Methodology

Every price on this site is computed the same way, from the same inputs, by the same rules. This page is those rules, with the real constants. It is free, carries no advertising and no sponsor, and every change to it is dated in the revision log at the bottom.

Where the prices come from

Completed public sales only. Two source classes feed the database: marketplace sold listings, supplied under a flat commercial data licence, and auction-house results from Heritage Auctions, Goldin and Fanatics Collect. Asking prices never enter a price. The unit of pricing is the exact triple of card, variant and grade. A PSA 9 is never pooled with a PSA 10, a shadowless printing is never pooled with an unlimited one, and one grader's 9 is never pooled with another's.

The window

The default window is the trailing 30 days. Fewer than 5 sales in 30 days widens it to 90. Fewer than 3 sales in 90 days and we publish no price: the page shows the sale count, the words "insufficient data", and the last sale with its date.

What we exclude, and how

Hard exclusions first, each listed on the page it affects: multi-card lots, proxies and reprints, damaged slabs, listings with a quantity above one, and sales our matcher could not tie to the exact card, variant and grade at 0.92 confidence or better.

Then one statistical rule. Prices move to log space, and each sale gets a modified z-score against the window's median, scaled by 1.4826 times the median absolute deviation and floored at 0.01 log units. A sale scoring above 3.5 is set aside and shown as excluded. Below 5 sales the rule does not run at all, and the page says so.

And a cap. If the rule wants to reject more than 20% of a window, that window is usually two populations, not one population with outliers: an unseparated variant or qualifier. We then reject nothing, publish the plain median, send the cell to human review and cap its confidence at Thin until a person resolves it. Trimming a second population produces a confident, wrong number.

The estimator

The published figure is the Slabpricing Market Value, SMV. When 10 or more sales remain after the checks above, the lowest 10% and the highest 10% by price are dropped. Each remaining sale is weighted by age with a 14-day half-life, so a sale two weeks old counts half of one from today. SMV is the weighted mean of that set. It is the only estimator in the product: the tier badge, the gold figure, the indices and the portfolio all read this one number.

Beside it we publish the kept count, the rejected count, the window in days, the raw median, the interquartile range, a confidence interval, a volatility figure and the source mix.

Evidence Grades

Four grades, and they gate what we will claim:

• Confirmed — 8 or more sales in 90 days. A price, a change figure, indicators.
• Thin — 1 to 7 sales. A price, no change figure. A handful of sales is not a market.
• Stale — no sale in 90 days, history exists. The last price stands, with its age printed.
• Unrecorded — no public sale ever recorded. Any figure shown is modeled from the card's own grade curve, marked as modeled, and never enters an index, a portfolio value or a deal score.

A price change is printed only on Confirmed cells. Everywhere else the field is null, never zero, so a fake move cannot render.

The gold conversion

Divide the SMV by the day's London PM gold fix, USD per troy ounce. Source: metals.dev. The Bundesbank series this page previously named is retired and is not the source of our figure; today's reference is carried as unverified until the backfill runs, and the page will say so for as long as that is true. The result is a weight in troy ounces. Grams appear only on this page: 1 troy ounce = 31.1034768 g, exact. Everywhere a buyer sees gold, the unit is the ounce.

The ounce ladder

A card at or above one full troy ounce is a Gold Tier slab. Tier assignment divides by a 30-day trailing mean of the London PM print, not the day's spot, so a badge is not granted or revoked by one loud day in the metal. Promotion requires the ratio to hold the line on 7 consecutive daily recomputes; demotion requires it to sit below 95% of the line for 14. One batch runs per day at 16:15 London, never on page load. Tier history is append-only: when a badge changes, we can show the rows that changed it.

When we refuse

Most graded cards trade a few times a month. Momentum indicators built for equities need hundreds of observations, so on most cells they are not computable. The page then prints the requirement and the count we hold, then shows a substitute labelled as what it is. A reading we cannot support is not published slowly or faintly. It is not published.

The accuracy record

From launch, every published SMV is scored against the next realised sale of its cell. The record answers one question: half the time, how far does the next sale land from our estimate? It is published monthly, split by liquidity band, with the indicator refusal rate beside it, and no month is ever restated.

Dates on this page

The modified date on every Slabpricing page reflects the last change to its data or its method, never the last deploy. The sitemap follows the same rule. This page carries a version number, and every change to a constant above is dated and kept in the revision log.