Valuing a lot before you bid

Published 17 October 2026 by CompsAPI

Estate lots, storage units and liquidation pallets all pose the same question under time pressure: what is this box worth, and what should I bid? Here is an actual ten item lot priced from completed eBay sales, including the row that nearly cost a fifth of the valuation.

The lot, priced

Ten lines, quantities included, three years of completed sales, medians with the 25th to 75th band and the sale count behind each.

Every row returned over a hundred sales. Nothing was flagged thin. A script would have called all ten of these confident and moved on.

The row that is wrong

Look at the Featherweight. Median $39.99, but the 75th percentile is $237.50. A distribution where the upper quartile sits six times the middle is not a price, it is two different markets stacked on top of each other.

Sorting the cheap end shows what they are: a record book at $9.35, a metal bobbin at $9.95, an instruction manual at $21.66. Sellers title those "Singer Featherweight 221 Sewing Machine Instruction Manual", so every word of the search matches and the accessory market, which is far larger by unit count, owns the median.

The machine is in there. It is just outvoted.

Two obvious fixes, both of which failed

This is worth showing because the instinct is right and the result is not.

Requiring the word "machine" and used condition moved the median to $100.00 over 110 sales. Better, still wrong, and for an obvious reason in hindsight: the accessory listings contain the word "machine" too.

Adding minus words for manual, book, bobbin, attachment, foot, controller, case and parts made it worse: $39.59 over 235 sales, with motor belts at $11 now at the bottom. Exclusions make eBay widen the match rather than narrow it, so you trade one set of accessories for another.

What worked: a price floor

Set a floor the accessories cannot reach.

q=singer featherweight 221&min_price=150&days=1095

That returns 34 sales, median $300.00, 25th to 75th from $219.99 to $400.00, and the cheapest rows are now genuine machines: a 1933 school model at $220, a 1940s AL series at $235.

Use this honestly. A price floor does not discover a value, it separates two populations you already have evidence for, and the evidence here was the bimodal spread in the first result. Applying a floor to an item with a single tight distribution does nothing but delete the cheap half and flatter you. The floor is a scalpel for a specific, visible problem, not a way to make a number bigger.

What that one row was worth

One row in ten, and the lot moves by $260, about 21%. In a room where you are bidding against someone who checked, that is the entire margin.

It cuts the other way just as hard. If a lot's value is concentrated in one or two items, as it usually is, then being wrong about those items is being wrong about the lot, and the eight easy rows provide false comfort.

A checklist for a lot

Price the whole list first, then audit the spreads. Sort by the ratio of p75 to median and look at the top three. A wide spread is where a mistake can hide.

Weight your attention by value, not by row count. The items worth the most deserve a manual look at actual listings. The $20 rows do not.

Widen the window before loosening the query. Three years on anything collectible, and treat fewer than eight clean sales as a direction rather than a number.

Remember these are gross sold prices. Selling fees, shipping, and the time to list ten categories of thing all come out before you see any of it. Whatever your fee and shipping assumption is, apply it before you bid, not after.

Check the negotiated sales. On accepted Best Offers eBay shows the asking price, not what was paid, and we measured that at a median 14.9% overstatement across 5,859 sales. Higher value lot items are exactly where offers happen.

Doing it at speed

The pricing run above is the bulk script pointed at a ten line CSV: one request per item, three year window, output straight to a spreadsheet. Ten items is ten requests, so the free tier covers ten lots a month with no card.

What the script cannot do is notice that a Featherweight is really two markets. That still takes a person reading the spread, which is why the spread is in the output at all.