How Susbid checks an auction

Principles

• Only what’s publicly visible. Every sign is something any buyer could check on eBay by hand. Susbid just does it faster.
• Signs, not verdicts. A sign is “present” or “absent”, a pattern is “observed”. Susbid never says a seller is shill bidding, and never gives a risk score. The only summary is a count, such as “1 of 3 signs present”.
• Uncertainty is visible. Thin data says “not enough data”, never “no signs”. Masked bidders that match are “likely the same bidder”.
• No lists of stores. No leaderboards, no “most flagged”, no indexed store pages. A check’s result appears only to the person who ran it, and a store report only to someone who looks the store up by name.

Where the data comes from

Susbid never fetches eBay pages from its own servers. Everything comes from what you capture in your own browser or app: a screenshot or PDF of the auction’s bid history page and, optionally, of individual bidders’ detail pages.

• Bid history page: every bid with its amount and time, each bidder’s masked name (such as a***b) and feedback score, and a section for retracted or cancelled bids.
• Bidder detail page: that bidder’s last 30 days, including total bids, items bid on, the share of their bid activity with this seller, and retractions over 30 days and 6 months.

A vision model transcribes the capture. It copies values; it doesn’t judge them. You then see everything it read, with anything it wasn’t sure of highlighted, and the report doesn’t run until you’ve confirmed or corrected each highlighted value or marked it unknown. The rules that decide whether a sign is present are ordinary code, the same for every check.

The signs and their thresholds

These are starting thresholds. A sign only belongs on the list if it is uncommon in ordinary auctions: the target is that each appears in fewer than 10% of ordinary trading card auctions. Thresholds are being tested against a hand-recorded baseline of ordinary card auctions, and will be tightened, or a sign dropped, where they flag too much ordinary bidding. Changes are noted here.

How bids are ordered

eBay sorts the bid history by amount. In an ascending auction accepted amounts only go up over time, so Susbid orders bids by amount rather than by the times in a screenshot, which avoids time-zone guesswork. Times are shown as eBay displays them.

What counts as a “small” bid

For small consecutive bids, a step is small when it is no more than three of eBay’s bid increments above the bidder’s previous bid (see the increment table), and each bid must be answered at once by the leader’s automatic bid, at most one increment higher.

Same bidder or not

eBay masks bidder names to their first and last characters. Two bids are treated as the same bidder only when the masked name and the feedback score both match, and even then the report says “likely”. Susbid never tries to unmask anyone.

Feedback is context, not a sign

Low or zero feedback is common among genuine buyers, because buyers often don’t receive feedback. The report shows each bidder’s score but never counts it as a sign.

What the research says

The UCI Shill Bidding Dataset covers 807 completed eBay auctions of the same phone from 2017, with 6,321 bidder-in-auction records. Researchers labelled 675 records (10.7%) as suspicious using clustering; those are statistical labels, not confirmed cases. Two patterns stand out: most ordinary bidders score near zero on concentrating their bids with one seller and on repeatedly outbidding in small steps, so those are distinctive. Bidding early and rarely winning are common among ordinary bidders, so on their own they are not.

That is why Susbid leads with seller concentration and retractions, and treats timing and win rate as weak on their own.

Store reports

Every confirmed check also feeds its store’s report, which pools the auctions anyone has checked from that seller. Susbid never monitors a store in the background, so coverage follows usage: popular stores fill in, small stores stay thin, and every store report says how many auctions it rests on and over what dates.

• Independent contributors. Store-level signs are only evaluated once checks come from at least three different people, so one person can’t build a store’s pattern alone.
• Cross-checked captures. When two captures of the same auction disagree (a different bidder at the same amount, a bid that vanished without a retraction, a different seller), neither counts toward the store until a person has reviewed them. Captures taken at different times may differ in the number of bids; that’s expected.
• Likely the same bidder. Across weeks a bidder’s feedback grows, so masked names match when the scores are within a few points (or 2%). That makes a match likely, not certain.
• No lists. A store report appears only when someone looks up that store by name. There is no index of stores.

Live auctions

A check of a running auction reflects the bids so far, says so, and shows the time left when captured. eBay is testing extended bidding in some categories, where a bid in the final minute resets the clock, so an auction can run past its scheduled end. The bidder-detail signs are usable mid-auction because they cover each bidder’s last 30 days.

What is stored

Screenshots are deleted as soon as they are read. Susbid keeps only what was read from them: masked bidder names, feedback scores, bid amounts and times, retractions, and bidders’ 30-day summaries. See the privacy policy.

Limits

• Susbid sees only what eBay shows buyers. eBay sees account links behind masked names; Susbid can’t and won’t.
• A capture can be cut off or misread. That’s why you confirm what was read before the report runs.
• v1 covers eBay US. Other eBay sites may show bid histories differently.