Comparable cards
The model learns how age, era, set, rarity, finish, Pokémon, illustrator, collector form, and other card characteristics relate to market price.
Collector analytics · RVM V0.3
Relative Value asks a focused question: compared with cards that share similar collector characteristics, does today's market price look unusually low, typical, or unusually high?
The short version
The model learns how age, era, set, rarity, finish, Pokémon, illustrator, collector form, and other card characteristics relate to market price.
Published Pokémon popularity bands are included carefully. Their influence changes with rarity and collector form so an ordinary card is not treated like a premium chase card.
For supported Scarlet & Violet sets, sourced rarity pull rates help represent scarcity. The model uses pull difficulty—not the current dollar price of sealed products.
Instead of presenting one supposedly perfect price, the model produces an expected range and a confidence tier based on its historical error on similar cards.
Under the hood
RVM V0.3 is a peer-pricing model, not a future-price forecast. It models the natural logarithm of market price with a regularized Ridge regression so very large prices do not dominate the fit. Categorical details are encoded as peer groups; numeric inputs are standardized and missing context is handled explicitly.
Model midpoint = exp(predicted log market price)
Peer range = exp(predicted log price + calibrated lower/upper residual)
Relative Value label = current price compared with that peer range
Pokémon species, illustrator, role/type, collector form, and collector-number pattern.
Rarity, finish, edition/subset, distribution class, set, era, and release age.
Official 2020 Pokémon of the Year top-30 popularity bands, conditioned by rarity and collector form.
Estimated packs for supported Scarlet & Violet rarity groups only; unavailable groups remain missing rather than guessed.
The “Why this estimate” cards describe model signals, not causal proof. A Pokémon or illustrator can appear as an upward signal because comparable cards historically priced higher after the other recorded characteristics were considered.
Reading the result
The labels describe the relationship between a card's current market price and its modelled peer range. They do not predict where the price will go next and are not recommendations to buy or sell.
The current market price is below the model’s expected range.
The current market price falls inside the model’s expected range.
The current market price is above the model’s expected range.
Confidence
High, Medium, and Low confidence describe how accurately the model has historically handled cards with similar characteristics. A High-confidence result can still be wrong; it simply comes from a more reliable part of the model.
Historically most reliable
Moderate historical reliability
Use the range more cautiously
Cost to pull
A sourced rarity hit rate is divided across the eligible cards in that rarity bucket. This assumes cards within the bucket are equally likely unless stronger evidence is available.
The reciprocal of that probability gives the average number of packs associated with one copy. Random openings can vary dramatically from that average.
Estimated packs are multiplied by a dated effective pack price. For special sets, product price is divided by the documented number of packs in that product.
Dollar cost-to-pull is display-only. Current sealed-product prices can reflect the same demand affecting singles, so feeding those dollars into the model would risk circular pricing.
Validation
V0.3 was evaluated on held-out cards, held-out sets, multiple random seeds, and every supported set individually. The released payload exactly matches the reviewed shadow model, while V0.2 is retained as a rollback version.
Boundaries
The model does not see card condition, grading, short-term sales momentum, population reports, seller liquidity, or private transactions. It uses the published market variant supplied to the site, so an unusual printing or stale price can also distort the comparison.
No price prediction. The range describes comparable-card pricing now, not next month.
No invented scarcity. Missing pull-rate evidence stays missing.
No sealed-price feedback. Dollar cost-to-pull never enters training.
No automatic retraining. V0.3 remains hash-frozen until a separately reviewed release replaces it.
Next data phase
The next coverage project will add sourced pull-rate and pack-price context for English sets from Sword & Shield onward. Sets will only be published when the source, sample size, rarity mapping, and pack economics can be documented clearly.