Standing Pre-registration: Six Frozen Water CHF Models Re-Scored on Later Blind Sets, Each Under Its Own Frozen Rule

This programme has frozen and published critical heat flux (CHF) models in versions 1 to 7, besides the setup-aware S and B* models and the hybrid, which are scored under their own protocols and which this rule does not cover. Version 1, version 2's regional model, version 3 and version 5, models for water, have each been scored on the blind data of their own pre-registered test, each passing, failing or void by its own pre-registered rule; version 4, frozen for refrigerants only, failed its own test on R-134a and R-12; versions 6 and 7 await their test. This record pre-registers a standing rule: every covered water model (version 1, version 2's regional model, version 3, version 5, version 6 with its Addendum A1, and version 7) is scored again on later blind water CHF point sets the rule admits, each by its own original pass rule against the 2006 look-up table and in its own frozen domain (version 1 and version 2's regional model, which froze none, in a domain the rule sets now). Each verdict reads as that version's frozen rule re-applied to the set, never as a new test of a changed model, and every verdict is published whichever way it falls. A set is scored only after its own test has been scored, from that test's hashed transcription; points from laboratories the programme has already seen are removed by a closed, frozen list. As frozen, the rule admits sets run with version 5's or version 6's evaluator, so in practice the version 6 test set; any other set joins only by an addendum made before it is opened. On the version 6 set, versions 6 and 7 are quoted from their own runs and version 1, version 2's regional model, version 3 and version 5 are scored. Before the record was frozen, the rule's code had to reproduce the published results of version 1, version 2's regional model, version 3 and version 5 on their own spent test files, and two further checks covered the version 1 adapter and the shared preparation code; versions 6 and 7 have no scored test yet, so no check of their own. All six checks passed in their one counted run (stage two, 24 October 2026): version 1's 14.75% against 11.19% on 888 KAERI tubes, the regional model's 14.12% against 10.94%, version 3's void result, version 5's passing result, and the adapter and preparation checks. The log and run folders are in the deposit. The deposit holds the frozen rule (text and PDF), its record of hashes with the six PASS lines, the seen-laboratories list, the version 6 source record, the run log and the six check folders, and the code with its test suite. The rule states its limits, among them narrow domains that will leave many sets void and up to 18 correlated rules per set, with no programme-level verdict drawn. Archive SHA-256: 7f99ccaaf7f0985378b9a6a55847be9118d308db7fd2ef14c3da55c3b445be36