Synthesis analysis plan v0.64 (frozen): the 2006 look-up table, standard correlations and laboratory offsets on opened water critical-heat-flux sets

This record is the frozen analysis plan for a synthesis paper of Guarino Industries LLC's critical-heat-flux (CHF) programme. The plan says how the 2006 CHF look-up table, standard correlations and mechanistic models fare on the programme's already opened water data sets once laboratories are accounted for, whether their errors and laboratory offsets go with the decade in which the experiments were run, and where the programme's frozen model versions stand beside them on their published figures. It is an exploratory, information-only analysis, not a blind test: every figure it will produce is labelled post hoc, and it has no hypothesis, no pass rule and no verdict. Its value is in being fixed before it is run. Sets, methods, code and outputs are frozen here, the run is done once, and every output is published whatever it shows. It has five parts: A, standard correlations; B, laboratory as a random effect; C, prediction intervals; D, mechanistic models; E, era of measurement; and a quoted section, Q, that places the programme's frozen versions beside them on their published figures only. The run waits for three gates of other tests of the programme (plan Section 1). No results exist. Nothing in this record has been run on any real data row of the sets it names. The plan's first line still reads "NOT FROZEN", as every draft of it did, and the Biasi rules v6 say "Draft for plan v0.62; not frozen". The files are frozen as they are, with those words: the freeze is the author's approval of this package, as the package README says. A failure found before the freeze, and the rule added after it. The plan's statistics code was checked before the freeze on made-up grouped data (STATS_DRY, 70 sets). In double precision, pass rule R3 failed on one constructed set, d4: the plan's estimate of the between-unit standard deviation tau was 11.117628 against the independent reference's 11.118139, and the reference profile at the plan's estimate lay 1.5e-7 below its maximum, against a stated allowance of 1e-9. A 50-digit recomputation showed that the double-precision log likelihoods on d4 are in error by about 2.2e-7, more than the rules' tolerances, and that at 50 digits the plan's estimate on d4 is the worse one, by more than both stated tolerances (its log likelihood 1.99e-8 and its profile 1.75e-8 below the reference's). The author ruled to keep the code and to add a precision clause to pass rules R1 and R3: where a 50-digit recomputation shows the double-precision log likelihood on a set to be in error by more than a rule's tolerance, that rule is judged on that set at the measured error. As technical choices under that ruling, the analyst (the AI assistant that drafts and codes for the programme) set the widened tolerance at the larger of the stated tolerance and twice the measured error, as a margin, and ran the measurement on every fitted set; at one times the measured error every verdict is the same. The clause was added after the failure had been seen, and the plan says so. Measured on all 62 fitted sets, only d4's error exceeds a stated tolerance (2.17e-7; the next largest is 1.4e-10); under the clause d4 passes R3 (gap 1.50e-7 against 4.35e-7), and R1 and R3 pass on all 62 sets. d4 passes only because double precision cannot decide the rules there, not because the plan's estimate is within them. The double-precision failure and the 50-digit figures are published unchanged in the checks file, prefreeze_v063/CHECKS_v0.63.txt. Version 0.63 also corrected a statement of v0.62: a second profile peak in sigma, which v0.62 listed as a path the drawn STATS_DRY sets do not reach, is reached by 21 of the 66 drawn sets. The resumable-run check was run on the made-up dry inputs: a dry run paused, then killed outright, then resumed gave outputs byte-identical to an unbroken run. Version 0.64 corrects v0.63's attribution of the precision clause (items 1330 and 1331 of the programme's development log, which is not part of this deposit): v0.63 presented the whole clause, the factor 2 included, as the author's ruling. It changes no code, rule, tolerance or verdict; its checks are in prefreeze_v064/CHECKS_v0.64.txt. Freeze rule (plan Section 12): the plan is frozen when the author approves a final draft after independent checks, with the driver and its wrappers, the reference step, the source key, the era coding, the made-up dry rows, STATS_DRY and its results, the software pins, the model code, the checks files CHECKS_v0.63.txt and CHECKS_v0.64.txt and their outputs, the Biasi rules, grid counts and hand cases, the fixed-inlet files, and the Bowring drop addendum with its errata. Any change after the freeze is a dated addendum, published before the run. The package holds the plan as text and as a line-for-line PDF rendering, the addendum dropping Bowring (1972) from Part A, the errata (ERRATA.txt), the code under the MIT License, the made-up dry input files and hand cases, the checks files with the folders of their outputs, a README with every file's sha256 and the commands that run the checks, and a SHA256SUMS list. It holds no source article, no extract of a source article, and no real data row. The 2006 look-up table code and data are frozen with version 3 of the programme and are not repeated here. Source documents were obtained through the Interlibrary Loan service of J.Y. Joyner Library, East Carolina University, by the author as an ECU student. The author thanks the Interlibrary Loan service of J.Y. Joyner Library, East Carolina University. Licences. Code (every .py and .sh file in the package): MIT License, Copyright (c) 2026 Guarino Industries LLC. Text (the plan, its PDF rendering, the addendum, the errata, the rules, the checks files CHECKS_v0.62.txt, CHECKS_v0.63.txt and CHECKS_v0.64.txt, the README, and the metadata sheet METADATA.txt with its PDF rendering METADATA.pdf, both in the package): Creative Commons Attribution 4.0 International (CC BY 4.0). Made-up rows and computed check outputs (the dry input files in synth_dry/, the hand cases, the STATS_DRY sets and results and the dry-run records in prefreeze_v063/, and the check records in prefreeze_v064/): MIT License, with the code. Bibliographic coding files (ERA_CODING.csv, S46_SOURCES.csv and synth_refs/outputs/S3_REFERENCES.csv): CC BY 4.0 with the text. A file named above under CC BY 4.0 is under CC BY 4.0 wherever it sits, the checks files inside prefreeze_v063/ and prefreeze_v064/ included. Every other file in the package (among them requirements.txt, extract/BIASI_GRID_COUNTS.txt, synth_refs/outputs/run.log, the SHA256SUMS lists and the check records) is released under the MIT License, with the code. Declaration on the use of artificial intelligence. The plan, the code, the checks and this package were drafted and built by an AI assistant (Claude, Anthropic) working for the author. Independent checks were run by separate AI agents that did not do the work they checked. The author made the intellectual commitments of the programme, ruled on the STATS_DRY failure, and approves the freeze. The author is responsible for the content. The metadata sheet in the package (METADATA.txt and METADATA.pdf) holds this description as it stood before four small wording changes made on this page before publication.