null-results

An evaluation methodology for context-augmentation experiments: a format-matched control that separates whether added context helps because of its information or its format. Pre-registered instrument, mutation-audited scorer, worked example in SVG reference resolution.

A paper-trading app and the earnings research behind it, in which twelve promising results died to their own controls and none survived.

A measured null result: 11,953 live crypto alerts, no directional edge, and the four ways I nearly fooled myself.

MSc thesis: 13 classifiers all performed at chance predicting in-game spending on public data, while a control on genuine game-level data hit 0.95 ROC AUC. Revenue Gini 0.87.

A backtesting harness built to refuse false results — and the closed study that used it. 24 pre-registered hypotheses, three sealed verdicts, no tradeable edge found, and no funded-account fee ever paid. Controls run in ten seconds with no data, key or network.

An honest empirical test of subjective/objective reconciliation for forecasting democratic backsliding using ensemble learning.

Finite-field audit notes, reproducible tests, and LaTeX preprint for the e-5-137 parameterization.

Archival repository for the Zenodo paper The Asymmetric Burden of Proof.