An EST-Compliant Tier-2 Evaluation on NYC TLC (2019–2023) This note reports a scope-gated Tier-2 result under EST discipline. The key finding is not a "best" estimator, but when pooled evaluation becomes invalid—and how preregistered windowing restores interpretability without erasing the pooled failure. Abstract Real-world socio-technical systems are rarely stationary, and this poses a fundamental challenge for empirical evaluation of structural indicators that assume estimator coherence across time. We present a Tier-2 empirical study applying the Force–Information–Time (FIT) framework with Estimator Selection Theory (EST) to New York City Taxi & Limousine Commission (TLC) Yellow Taxi trip data from 2019–2023. Our goal is to evaluate whether a cost-family of constraint estimators exhibits sufficient coherence to support interpretation of Information/Constraint regime signals (P11). The central finding is this: pooled evaluation across the full period fails the coherence gate (Spearman ρ = 0. 543, below the 0. 6 threshold), while all per-year windows pass, as do preregistered pre-COVID and post-COVID macro windows. This pattern is consistent with structural level shifts—something akin to Simpson's paradox—rather than estimator noise. 235 We formalize this outcome as a scope-limited success (OKPERYEAR, OKPERWINDOW), while explicitly preserving the negative pooled result. The study demonstrates that under EST discipline, failure of pooled coherence is itself a diagnostic signal of regime heterogeneity, and that meaningful structural interpretation requires preregistered phase conditioning.
Qien Huang (Thu,) studied this question.