Pre-registration, Version 7: Inlet Subcooling and Pressure Drop in the Water CHF Blend, with a Bound by Pressure Domain, Blind-Tested on the Version 6 Articles

Pre-registration of a blind test, frozen on 12 October 2026 before any CHF value of a test article was seen. Version 6 of this programme (doi:10.5281/zenodo.23288012) blends the 2006 look-up table, a π table and a corrected table to predict critical heat flux (CHF) in water tubes; its Addendum A1 (doi:10.5281/zenodo.23289282) adds a conservative bound. Its test articles, Previti et al. (1966, SORIN Saluggia) and Würtz (1978, Risø), have not been read. Afterwards, six development rounds on 23,867 NRC rows, each predicted with its laboratory left out, looked for what version 6 is missing, each under a rule written before it ran. Two dimensionless groups helped: the inlet subcooling number and a homogeneous estimate of the tube’s pressure drop over its pressure. With both in every local fit, the relative RMS error fell from 8.49% to 6.41% and the median absolute error from 3.37% to 2.97%. No round reduced the margin a 1% safety bound needs by the point its rule required; it stayed between 19.35% and 20.23%, set mostly by a few laboratories that run high. By the stated rules no candidate qualified; version 7 and its bound by pressure domain were chosen by the author after the results were seen, and the record says so. Every failed round is reported. Version 7 will be scored once on the version 6 test articles, after version 6 and Addendum A1, on the same transcription and point sets. H7 and H7b: version 7 beats the 2006 table on relative RMS and median error. H7c: version 7 beats version 6 on RMS. H7d: its bound (0.7853 at 1 MPa and below, 0.8397 above, to 20 MPa) lies above the measurement on at most 1% of points. H7e: version 7 has a smaller share of predictions more than 10% high than the table. H7f: its bound needs less margin than version 6 given a bound by the same rule and split, with no more points above. Each is reported on its own, with no overall verdict. The record states in advance that H7e is more likely to fail than pass, and that version 7 overpredicts large tubes at 8 to 14 MPa more often than version 6 on development data, which matters for Previti’s 17.1 mm section. An independent AI audit found no critical problem; its eight major findings were fixed before freezing, and the author chose the comparator of H7f. Archive SHA-256: c663aa414d3e69e139d33f51432c40011321a49a6aa3cafd3456a3e442fcb93b