Pre-registration, Version 8: Does the Tube Wall, the Heating Method, the Detection Method or the Coolant Change Water CHF? A Blind Test in Matched Same-Rig Comparisons

Version 1.3 is a correction of record. Reference [15], cited for the wall properties at 300 K in Table 1 and materials_v8.csv, named the 6th edition of Incropera et al., Fundamentals of Heat and Mass Transfer (Wiley, 2007), which was never checked: the programme held no copy of any edition when the values were entered. The first copy it obtained, the 7th edition (T.L. Bergman, A.S. Lavine, F.P. Incropera, D.P. DeWitt, Wiley, 2011, Appendix A, Table A.1), received through interlibrary loan on 19 October 2026, matches the density, specific heat and conductivity of all nine rows exactly, and it is the corrected reference. Whether the 6th edition prints the same values was not checked. The version 1.2 archive is carried byte for byte and unopened, with CORRECTION_1.txt, its PDF and a script that repeats, by code, the comparison with the analyst's transcription of the 7th edition (48 checks, 0 discrepancies). Nothing in the protocol, the hypotheses, the scoring code, the test pool or any value changes, and the test has not been scored. An independent review of the programme's published records, run by a separate AI agent, found the point.

Version 1.2 adds Addendum A2, written on 15 October 2026 before any test source was opened and before any scoring. An independent review of the programme found four defects in the frozen scoring code and its description, and the addendum corrects them. Scoring will use evaluate_v8r.py, which differs from the frozen evaluate_v8.py only in lines marked A2; the frozen file stays in the record. (1) In the two-sided sign tests a rig mean of exactly zero counted as negative, so six rigs showing no difference at all would have passed (p 0.0312); a zero now counts against both directions, as the protocol requires. (2) The 2006-table correction was applied only when a source's text read exactly "vertical" and "up"; it now also accepts wordings such as "upflow" and "upward", and every pair given no correction is logged. (3) H8g-2 can pass with no heavy-water difference when a rig states a large uncertainty (above about 7 to 9%), so the reading that latent heat accounts for the difference is claimed only if H8g-1 also passes; this rule is the author's. (4) The protocol said the self-test finds the built-in effects, but H8a-M missed a built-in +0.08 material effect (p 0.0625), so a failed test will be reported as not finding an effect, never as showing that none exists. No hypothesis, tolerance, threshold or test source changes, and every version 1.1 file is carried over byte-identical. The frozen and amended self-tests give the same verdicts and p values, and the consistency pass ran 36 checks with 0 discrepancies.

Version 1.1 adds Addendum A1, written on 12 October 2026 after version 1.0 was published and before any CHF value of any test source was seen. Nine test sources had arrived by then; each was forwarded unopened and filed unopened with its SHA-256, and the classification agent had not run. Nothing in the protocol, extraction instructions, scoring code or test pool changes, every version 1.0 file is carried over byte-identical, and no verdict changes.

The addendum adds a descriptive split of the scored pairs by boiling-crisis regime. Two development sources read after freezing disagree on the direction of the wall-thickness effect: Hewitt et al. (1965) found a thicker stainless wall burning out 6 to 9% higher in film dryout at exit qualities of 0.60 to 0.78, while Lee et al. (2020) found thicker 316 stainless lower at exit qualities near zero. A wall takes part in a boiling crisis only as deep as heat diffuses while the crisis develops, which reaches the wall thickness in film dryout (0.1 to 1 s) but not in departure from nucleate boiling (milliseconds). The split classifies each round-tube pair as DNB-type (mean exit quality below 0.1), transition, or dryout-type (above 0.3) and reports rig means by regime, with no p value and no verdict, so a reader can see whether pooling the regimes hid opposite or vanishing effects.

The addendum also records an erratum to version 1.0 (Lee’s tubes had 10 K inlet subcooling and exit quality near zero, not “strongly subcooled”) and two pieces of prior work the record should have cited: the thermal activity parameter of the pool-boiling literature, of which the record’s group Pi_w is a dimensionless form, and Hewitt et al.’s 1965 remark that wall-thickness effects may explain small discrepancies between laboratories. The consistency pass of the addendum is deposited with both of its runs; the first found five undefined symbols, corrected before freezing.

Archive SHA-256 (version 1.3): 116350f4286fcd9e5149fba060eecf6c174b0d8fa5286aec1b02702f5aaf5bb7; the version 1.2 archive inside it: 837631cc8cddebbea452aadbe3082b0a0d72fc9f18702aac9ec8b055f37f0f65