A reproducible audit of how the field forecasts AI and data-centre electricity demand, scoring 11 published forecasts on transparency, internal consistency, revision behaviour, and comparability. It scores forecaster behaviour; it is not itself a forecast, and not a bubble call. Of the 11 forecasts, only 3 are reproducible from public method and data (7 are confirmable at their primary source, and 10 are verified against some credible source), while the single highest figure (BCG, about 1,050 TWh) cannot be verified at all because it is paywalled. The forecasts are expressed in six incompatible formats and disagree by multiples at the US 2030 horizon without sharing a common scope definition; among the revisions that could be documented, all have moved upward. A companion reproduce.py computes the four signals from the shipped dataset, so the headline figures are derived rather than asserted. Version 0.2.1 applies two precision corrections from an external adversarial review (an EPRI revision figure reconciled to the dataset, and the Goldman revision restated as a change in the demand level rather than in percentage points); no finding changed. Independent analysis, not investment advice. Produced with assistance from Claude Opus 4.8, an Anthropic model.