Recovering the Designer's Ledger: Inverse Hedonic Pricing of Card Costs in a Collectible Card Game

Collectible card games (CCGs) price every card in a single integer resource, yet the rules mapping card text to cost are rarely published. We cast their recovery as an inverse hedonic pricing problem and propose ZHIZHI (Zero-residual Hedonic Inversion, Zero-shot Hedonic Imputation): Stage 1 inverts effect prices from anchor cards that differ in a single effect, so most prices are identified exactly; Stage 2 imputes any card's value from its text without refitting. The value-to-cost ratio is reported as the AMBER score (Attribute-Matched Budget-Efficiency Ratio), with 1.00 as par and repeatable triggers booked at half an activation. Applied to Riftbound, Riot Games' League of Legends card game, the framework fully prices 497 of 702 costed units and spells and 69 of 107 gear cards, extending a linear additive core with mechanism models for hand cards, selection and on-play effects. The table exposes latent design regularities: a 50% rule for repeatable contingent value (one-off conditions mostly at zero), free play-timing keywords, a 1 C same-target complementarity premium, and integer-rounding absorption of small effects on cheap units. The AMBER score has median 1.00 (interquartile range 0.86–1.07; 0.80–1.09 excluding anchor cards); calibrated prices predict held-out cards, and a nested ablation attributes nearly all error reduction to on-play adjustments. In a preregistered blind test on the next set, the frozen model priced 60 new cards with RMSE 1.73 C (threshold 1.84) and median score 1.00; booking triggers at half an activation lowered its error (RMSE 2.00 to 1.30 C). Version 4 adds the preregistered blind test of the frozen model on the fifth Riftbound set, Radiance. The model (pricing rules v6 and the value table) was frozen and registered on 24 September 2026 (concept DOI 10.5281/zenodo.22928382), before the set's preview season; the blind pricing of Radiance was published as version 3 of that record (10.5281/zenodo.23306713) before any cost was read. Main results (Section VIII): both primary hypotheses hold. On 60 fully priced units and spells the frozen model predicts printed cost with RMSE 1.73 C against a registered threshold of 1.84 C, and the median nominal AMBER score is exactly 1.00. Booking repeatable triggers at half an activation is supported out of sample (H9a; H9c: RMSE 2.00 to 1.30 C on 19 units). Timing keywords are again free (H8). Hypotheses with too few cards in this set (H3, H4, H5, H7, H9b, H10) are reported as untestable. Exploratory additions (Section IX): first prices for the set's new keywords (Disarm, Show Off, Bomb, Deploy); the designer's price for stun fell after the first set (any stun 2 C from the second set on); "can't move this turn" is priced at −2 C, unlike exhausting a unit (−1 C). Other changes: abstract, contributions, limitations and conclusion updated; related work extended; Data Availability cites the blind-pricing record. Tables II and V mark the revised stun price and the new "can't move this turn" price; all statistics outside Section IX still use the frozen v3 prices and are unchanged.