Pre-registration, Version 3: A Near-Critical Dimensionless Critical Heat Flux Model Against the 2006 Look-Up Table

Two pre-registered tests of a compact Buckingham π model for critical heat flux (CHF) have failed against the 2006 look-up table. The second failed above 20 MPa, beyond the pressure range of its training data. This third version adds the reduced distance from the critical pressure, 1 − P/Pc, to both terms of the model and is fitted to minimize the error the test scores: the critical heat flux of a tube with fixed inlet conditions. Trained on 2,553 high-pressure rows of the NRC database and the 24 near-critical Oettig et al. series, it has eight parameters. When the Oettig series are held out and the model is fitted on NRC data at or below 20.0 MPa alone, it predicts them with 7.78% relative RMS error, against 10.94% for the table. Those series shaped the choice of the new terms, so this is not a blind result. The blind test uses two articles (Chen et al. 2018, ASME JNERS; Pan et al. 2016, IJHMT) that had not been received when this was frozen. H3 (primary): the model beats the table on saturated CHF at 14.0 to 21.0 MPa. H3b: it comes within 2.0 percentage points of it. Every change made after the first freeze is recorded in hashed addenda, including a superseded model and the in-sample results that prompted each change. The result will be published as a new version of this record whichever way it falls. Archive SHA-256: f9d9cfa623b0779d936430a02cbb270ead2b2f8d476bac7699e873e879841333