Information Leakage at Population Scale: An Evaluation of the Polymarket Insider-Relevant Subpopulation, 2020-2026

Referee: The central scope claim—that only 88 of 12,708 markets (0.7%) yield computable ILS-dl values and that resolution semantics are the binding constraint—depends entirely on the authors' binary manual classification of markets as computable versus ambiguous or unclassifiable. No explicit decision protocol, criteria for 'genuine resolution-criterion ambiguity,' or inter-rater reliability statistic is supplied. Because resolution wording is interpretive, even modest reclassification rates would materially change the reported effective domain and the paper's primary conclusion.

Authors: We agree that an explicit decision protocol is needed to support the classification. The binary distinction was based on whether resolution criteria permitted unambiguous identification of both T_event and a valid anchor date without interpretive ambiguity in outcome determination. We will add the full decision criteria and protocol as a new appendix or subsection in the revision. The classification was performed by the lead author; we will explicitly discuss this as a limitation and note that inter-rater reliability could not be computed. We maintain that the 0.7% computable rate is unlikely to change materially under modest reclassification, given that most exclusions arise from clearly ambiguous resolution language rather than borderline cases. revision: partial

Referee: The hazard-decay baseline correction is introduced to address negative raw ILS-dl medians and produces the reported heterogeneous findings (regulatory_formal post-2024 near zero; regulatory_announcement post-2024 with 95% CI entirely below zero). This adjustment is presented without an external benchmark, pre-specified validation, or full specification of its free parameter, making the post-correction signals dependent on an ad-hoc choice whose impact on the leakage interpretation is not quantified.

Authors: The hazard-decay correction was introduced post-hoc after observing consistently negative raw medians, which we view as indicating baseline model misspecification rather than true negative leakage. We will fully specify the free parameter (decay rate) and its selection procedure in the revised text, along with a sensitivity analysis showing how the post-2024 heterogeneous signals vary with plausible parameter values. As an exploratory adjustment developed during analysis, no pre-specified validation or external benchmark exists; we will add explicit discussion of its ad-hoc character and limitations on generalizability. revision: yes

Referee: The independent-second-pass T_event validation is reported at 57.8% exact-date agreement against a 90% ex-ante criterion, and only 12 of 88 markets satisfy anchor-sensitivity. Without details on the exact rules applied in the second pass, error propagation, or how disagreements were resolved, it is difficult to evaluate whether this result undermines the framework's reliability claims.

Authors: The 57.8% agreement is presented as a substantive finding that the framework does not meet the pre-specified 90% target, reinforcing the paper's conclusion that resolution semantics and event dating remain barriers. We will expand the methods section to detail the exact rules used in the independent second pass, the approach to error propagation, and the resolution of any date disagreements. This will allow readers to assess the reliability implications directly. revision: yes