How We Calculate Profit: What Counts and What We Threw Out
An open methodology instead of promises. Which marketplaces do not count as a seller and why, how stub prices are filtered, what the "N against one" rule does, and where the line runs between a listing and a real sale — with numbers for how many deals each rule removed.
Why publish a methodology at all
Every price comparison service shows a "profit" column. Almost nobody explains where it comes from — which is convenient, because the prettiest numbers in such tables usually mean an error rather than a find.
We decided to go the other way and publish the rules together with the price of each one: how many deals the rule removed and how much "profit" disappeared with them. The numbers below come from our own measurements on the live database, not from general reasoning.
The main thing to know before the details: the table currently holds about 219556 rows. Every rule below reduced that number, and we consider that an improvement rather than a loss.
AK-47 | Redline
★ Karambit | Doppler
★ Sport Gloves | Pandora's Box
Kilowatt Case
Sticker | Crown (Foil)Rule 1. Trade sites do not count as a seller
On some platforms the price is not money but a valuation of an item for swapping it against another item. You cannot sell there for real money, so treating such a price as revenue is incorrect.
The measurement showed the scale: letting trade sites into the calculation as the selling side produces 41719 deals and one and a half million dollars of paper "profit" — more than all other marketplaces combined.
So the selling side is switched off for trade sites. The buying side stays: you really can buy there for money, and that is an honest transaction.
Rule 2. The overpriced seller — "N against one"
The second layer of pretty numbers is a single marketplace priced noticeably above everyone else. The gap looks like profit but means that this one price is inflated.
The evidence here is not an absolute threshold but the agreement of independent marketplaces with each other. If an item has at least three sources and the selling price is twice the median of the rest, that marketplace does not count as a seller for this item.
| Number of sources | What we do | Why |
|---|---|---|
| one | no judgement | nothing to compare against |
| two | no judgement | the median degenerates |
| three or more | compare with the median of the rest | agreement of several is evidence |
Note the boundary: the rule judges an item, not a whole marketplace. The same platform can be inflated on one skin and perfectly normal on another.
Rule 3. Stub prices never reach the database
Sometimes a source returns a physically impossible price: a case at five cents next to a price in the hundreds of thousands, an item at three dollars next to eleven thousand. That is not a market but mixed-up units or a broken response parse.
Such prices are cut off at import, before they enter the calculation, and the filtering is printed to the log — silent filtering is more dangerous than the stub itself, because it hides the moment when normal goods start being lost.
We do not consider this defence finished: the latest check found 49 names where the spread between marketplaces exceeds a hundredfold, and two names where it exceeds a thousandfold. That is open work, not a solved problem.
Rule 4. A dead source means a frozen price
If a marketplace stops responding, its last prices stay in the database and look current. On a falling market such a frozen price automatically turns into a "great deal" and floats to the top.
So dead sources are kept in a single list, cut off at import, excluded when a price is chosen, and shouted about in the collection cycle log. Four places — because one is never enough: a ninth reader of the data always turns up.
Rule 5. Profit is counted from listings, and we say so plainly
This is the most important limitation, and almost nobody spells it out.
An item has two prices: what it is offered at and what someone will take it for right now. The second is available from just one source out of twenty and covers about 9.4% of the catalogue. So for 91.1% of rows in the calculation, the selling price is a listing.
What that means in practice: the displayed profit is correct for a patient sale and overstated for a sale today. By our measurement an instant buyout returns a median of 81% of the listing price.
How that looks on live data right now:
What always enters the calculation
The selling marketplace's fee — deducted from the trade amount, not from the profit. The buying marketplace's fee and terms, where they change the effective entry price. Trade hold: a deal with a delivery delay is flagged, because the price can move during it. The freshness of both prices: a deal is live only as far as the older of the two prices is fresh. Order book depth where it is known: a single buy order is not a market.
What we deliberately do NOT do
We do not tune the top of the list. Beautiful percentages in the first rows almost always mean a data defect. We would rather remove a row than keep it for looks.
We do not promise a deal will execute. Between the calculation and your trade stand time, the hold, and another buyer who may get there first.
We do not count as profit what cannot be withdrawn. Money inside a closed ecosystem and money on a card are different things, and they cannot go into one column.
A practical way to test any comparison service, ours included: take a row from the top of the list and check how many independent marketplaces agree with the selling price. If one agrees and the rest are noticeably cheaper, you are looking at an inflated seller, not a find.