The File That Counted Itself Twice: Paper #224's Deposited Catalog, Its Bin Widths, and the Spread That Survives Both
Pays the two grade-A debts Paper #224 (One Ruler, Three b-Values) printed against itself, by exact recomputation from its own record. (1) The deposited-file kill (D-220-237-19): recomputing the three Gutenberg-Richter estimators from the deposited California ml catalog gives spread 0.2531 in every state of the file, so the kill condition (spread at or under 0.0639) does not fire and #224's priced claim survives -- but the deposited file itself is corrupt: it holds 20,660 rows under SHA-256 9c2a5557... where the manuscript prints 10,330 rows and SHA-256 2cee4642... Dissection proves it is the correct snapshot concatenated to itself byte for byte (every event exactly twice; the first half hashes to the printed value), and a sandbox reproduction proves the mechanism: the original notebook's input glob matches its own output snapshot, so one re-run doubled the file. The values survive because every estimator in the protocol is invariant under exact duplication; only maximum likelihood's formal error moves (0.0207 to 0.0147). The provenance sentence is killed at full size in DEBT-224/KILLS.md. (2) The bin-width spread (D-220-237-20): the catalog is mixed-precision (39.98 percent of headline magnitudes land exactly on a tenth, against ~10 percent expected under pure hundredth rounding). Exact grouped likelihood gives b = 1.4288 at the finest rounding present, a mixed-precision fit gives 1.4478 (tenth-rounded share 0.332), and the continuous limit is 1.4526 -- the printed headline MLE of 1.2445 (Utsu width 0.1) is low by 0.184, 8.9 times its printed formal error. Corrected estimator spread: 0.4374, 6.85 times the largest formal error (printed: 0.2531, 3.96 times); the straddle of b = 1 stands. Attacks: adversarial independent re-binning cannot push the spread under 0.0983 (under 0.1999 with defensible widths); 200 bootstrap resamples never return a spread at or under the kill bar; the b-positive estimator splits under its own binning (1.1153 raw, 1.5006 corrected) and adjudicates nothing. The completeness sweep's printed cut labels 2.8/3.2/3.8 are corrected to the notebook's actual quarter cuts 2.75/3.25/3.75. Graded DERIVED-INDEPENDENT.
Authors
- William Alexander Aldred
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-02
- DOI
- https://doi.org/10.5281/zenodo.23926530
- Primary Topic
- earthquake and tectonic studies
- Type
- preprint