Tianlang (Jackie) Chen § 3 · PROJECTS
Enter the 3D Website
TIANLANG (JACKIE) CHEN · PERSONAL WEBSITE § 3 · PROJECTS
§ 3

Projects

2 SPECIMENS · AI EVALUATION · 2026
Independent Research AI Evaluation
PR-2026-01 time-series foundation models in progress
PR-2026-01 PROJECT 01

Auditing Regime-Stratified Evaluation for Time-Series Foundation Models

Tianlang (Jackie) Chen

Developed a pre-registered self-audit framework for regime-stratified evaluation of time-series foundation models on traffic forecasting. The project uses meaning-destroying but structure-preserving null controls, label-family sensitivity analysis, cross-dataset checks, and interval-calibration diagnostics to test whether a reported conditional failure is genuinely attributable to the named regime. It reframes a refuted headline as four measurable preconditions for trustworthy stratified evaluation.

◌ Public GitHub Demo and Paper in Preparation
Independent Research LLM Evaluation
PR-2026-02 measurement systems in progress
PR-2026-02 PROJECT 02

Auditing Decision Rules and Reproducibility in LLM Evaluation

Tianlang (Jackie) Chen

Built an audit methodology for rubric-based LLM evaluation that examines two properties hidden by point estimates: whether a decision rule can resolve the effect size it claims, and whether the measurement pipeline reproduces its own claim-bearing statistic. Exact bootstrap-support analysis and controlled reruns expose low-resolution thresholds and run-to-run movement, turning evaluator reliability into a measurable part of the result rather than an untested assumption.

◌ Public GitHub Demo and Paper in Preparation
FOL. 022 · PROJECTS
2 SPECIMENS · v1.0