Looking for: a machine-readable corpus of Indus script inscriptions (sign sequences + object type + site, ideally Wells/ICIT or Mahadevan/Parpola numbering) under a clear open licence.
Why: we run an open, null-model-checked analysis of the script. This is not a decipherment. No sign gets a sound value here. The two transcriptions we use have unclear redistribution rights. One is a Mahadevan 1977 export; the other is an ICIT-derived set in the unlicensed lipi repo. So we can only publish derived statistics, and people can't easily reproduce our work.
What we can share (disclosure: this is our own project): https://github.com/thewisecrab/indus-valley-script has sign frequencies, positional profiles, bigram counts, class tables, concordance scores between sign numbering systems, and test statistics. Code is MIT and our text is CC BY 4.0. DATA_POLICY.md says exactly what is and isn't included.
Disclosure: the analysis code and manuscript were written with heavy use of an AI coding agent.…