Building Tools, Not Just Papers: Why I Code

When I started this research, I chose to develop both manuscripts and inspectable research systems for studying miscalibrated advice.

CheckMyCoach is more than a manuscript concept: it includes 44 passing tests, a functioning API pipeline, a Streamlit frontend, and an observed API cost of approximately $0.001 per attempted correction in the evaluated configuration.

Why does this matter? Because trust calibration research benefits from inspectable implementations. Implemented prototypes make assumptions, failure modes, and future human-study designs concrete.

For released artifacts, I provide quick-start instructions, environment information, tests, and versioned schemas where available.

Reproducibility status is reported separately for each project; independent reproduction and cross-lab convergence are not assumed from documentation alone.