Replication package for "Plain Honest Exemplars Match Reflection-Formatted Priming: In-Context Honesty Priming and Compliance with Deceptive System Directives in Five Chinese LLMs"

Replication package for the IEEE ICCC 2026 paper. Contains the two black-box benchmarks (30 fabrication + 34 deception scenarios) with their machine-generated no-directive variants, the three token-matched exemplar libraries (reflection / plain-honest / neutral), all generation/judging/analysis code, the scored result files from three independent LLM judge families, human-validation annotations, a post-hoc run manifest, separated failure logs, SHA-256 file hashes, and the statistics output from which every table and figure in the paper is generated. Run `python reproduce.py` to regenerate the statistics, agreement metrics, and figures. See README.md for details.