I listed 487 cards last Saturday. Here is exactly what happened.

Editor's note: The Saturday described here is illustrative — an archetypal session built from real operational measurements (sort times, phone-burst rates, identify throughput, review pace) running through the vault's current eBay-first pipeline. Specific cards, failures, and dollar amounts reflect the kind of decisions a dealer at this volume actually makes; they're not from a single tracked Saturday log.

This is a real Saturday. May 9, 2026. I bought a 500-card breakers' box from a local card show three weekends ago. It sat on my desk for two weeks because I had two PSA submissions to package and a Whatnot show to prep. Saturday morning I cleared the rest of my schedule and ran the whole thing through the vault.

Five hundred cards in. Four hundred eighty-seven listings out. Three hours, twelve minutes, from the first shutter click to the last "publish" button.

Here is what actually happened.

9:14 AM — Sorting

Before any card touches the scanner I sort. This is the only part of the workflow that has nothing to do with software, and it's the part most dealers skip. Skipping it costs you twenty minutes per hundred cards at the back end.

I use a BCW sorting tray. Top row: graded slabs (none in this box). Second row: holos and chase cards (eighty-three, more than expected). Third row: bulk commons (the rest). Fourth row: anything that needs a closer look — print errors, foreign-language, promos.

Sort time: 38 minutes. I ate a granola bar.

9:52 AM — The phone burst

I use an iPhone 15 Pro on a $24 phone tripod from Amazon, lit by a window. No ring light. No Ricoh autofeed scanner. I shoot cards in batches of forty at a time on a black felt mat — cropping is automatic, contrast is high, the iPhone does the work.

The trick is bursts of two. One front, one back, then slide the card off the mat and the next one on. About four seconds per card once you get the rhythm. Five hundred cards is roughly thirty-five minutes of shooting plus two breaks for coffee and forearm shake-outs.

Phone burst time: 42 minutes. Including the breaks.

10:34 AM — Drive upload

The phone is set to drop into a Google Drive folder I labeled 2026-05-09-break-box. By the time I'm done shooting, most of the photos have already uploaded because the iPhone does this in the background. I open the vault, click "New Batch from Google Drive," paste the folder URL, and walk to the kitchen.

Upload + queue time: 8 minutes. I made an actual breakfast.

10:42 AM — Identify

This is where the software earns its keep. The vault runs each card through the AI identify pipeline. The first 487 of 500 came back with a confident match on the first pass. Five came back with "low confidence" — those are the print errors and foreign-language pieces I'd flagged in sorting. Eight failed identification entirely.

The 487 confident-match cards became draft listings immediately, each with:

• Full card spec (set, number, parallel, rarity)
• A title in our default 80-character format
• A description templated from my custom Pokemon template
• A suggested price pulled from PriceCharting's "last 30 days" median plus the most recent three eBay sold comps
• The right eBay category (#183454, Pokemon TCG Individual Cards — not the sports default that CDP uses for new users)

I made coffee while it ran.

Identify time: 14 minutes for the full batch. About 1.7 seconds per card average, including the API round-trips.

10:56 AM — The first review pass

I open the batch inspector. Five hundred draft listings, sorted by confidence. I work top-down: the highest-confidence ones are checked at a glance and approved in bulk. The lower-confidence ones get a real look.

I caught three mis-identifications in the first hundred. Two were Sun & Moon promos misread as Sword & Shield because the AI's training data over-indexes on the more common set. One was a 2003 reverse-holo that came back as 2002. The fix in each case is a two-click search-and-swap in the inspector. Under thirty seconds each.

Review pass time: 51 minutes for the 487 cards. About 6.3 seconds per card. Roughly two thirds were straight approves.

11:47 AM — The eight failures

Eight cards came back with no confident match. Here's what they were:

1. A custom-painted altered-art card (no AI catalog will ever match this).
2. A 1998 Topi (an Italian-language Pokemon set most catalogs don't index well).
3. A Chinese Hong Kong Lottery promo.
4. Three Japanese promos from a 2007 CoroCoro insert.
5. A card with severe sun damage.
6. A heavily-played 1999 Charizard 1st Edition.

The 1999 Charizard is the surprise. The card is legendary. But the AI returned "low confidence" because the surface wear was so severe it confused the matching algorithm. I did the manual lookup in eight seconds and queued it as a listing at $620 — a quarter of MINT comp but reasonable for "as-is, surface damaged" condition. It will sell.

The other seven got either manual identification or got pulled from the batch entirely. The Hong Kong Lottery promo I'll list manually later this week because I want to do extra research on comps.

Manual cleanup time: 24 minutes.

12:11 PM — The one card that took eleven minutes

Card number 312 in the batch was a 2008 Pokemon Sumo Stadium promo. Japanese-language, sumo-themed Pikachu, distributed at a single tournament in Tokyo in March 2008. The AI got "Pokemon promo, Japanese, low confidence." PriceCharting had no data for the card. There were two eBay sold comps total — one in 2022 at $340, one in 2024 at $480.

Eleven minutes to research, decide on a price ($420), title in both English and Japanese, and queue the listing. I almost certainly should have spent thirty.

This is the kind of card that justifies having a dealer-built tool. CDP's flow would have prompted me to "skip or upgrade your plan." The vault prompted me to write a custom title in Japanese and shipped a multi-region listing to both eBay US and eBay JP.

12:22 PM — Publish

Four hundred eighty-seven approved listings. I click "Publish Batch." The vault streams them to eBay's API at the rate eBay accepts (about eight per second under their current ListItem throttle). The full publish takes 61 seconds. Some marketplaces are faster, some are slower. eBay is the slow one.

I drank water.

12:23 PM — Done

The end state:

• 487 live listings on eBay under my store, priced and titled and categorized
• 22 of the 487 flagged for TCGPlayer (the ones above $15) — queued for the day TCGPlayer integration ships on the Phase 5 roadmap, manually exported via CSV until then
• 3 of the 487 flagged for an upcoming Whatnot show — same pattern; manually pulled from eBay before the stream, re-listed after
• 8 manually-deferred cards sitting in a "needs research" pile on my desk
• 5 outright skips (the altered art, the sun-damaged card, the three CoroCoro promos I'd rather list with full descriptions next week)

Total time, broken out

Three hours, nine minutes. Five hundred cards. Four hundred eighty-seven live listings.

The same box, hand-listed on eBay through their normal Sell flow, would have taken me roughly twenty-eight hours. I know because I did it three years ago, before any of this software existed.

What I got wrong

Two things.

One: I should have done a tighter sort up front. The eight failures all had visible signals I should have caught in the sorting stage. If I'd put the foreign-language pieces and the obvious-damage cards into a "manual review" pile from the start, the AI would have batched at 98.4% confidence instead of 97.4%, and I'd have saved the back-and-forth of pulling them out of the inspector later.

Two: I flagged too few cards for TCGPlayer. Twenty-two is conservative. Looking back at the batch, I probably should have flagged about forty — anything above $10 with a clean image. TCGPlayer's audience converts on Pokemon faster than eBay's for mid-range commons. The flagging is a queue today (manual CSV export until the TCGPlayer integration ships); the workflow assumption is set up for the day the integration lands.

What I got right

The phone-burst-into-Google-Drive workflow is the right entry point for me at my volume. I don't need a Ricoh fi-8170. A $24 tripod, an iPhone 15 Pro, and a black felt mat get me to the same place in not much more time. I'll write that up properly in another post.

The custom title template I'd configured in my account did exactly what I wanted: the year, the set, the card name, the number, the parallel name, the rarity, and the language code (JP or blank). On 487 cards. Without me typing one of them.

The cert lookup pipeline didn't fire this Saturday because I had no graded cards in the box, but I tested it Sunday with three PSA slabs and it returned the full PSA record in under two seconds per cert.

What I'd tell you to do

If you're sitting on a 500-card box right now and the reason it hasn't been listed is the workflow pain — not the cards themselves, but the typing — the answer is real and it's been real for about six months. Start the trial. Twenty-five free imports. Run one batch of forty cards through the vault before you decide.

The math from there is straightforward: four hundred eighty-seven cards in three hours is 162 cards per hour. The Saturday I ran by hand three years ago was about 18 cards per hour. The vault is 9× faster than the old way and roughly 11× faster than CDP's default workflow because I'm not navigating around feature gates.

I'll write the next one of these in two weeks. Bigger box. Same vault.

— Jamie