Preregistration: Blind Cost Prediction for Riftbound Set 5 (Radiance) with a Frozen Designer-Pricing Model

Abstract

Preregistration of a blind out-of-sample test of a designer-pricing model for the Riftbound trading card game. The model (pricing rule set v6, described in the preprint https://doi.org/10.5281/zenodo.22909590) is frozen on 24 September 2026, before the preview season of Set 5 (Radiance). Radiance cards will be priced from their text without seeing their costs, and eight hypotheses are tested with pre-specified criteria. Version 2 adds an addendum, written before any Radiance card was priced, registering two secondary hypotheses: H9 (repeatable triggers are booked at about half an activation, a quarter under an independent condition) and H10 (one-off "if…then" conditional bonuses are booked at zero). The original files are unchanged. Addendum SHA-256: b06f822bf10b76abe036663a07fb3528a8d9a10d62c76446f15be7514f38129b; scoring script score_h9_h10.py: 2b5a02c5fd8f96cc506d02c910321131863386042c8e8b5fd60f9ca38aed9ccd.Blind pricing of the Radiance set, produced from card text only before any cost was viewed. The original file (新系列_盲定价.csv) was then reviewed by the first author before unblinding; four reading/omission errors were corrected, each with the card-text basis listed in 修改清单_草案.csv. The primary analysis uses the revised file (新系列_盲定价_修订版.csv); the original is reported as a sensitivity analysis. Hashes in SHA256_上传包.txt and SHA256_修订.txt.