Data and code of computational cryptanalysis of the D'Agapeyeff cryptogram (1939)

This repository contains the raw data and source code underlying a computational cryptanalysis attempt on the 196-symbol challenge cryptogram published by Alexander D'Agapeyeff in 1939, at the end of the first edition of his book Codes and Ciphers (Oxford University Press). The cryptogram has never been solved. This dataset is the companion of a full research report (methodology, results, discussion) documenting the entire investigation; it provides everything needed to verify, reproduce, or continue the search. The repository is organized into four parts: the Python search code (one script per linguistic hypothesis and cipher family), the training corpora and derived language models, the raw results of every search performed (compressed, several million logged combinations), and the exact coverage status of each search at the time it was stopped. A detailed manifest of every file, its provenance, and the exact format of the results is given in the accompanying README. Four classical cipher families (substitution, columnar transposition, Fleissner grille, double transposition) and a Four-square digram cipher were tested systematically under three linguistic hypotheses (English, French, transliterated Hebrew), using a calibrated quadgram language model and simulated annealing. Every result was checked against a null-baseline calibration (the same search run on a randomly shuffled version of the cryptogram, at equal trial volume) to distinguish genuine signal from statistical artifacts arising from testing very large numbers of combinations. At the project's closure, methods A through D reached guaranteed 100% coverage over two independent systematic sweeps (main run and null baseline), with no statistically distinguishable signal detected. The Four-square cipher was covered to 20.9% of its search space before the project was closed for lack of further computing resources, again with no signal found. The dataset therefore documents a rigorous negative result, published for transparency and to allow the search to be resumed exactly where it left off. The underlying raw results range up to several gigabytes and exceed arXiv's submission size limits, hence their publication here rather than as supplementary material to the report. Code and data are released under the CC BY 4.0 license. The companion research report is currently being submitted to arXiv, after what data will be publicly available. A parallel French-language version of this dataset and report is published separately on Zenodo.