross-meal personalization. Food & You further weakens the idea that larger glucose excursions are automatically more informative.