How we decide what to bid on
Surplus auctions move faster than any person can read them. Here is how we filter thousands of lots down to the few worth bidding on, and where we stop.
There are more surplus lots posted every week than any person can read. Most are worthless. A few are badly underpriced. The entire problem of this business is telling those apart quickly enough to still be there when the auction closes.
We do not do that by hand anymore.
The filter
We built tooling that watches the auction sites we care about, pulls every new lot, and does the first pass for us. It reads the listing, works out what the equipment actually is (which is harder than it sounds, because the title is frequently wrong or missing the model number entirely), and then goes and finds what that equipment sells for.
Language models do a specific and narrow job in that pipeline. They are good at reading a badly written, misspelled, abbreviation-riddled government listing and working out that “LAB EQUIP MISC, 1 EA, SEE PHOTO” is in fact a specific instrument with a specific model number. That identification step used to be the bottleneck. Now it is not.
Everything downstream of identification is arithmetic, and we do not let the model do arithmetic. Comps come from real sold data. The bid ceiling is computed, not guessed. The model tells us what the thing is. It does not tell us what to pay.
I am being deliberately vague about the details here, for reasons I hope are obvious. The edge is real and I would like to keep it.
Where we stop
The filter produces candidates. Then the constraints kill most of them.
Can we physically get it? We have a car and a trailer. That defines a hard ceiling on size and weight, and a practical radius around Philadelphia. A perfect lot in Arizona is not a lot, it is a fantasy.
Can we ship it, or does the buyer collect? Some things are worth buying only if they can go in a box. Others are worth a drive.
Is the market real? This is the one that catches people. A thing can be worth $40,000 and still be a bad buy, because there are four buyers on earth and none of them are shopping this month. Thin markets need patience and a channel, and if we do not have both, we pass.
What is the honest downside? Every lot is bought untested. We assume it is broken and ask whether the price still works. If it only works when everything functions, it is not a deal, it is a bet.
What we deliberately avoid
Anything with installed alkaline batteries, because they leak in storage and take the device with them. We will bid on those for parts and not a dollar more.
Anything where the value depends entirely on software we cannot get. This one is important enough that it has its own post coming.
Anything a hobbyist would overpay for on principle. We are not here to win auctions. We are here to buy below what something is worth, and a lot with ten emotional bidders on it is not that.
The part that does not automate
Somebody still has to look at the photographs. Somebody has to notice the scuff on the enclosure, the missing cable, the serial number that dates the unit to a year when the manufacturer changed the controller board.
The tooling gets us to a shortlist of things worth thinking hard about. The thinking is still the job.