Data for: Breaking the Illusion of Identity in LLM Tooling

Experimental data for “Breaking the Illusion of Identity in LLM Tooling” (Miller 2026). Contains 1,560 JSON API response files from 780 two-turn conversations with Claude Sonnet 4 under two conditions (default register vs. constrained voice model), 30 software-development tasks × 13 replicates. Derived results include per-conversation marker counts (7 rules, 82 regex patterns), per-rule compliance verdicts, RoBERTa-based AnthroScore values, and summary statistics. Directory structure:- default/ and constrained/: turn-1 responses- follow_up/default/ and follow_up/constrained/: turn-2 responses- results/: marker_counts.csv, compliance.csv, anthroscore.csv, summary.json- prompts.json: the 30 task prompts