I'm a solo dev who spent the last 6 months building a sports card scanner app from scratch. Along the way I hit an interesting problem: every existing scanner on the market gives different prices for the same card, and almost none of them tell you why.
So I scanned the same 2018 Topps Update Shohei Ohtani rookie card on 5 different pricing apps. The results:
- CollX: $87
- Ludex: $74
- Card Ladder: $91
- PriceCharting: $68
- My app (Cards AI): $79
That's a 34% spread on a single card. None of the apps was technically "wrong" — they were just using different data sources, different methodologies, and different condition assumptions.
If you're building anything in this space, or just curious how card pricing actually works under the hood, here's what I learned.
The 4 reasons scanners disagree
1. Database lag
Most pricing apps don't pull live data, they query a database refreshed on a schedule. Card Ladder ($15/mo) and PriceCharting both use this model. If a card's market moves overnight, the database doesn't reflect it until the next refresh. Could be hours. Could be weeks.
Apps that scrape live data (mine included) avoid this lag but pay a different tax: occasional empty results when scrapers fail, plus weird outliers from edge-case eBay listings.
2. Asking prices vs sold prices
This is the biggest one. Many "pricing" apps display asking prices (what sellers list cards at), not what they actually sold for. Asking prices on eBay run 20-50% above true market value because sellers are optimists and most listings never sell.
Card Ladder claims to use sold data. PriceCharting blends both. Some smaller scanner apps quietly use asking prices because they're easier to scrape.
CollX uses a "market average" that includes asking prices. Their prices run ~22% above actual eBay sold values on average. Ludex weights completed sales more heavily and runs ~8% above. Cards AI is built around eBay sold-and-completed listings only.
3. Condition assumptions
Every app makes a guess about your card's condition. Most assume "raw, played-with condition" by default. None do this reliably without you confirming the grade manually.
A 1989 Upper Deck Ken Griffey Jr. rookie is a $30 card raw and a $1,500 card in PSA 10. Same card. Different number.
4. Geography
For high-end cards, the same card sells for different prices on US eBay vs Japanese Yahoo Auctions vs European auction houses. Live scrapers that pull from multiple regions average across them. Database apps that only track US sales miss the gap.
The honest test: how to verify any pricing app
You don't need to take any app's word for the value of your card. Here's the 2-minute manual check:
- Open eBay
- Search the card name exactly (year, set, player, parallel)
- Add the filter: Sold Items
- Sort by Recently Sold
- Look at the last 5-10 results, throw out obvious outliers
- The middle of the remaining range is your real market value
If any pricing app's number matches this, trust it. If not, the app is either lagging, using asking prices, or making a different condition assumption.
What I built differently
Cards AI scrapes real eBay sold listings live on every scan. You don't get an averaged "market price" — you get the actual last 15 sold listings with dates, prices, and a link to each one on eBay.
If the number looks wrong, you verify it yourself in 10 seconds.
Stack notes for the curious
- React Native / Expo on the frontend
- Gemini 3.1 Flash-Lite for card identification (multimodal, falls back to 2.5 Flash on rate limits)
- Apify scrapers for pricing (caffein.dev/ebay-sold-listings for sports, parseforge/tcgplayer for TCGs)
- Supabase edge function orchestrates the pipeline + writes results
- RevenueCat for subscriptions
- Cost per scan: ~$0.008 all-in (Gemini + Apify + edge function compute)
- $39.99/year flat pricing, possible only because cost-per-scan is so low
Lessons for other indie hackers in this space
The unit economics of "scrape on every scan" only work if your pipeline is cheap. Mine is ~1 cent per scan. That gave me room to skip the monthly subscription play entirely.
Transparency is the moat against bigger competitors. Card Ladder is opaque. CollX has marketplace fees. Ludex is closed about methodology. I show every comp. People notice.
AI grading is mostly oversold across the entire market. I built it because users asked, but I'm honest in the app: no AI grade is a real grade. It's a pre-screening tool for deciding which cards to submit to PSA. The market only pays for actual PSA slabs.
On the question of "should you use my app vs others"
Honest answer: depends what you're optimizing for.
- Casual collector? CollX has a better UI, a marketplace, and a usable free tier. Use them.
- Modern set builder? Ludex catches parallels 85% of the time vs CollX's 65%. Pay the $20/mo if you collect refractors.
- Portfolio tracker? Card Ladder's analytics are unmatched.
- Want to verify every price? Cards AI. That's the whole pitch.
You can read my full honest comparison of CollX vs Ludex with the 50-card test data on my blog if you want the depth:
https://www.cardsaiapp.com/blog/collx-vs-ludex
Cards AI is $39.99/year (about $3.33/month) with a 7-day free trial. iOS only for now. Built solo from Morristown, NJ. Real eBay data. No subscription tricks.
If you're building in any data-pricing space (sneakers, vinyl, watches, art), the same patterns apply. Live scraping with transparency beats opaque database averages. Almost always.