Deal Detection Methodology
A deal is an active listing priced below what the card is worth, and confirmed to be the card it claims to be. Every deal on TCGBerg starts from the same yardstick as the rest of the site: the TCGBerg fair value for the specific atom, the (printing, grader, grade) tuple, that the listing represents. A listing becomes a deal candidate only when its all-in price (item plus shipping) sits below that fair value, and only when the fair value itself is trustworthy and current. Candidates then run a multi-stage verification cascade that cross-checks the grader, the grade, the printing, and where possible the slab's certificate, before the listing is allowed into the deal screener. The goal is a feed you can act on: genuine discounts on correctly identified cards, not keyword-stuffed titles or stale prices. This page explains the yardstick, the gates, the cascade, and the limits.
Maintained by the TCGBerg research team · Last updated June 30, 2026
What Counts as a Deal
A deal is measured against fair value, atom by atom. The unit is the atom: the specific (printing, grader, grade) tuple, not the card as a whole. A PSA 9 of one printing and a PSA 10 of another are different atoms with different fair values, so a listing is only ever compared to the fair value of the exact atom it represents. Fair value is the confidence-scored estimate documented in the Fair Value methodology; deal detection consumes it and never recomputes it.
The comparison uses the listing's all-in price (item price plus shipping), converted to USD. The discount is simply how far that price sits below fair value: (fair value minus price) divided by fair value. A listing at or below fair value is a candidate; a listing above fair value is not.
Because a single atom can have many simultaneous listings, we keep only the strongest few candidates per atom rather than flooding the feed with every below-fair listing. Those survivors are then verified and scored, and the best are published. So a deal is not just any cheap listing: it is a top-ranked, below-fair, verified listing on a card whose fair value we actually trust.
The Gates a Listing Must Clear
A big discount against a number we do not trust is not a deal. Before a candidate is ever scored, it must pass several gates that protect against false bargains:
Confidence floor. The atom's fair value must clear a minimum confidence (medium or better by default). A discount measured against a shaky, thinly-evidenced fair value is meaningless, so those atoms never produce deals.
Freshness. The fair value must have been recomputed recently (within about three weeks by default). A stale fair value can lag a real market move, manufacturing a discount that has already evaporated.
A living market. The atom must show signs of an active market: at least a few sales in the trailing year and a sale within roughly the last six months. A card that has not traded in a long time has no reliable current value to discount against, however large the apparent gap looks.
Auctions near close only. An auction's price is its live bid, which climbs toward fair value as the close approaches. An auction only qualifies once it is close to ending (within about a day), because earlier its current bid is still rising and is not a lockable price.
A sanity ceiling. An implausibly large discount (on the order of 85 percent off or more) is treated as suspicious rather than as the deal of the century, and is held back. A gap that big almost always means a mislabeled listing or a wrong fair value, not free money.
The Verification Cascade
Surviving candidates run a cascade of checks. Each stage can drop a listing and records why, so every published deal carries an audit trail of what it passed.
L0, candidate detection and ranking. Joins live listings to current fair values, applies the gates above, scores each candidate, and keeps the top few per atom.
L1, marketplace cross-check. Pulls the listing's structured item details from the marketplace and checks that the grader and grade stated in those fields match what we parsed from the title. A listing whose own data disagrees with its title is dropped.
L2, printing attribution. Confirms the listing really is the printing we matched. This catches language mismatches (a German card posing as English), variant contradictions (an Unlimited print posing as Shadowless, or the reverse), and ambiguous multi-variant cards (for example a Pikachu whose title does not say which cheek colour it is). Unresolved ambiguity is dropped or sent for a closer look rather than guessed.
L3, certificate verification (PSA). When a listing carries a PSA certificate number, we look it up against PSA's own records and confirm the slab resolves to the same card and grade. A cert that points at a different card or grade, or that does not resolve at all, is dropped.
L4, automated review (optional). For the highest-risk candidates, an automated title-and-image review provides a final reading of whether the listing genuinely matches the expected printing. It is a backstop for the cases structured data cannot settle.
What Gets Filtered Out
Most of the work in deal detection is removing false positives. A listing is dropped, with a recorded reason, when it shows any of:
- An implausible discount. Gaps around 85 percent or larger are almost always a mislabel, not a bargain.
- A language mismatch. A non-English card listed as if it were the English printing.
- A variant mismatch or ambiguity. The asserted print run contradicts the card's attributes, or a multi-variant card cannot be pinned to one printing.
- A certificate or grade mismatch. A PSA cert that resolves to a different card spec or a different grade than the title claims.
- A dead or stale market. No recent sales, or a fair value too old to trust.
- An auction that is not closing soon. Its current bid is still climbing and is not a lockable price.
- A sold or unavailable listing. It has ended, sold, or gone out of stock since the last crawl.
The effect is a feed weighted toward listings that are both genuinely cheap and genuinely what they say they are.
How a Deal Is Scored and Shown
Verified deals are ranked by a quality score built from four components, so the feed leads with the listings most worth acting on, not merely the steepest discounts:
- Value. How good the discount is, with a sweet spot rather than a straight line. A solid, believable discount scores best; an implausibly large one is treated as a red flag, not a bonus.
- Believability. Seller and listing trust signals: authenticity guarantees, top-rated and high-feedback sellers, multiple photos, accepted returns, and whether the price sits sensibly within the cohort of comparable listings.
- Conversion. Practical buyability: a sensible price band, free shipping, Buy It Now versus auction, and how soon an auction ends.
- Confidence. How solid the underlying fair value is, carried through from the Fair Value engine.
The combined score sorts deals into quality tiers (strong, moderate, weak), and the screener can require a minimum tier before a deal is published. Each card surfaces short reason chips (for example a discount badge, an Authenticity Guaranteed flag, or an ending-soon countdown) so the why is visible at a glance. Auction deals are framed honestly: the price is labelled Current bid, because it will climb by close, while a Buy It Now price is a lockable discount.
Limitations
A few things the deal screener is not:
- It is a screen, not advice. A flagged deal is a data-driven candidate, not a recommendation to buy. Always look at the actual listing.
- It is marketplace-sourced. Deals come from active marketplace listings (US and EU). Prices and availability move fast; a deal can be gone by the time you open it.
- The links are affiliate links. TCGBerg may earn a commission when you buy through a deal link, but the commission never affects whether a listing is flagged or how it ranks. You still buy on the marketplace at the listed price.
- It depends on fair value. Where there is no confident, fresh fair value, there are no deals for that atom, by design.