Every pre-grade is an estimate. This page measures how close those estimates come to the grades collectors actually receive, company by company, and sets out the method in full.
GradeMate measures its accuracy by comparing, for each grading company, the grade predicted from photos with the grade users report receiving from that same company for the same card. A company's figures (mean absolute error, share of results within half a grade, share of exact grades) are published only once 30 eligible results have been collected; below that, the table shows how many have been collected so far. The grades received are self-reported by users and are not verified against certificates.
01 — Results
The predicted grade compared with the grade users report receiving from the same company. Recalculated every day.
No grading company has reached 30 results yet, so GradeMate does not publish an accuracy rate. Each row shows how many results have been collected so far.
PSAResults · 0
Collecting data — 0 of 30 results
BGSResults · 0
Collecting data — 0 of 30 results
CGCResults · 0
Collecting data — 0 of 30 results
SGCResults · 0
Collecting data — 0 of 30 results
PCAResults · 0
Collecting data — 0 of 30 results
TAGResults · 0
Collecting data — 0 of 30 results
HGAResults · 0
Collecting data — 0 of 30 results
02 — Key facts
03 — Details
A result is one saved card whose grade has come back from the grading company. Each rule below removes a case where the comparison would say nothing about the estimate.
The card must have been saved straight from an analysis run on GradeMate, which the server records at the time of the analysis. Cards saved before that record existed are left out: nothing proves which prediction they came from.
Each prediction targets one company's scale. A grade received from another company stays in your collection but is left out of the figures: comparing a PSA prediction with a CGC grade would measure the gap between two scales, not the error of the estimate.
When the photos do not allow a reliable estimate, the report says so and gives no grade. With no prediction there is nothing to compare, so these cards never count.
Results follow the same rules as the +1 credit offered for entering a grade, so that every grade in the figures has gone through the same checks. A card whose detailed report was never unlocked is not counted.
Grading takes weeks. A grade entered within 14 days of saving the card cannot be the grade the company returned for it, so it is ignored. Changing a grade later dates it again, and the 14-day rule applies to the new date.
A card counts once, whatever the number of times its grade is edited. A card deleted from a collection leaves the figures at the next daily calculation.
04 — Details
The number of cards that meet every rule above for that company. Read it before any other column: the same percentage means more on 300 results than on 30.
The average gap between the predicted and the received grade, in grade points, whichever the direction. 0.50 means the prediction is off by half a grade on average; 0.00 would mean every prediction was exact.
The share of results where the received grade is no more than half a point away from the prediction: a predicted 8.5 followed by a PSA 8 or a PSA 9 counts; a PSA 7 does not.
The share of results where the two grades are identical. It is the strictest measure: on a scale in half points, a prediction can miss the exact grade while being only half a point away.
Predicted minus received, averaged with its sign. Positive means the estimates run high, negative means they run low. A value close to zero next to a large mean error means the misses go both ways.
The dates of the first and the last result counted, so you can tell whether a company's figures describe last month or last year.
05 — Key points
Publishing a measurement means publishing its limits. These are the ones that matter most here.
Users type the grade they received. GradeMate does not ask for the certificate number and does not check the company's database, so a typing error or an invented grade is not detected. The rules above only remove grades that cannot be real.
Collectors may be more inclined to enter a grade when it matches the prediction, or when it disappoints. The figures describe the results that were entered, not every card that was analysed.
Grading involves human judgement, and the same card submitted twice can come back with different grades. Part of any gap belongs to the grading, not to the estimate.
The estimate sees what the photos show. A defect that a photo does not reveal, such as a faint surface scratch, shows up in the grade received but could not be in the prediction.
The grade field accepts 1 to 10 in half points, and PSA has no 9.5. PCA's 10+ is entered as 10, so it is compared as a 10.
Even at 30 results, a single card shifts a percentage by more than three points. Read the number of results next to each figure before comparing two companies.
06 — Steps
Every grade entered makes the figures more useful for the next collector deciding whether to submit a card.
Run the analysis on GradeMate, choosing the company you plan to use, and save the card to your collection from the report.
Send the card to the company you chose at analysis. A grade from another company is kept in your collection but not counted here.
In My collection, type the grade on the card's tile, in half points. You can change or clear it later.
From 14 days after saving, the first grade entered for a card with an unlocked detailed report earns one credit, once per card and at most 3 a month.
07 — Questions
It depends on the grading company, and GradeMate publishes a company's figures only once 30 eligible results have been collected for it. The table at the top of this page shows the current figures, or how many results have been collected so far.
No. The grades received are entered by users and are not checked against the grading company's certificate or database. The rules on this page remove entries that cannot be real results (entered too early, from another company, with no prediction to compare), not entries that are false.
Below a few dozen results, one or two unusual cards move a percentage by several points, and a rate calculated on ten cards would be noise presented as a measurement. GradeMate has set the threshold at 30 results per company; until a company reaches it, its row only shows its progress.
The prediction is made on the scale of the company chosen at analysis, and scales differ: a PCA 10 is not the equivalent of a PSA 10, for instance. Comparing across companies would mix the error of the estimate with the difference between two scales.
No. The report keeps the prediction made at analysis. The grade received is stored next to it, with the company that awarded it and the date it was entered. You can change or clear it at any time; a credit already received stays yours.
The figures count cards, not people: a collector who grades twenty cards contributes twenty results. Credits for entering grades are capped at 3 a month, which limits the incentive to enter grades for the sake of it.
08 — Read next
Then submit it, enter the grade you get back, and add your result to this page.