How the AI Decision Engine works

We tell you whether to hold, sell, or grade — and we'll show our working.

CardEtra is honest about what an AI can and can't know. No fake precision, no invented sale prices. Conservative ranges, explicit confidence, plain-language reasoning.

1. What you give us

Front and back photos

Two clear photos taken on your phone or scanner. JPG, PNG, or WEBP, up to 10 MB each.

Light metadata

Player name, year, set, rarity, and an optional note. The AI uses these to identify the exact parallel.

2. What the AI looks at

  • Player and year
  • Set, parallel, and serial number
  • Visible centering, corners, edges, surface cues
  • Whether the card is raw or already in a graded slab

3. What we return

A clear decision

HOLD, SELL, or GRADE — one of three, never a vague verdict.

A conservative GBP range

Deliberately wide when comp data is sparse. We never invent a single sale price.

Confidence level

Low, Medium, or High — reflecting how many close comparable sales likely exist for this exact parallel.

Plain-language reasoning

A few honest sentences explaining the recommendation. For raw cards, an optional grading note.

Sample insight

HOLD

MEDIUM CONFIDENCE

£90 – £140

Based on what we can see, this card has limited comparable sales — the range is intentionally wide. Demand for the player is strong, so holding is reasonable unless a notable offer appears.

4. Where we're honest about limits

  • No live market comps yet — values come from the model's training data and the metadata you provide.
  • No fake precision — ranges are intentionally wide for low-pop cards.
  • AI insights are illustrative, not financial advice or a guarantee of value.

5. What we don't do

We don't pretend to know the exact next sale price, we don't act as a broker, and our insights are not financial advice. Use them as a thinking aid alongside your own judgement.

Try it on a card you actually own.

Sign up and upload a card to see a real insight. No fake numbers, no broker tricks.