Interruptive Thinking: coded interruption events from one user's LLM working sessions

Coded data accompanying Garg (2026), "Interruptive Thinking: An Empirical Study of Operator-Side Control in AI Interaction", PPIG 2026. The dataset records 611 interruption events, moments where the user stopped or redirected a large language model, from 29 of the author's own working sessions (April–May 2026). Each session was coded by one or more LLMs (Claude, ChatGPT, Gemini) against a fixed codebook, giving 47 receipts. Files: events.csv (one row per event), receipts.csv (one row per receipt), codebook.md, a README, and reproduce.py, which recomputes the paper's reported counts. Raw transcripts and quoted user turns are not included. Session names are anonymised, and product and company names are replaced with placeholders. See the README for full details and limitations.