mbruton/keybridge-models

Shared decipherment models for historical substitution ciphers

Each model in this collection has been trained for one known historical cipher key. It reads transcribed ciphertexts written with that key and produces a draft decipherment. Testing a ciphertext against all models shows which known key, if any, it was written with.

Use the models with the notebooks or command-line tool at https://github.com/mbruton0426/keybridge. You do not need to download anything from here by hand.

Contents

File	Purpose
index.json	Lists every model with its symbols, transcription format, language and source. The tools read this file first.
<model id>/<model id>_model.pt	The trained model.
<model id>/<model id>_key.tsv	The key it was trained on.
<model id>/<model id>_training_report.txt	Training settings and accuracy on synthetic text.
<model id>/README.md	Description, source of the key, and contributor.
submissions/	Models contributed by others, waiting for review.

Models

ID	Cipher	Transcription format	Source of the key
maximilian_II	Letters of Maximilian II (1575)	symbols separated by spaces (single characters; convert one-character-per-symbol files with keybridge convert)	DECODE IDs 1471, 1412, 1407
maximilian_1472	Maximilian letter with a different key	symbols separated by spaces (named symbols)	DECODE IDs 1472, 1473
copiale	Copiale cipher	symbols separated by spaces	DECODE ID 2

All three were trained on 5,000 synthetic examples from the built-in English HistCiph plaintext, with rare-unit insertion at 0.02 (see each model's training report).

Contributing a model

Train a model for your key with notebook ① and tick Share my model. Your contribution will be reviewed and added to index.json.

Licence and citation

All models and keys are released under CC BY 4.0: you may use and adapt them for any purpose, provided you give appropriate credit. Please cite:

[citation to be added]

When you use a specific model, please also credit the original source of its key, given in that model's README.