PalletMath tells people how much to bid on a pallet. The only honest way to make that claim is to publish what actually happened, so this page is computed from real outcomes and updates itself. Right now it mostly says not yet.
Being pessimistic costs you a deal you would have won. Being optimistic costs you money on a pallet you should not have bought. Those are not the same mistake, so they are not measured the same way.
Loading live results…
The product says: the max bid is the price at which even a bad pallet does not lose money. The test is per pallet and binary — once a pallet has sold through, did what it actually recovered clear the all-in cost at the bid we recommended?
A single failure is not an average slipping. It means the promise was wrong on that pallet, which is why the bar is zero and not a percentage.
How close the expected recovery came, as a median across pallets. The bar is 35%, and that is deliberately loose.
The quantity underneath is genuinely wide: hand-comped sales from a real buyer recovered between 17% and 39% of stated retail depending on the item — a 2.3-fold spread. A tool claiming ±10% on top of that would be making a promise the data cannot support. We would rather launch honest than impressive, and tighten the bar as the evidence gets better.
If you bought pallets before using PalletMath, entering them is the fastest thing you can do for the model. Every line you sold has a realized net over its stated retail, and that ratio is exactly what every recovery rate in the model is currently guessing at — today those rates rest on eleven hand-looked-up sales. One backfilled pallet can outweigh that.
What a backfilled pallet cannot do is count toward the gates above. A score produced after a pallet has already sold is not a prediction: it is run against comps fetched today for a market that has since moved, with the outcome already known. Publishing that agreement as foresight would make this page flattering and false.
So backfilled pallets are counted separately and shown, never folded in. The test needs no declaration from you — it compares the date of the first score against the date of the first sale.
No figure is published until 5 pallets have been bought and sold through by at least 3 different sellers. Below that, any percentage on this page would be one person's book with a percentage sign on it — and a seller's realized prices are their margin, which is not ours to publish.
Counts appear from the start, because "no pallets have been sold through yet" is a fact about the product rather than about anyone's business.
What this page does not claim. Scoring quality is not the same as data quality. A pallet with no product codes is priced from published rules of thumb, and the score says so at the time — no aggregate here can repair a manifest that named nothing. These gates measure whether the numbers came true, not whether they were well-founded.
Every score is recorded when it is produced — expected recovery, worst case, the bid, the all-in cost, the match rate, the flags. Sales are recorded as they clear, by hand or imported read-only from a seller's own eBay orders. This page compares the two. Nothing is re-scored: grading today's model against yesterday's outcome would be marking its own homework.