From Residual Candidates to Population Inference: Auditing Early Excesses in Type Ia Supernova Forced Photometry

Abstract Candidate flags from one-sided residual thresholds are useful for follow-up but do not directly estimate a population fraction. We apply the Early-Excess Residual-Inference (EERI) audit to a ZTF-selected sample with ATLAS forced photometry, combining denominator reconstruction, conditional residual-sign tests, full-refit injection–recovery, point-level forensics, and label-blind external comparison. The sample ladder contains 1000 raw jobs, 920 secure associations, 761 objects with early coverage, 368 passing the native-band gate, 303 with valid interior fits, and 277 in the primary denominator. Two objects cross the positive rule and three cross its sign-reversed counterpart. The exact band–night block-sign randomization gives E [ N + ] = 2.25, P ( N + = 2) = 0.34375, and a two-sided probability-ordered p = 1.0; a complete-vector reversal sensitivity gives the same inferential conclusion. Neither diagnostic is a survey false-positive rate. No observed candidate passes the structural-confirmation rule. Of 41,550 primary full-refit injections, 444 cross the raw rule and eight survive single-point deletion, but none has two-band support. End-to-end efficiency is therefore zero for every tested proxy, so the frequentist confidence set remains the full physical interval and provides no occurrence constraint. The scientific result is an identifiability audit: a fixed denominator and apparently useful raw sensitivity do not substitute for recovery under the endpoint applied to observed data. Four of 17 Müller-Bravo catalogue objects enter the primary denominator; none crosses the EERI rule.

Hali tarjima qilinmagan