Indus signs written abroad combine in ways the home script does not: two pre-registered tests on inscriptions from the Gulf and Mesopotamia

Scholars have long observed that Indus inscriptions found in the Gulf and Mesopotamia carry sign sequences not found in the Indus Valley, and have proposed that the script was adapted there to another, probably Semitic, language. This paper tests that claim with two frozen tests, each committed publicly before scoring. A text's score is the share of its adjacent sign pairs never seen in home Indus texts; controls are texts from small home sites, each scored without its own site. On 17 abroad texts never analysed for this question, 83% of sign pairs were unattested at home, against 49% for 334 control texts (p = 0.0006). The effect holds for pairs of two common signs (67% vs 23%, p = 0.0002), so it is not caused by rare signs. A replication on a blind transcription of Laursen's (2010) Gulf Type seals gave 86% against 49% (9 seals, p = 0.005); after removing three possible duplicate objects, 6 seals still gave 84% (p = 0.015). Abroad, Indus signs were combined into new sequences, as expected if the script was writing different words. Descriptively, the human-figure sign, which closes texts at home, tends to open them abroad; this is sealed as a prediction for future finds. Data and code: github.com/RN-Top/Indus-Research-