Daily Buys

Why this engine looks nothing like Weekly Buys. Before writing it we scored the weekly ledger’s 72 marked picks factor by factor (diagnosis run 2026-08-30). The findings drove every design choice, and we publish them: print-clock rode on 67 of 72 marked picks at a 26.9% win rate and −3.07% mean alpha, so a factor attached to almost every pick selects nothing and is not a daily factor; stamp-freeze (0 wins in 5) and sealed-ev (0 wins in 4) are dropped; and static cohort evidence lost −4.68% mean alpha while the picked card’s own tape fell, so every daily factor is a per-card recorded signal and no pick may fight its own falling tape.

The three-factor bar. A card must clear at least three of seven typed factors, each computed nightly from recorded tapes with disclosed formulas: species-demand (the species’ demand-index percentile among all tracked species, at or above p80), finish-favor (the finish class in the favored tail of finish cohorts by median 90-day trend, p75 or higher, silent until the tape is deep enough), gem-drift (recent gem share at least 3 points below the lifetime rate: new top-grade supply tightening), float-tight (active TCGplayer listings under 0.5% of PSA pop, pop 500+), cheap-multiple (PSA 10 premium in the cheapest 35% of its era cohort on 10+ comps), velocity (2+ verified PSA 10 sales per week), and uptrend (the card’s own recorded tape up 8%+ over roughly 90 days). Two hard gates apply before any factor counts: a fresh entry mark from the daily price dump, and a 30-day own-tape trend no worse than −2%. At most four picks publish per day, ranked by factor count then summed factor strength; a picked card cannot re-enter for 45 days, so a single name can never compound across editions the way the weekly diagnosis caught.

Prices, marks, and vehicles. Raw picks enter and are marked at the TCGplayer market price from our daily full-catalog sync; slab picks (published when the cheap-multiple case rides a deep PSA 10 tape of 50+ recorded comps) enter and are marked at the eBay PSA 10 sold median from our graded series. Entry and mark always share one source, recorded in the append-only ledger on publication day and never edited. “Pay up to” is a discipline ceiling at entry +10%, not a valuation. Alpha is the pick’s return minus the moonstone Pokémon Index over the same window. Marks are unrealized mid-horizon readings; verdicts are issued when a pick reaches its stated horizon. The factor-attribution table above retires any factor that fails to earn alpha, exactly as the weekly series’ attribution loop exposed its losers.

Revision log. v1.0 · 2026-08-30: engine built from the weekly-ledger diagnosis (72 marked picks); factor set calibrated on the live 500-card feature universe (23 candidates cleared the three-factor bar on calibration day); day-keyed ledger opened alongside the untouched weekly ledger. Criteria changes will be logged here and never applied retroactively to logged picks.