How Consistent Is DCM Optic? Our Repeatability Data, Method and Limits

Any grading service, human or AI, should be able to answer a simple question: if you grade the same card again, do you get the same answer? This page is our answer for DCM Optic, with the numbers, how we measured them, and the limits of what they show.

The short version: every DCM grade is the median of three independent evaluations of the same photos. On a sample of 600 cards graded between September 8 and October 7, 2026, all three evaluations landed on the same final grade for 68.5% of cards and within one grade of each other for 89.3%. Because the median is published, a single outlying evaluation cannot set the grade by itself.

What "consistent" means here

There are three different questions hiding inside "is it consistent?", and they have different answers.

1. Do the three evaluations inside one grade agree? This is measured on every card we grade, and it is the bulk of the data below.
2. If the same photos are graded again from scratch, does the grade come back the same? We test this with repeat runs on a fixed set of cards.
3. If you take new photos and grade again, does the grade come back the same? This depends heavily on the photos, which is why every DCM report carries an image confidence letter. We do not yet publish a measured figure for this one, and we say so below.

Result 1: agreement between the three evaluations

Sample. 600 cards drawn at even intervals from all 6,842 cards graded on dcmgrading.com and the DCM app between September 8 and October 7, 2026. Every card in the sample had all three evaluations recorded. No cards were excluded for their grade, category or photo quality.

Measure. For each card we took the final grade from each of the three evaluations and measured the spread between the highest and the lowest.

Bar chart of how far apart DCM Optic's three passes landed in a 600-card sample: 0 points 68.5% (411 cards), 1 point 20.8% (125), 2 points 6.0% (36), 3 or more points 4.7% (28)

Looked at one evaluation at a time, 12.1% of individual evaluations differed from the median of their card, and 3.1% differed by two grades or more. That second number is the reason DCM never grades from a single evaluation: one run on its own would be noticeably less stable than the median of three.

Disagreement was spread fairly evenly across the four categories. At least one evaluation differed on centering for 21.3% of cards, corners 21.3%, edges 22.0% and surface 18.3%. Most of those differences are one point on one face, and many never reach the final grade, because the final grade is set by the lowest category.

Result 2: re-grading the same photos from scratch

In August 2026 we re-graded 11 cards three times each with the full production pipeline, on the same photos, and compared the runs with each other and with hand-verified reference grades.

This test was designed to compare two image settings, and the setting with the more stable result is the one DCM uses today. It is a small sample, so read it as a direction rather than a precise rate.

What we do to keep grades stable

• Three evaluations, median result. A single outlier cannot move the grade.
• A stricter bar for a 10. At least two of the three evaluations have to find the card flawless, and when the third evaluation cites a real flaw, the card is held at 9.
• Magnified inspection. Corners, edges and surface regions are re-examined on close crops so small defects are judged on detail rather than on a full-card view.
• Image confidence on every report. When the photos limit what can be seen, the report says so with a letter from A to D instead of guessing.
• A published, versioned standard. The rules live at the DCM Grading Standard, and we test engine changes against a set of hand-verified cards before we release them.

What these numbers do not show

We would rather you hear the limits from us.

• Agreement is not accuracy. Three evaluations can agree on the wrong answer. Agreement measures stability, and the hand-verified test above is the closer measure of correctness.
• New photos can change a grade. A different angle, light or camera can reveal or hide a defect. We have not yet published a reshoot study, and until we do, we make no numeric claim about it.
• A newer engine version can grade a card differently. Every report records the engine version it was graded with.
• No claim about other graders. DCM grades against its own published standard. These numbers say nothing about the grade PSA, BGS, SGC, CGC or anyone else would give.
• Photos cannot show everything. Authenticity, alterations and some surface defects are beyond what a photograph can establish. See grading limitations.

How to check a grade yourself

Every public DCM grade has a report page with the photos that were graded, the four subgrades for each face, a written reason for every deduction, and the image confidence letter. If a result looks wrong, you can request a review from the card page. The AI card grading accuracy page explains the method in more detail, and how DCM Optic produces a grade walks through a real card step by step.

Method note: agreement figures come from DCM production grading records (sample of 600 of 6,842 cards graded September 8 to October 7, 2026, selected at even intervals). Re-grade figures come from an internal test run in August 2026 on 11 cards with hand-verified reference grades, three runs per card. Last updated October 2026.