CS2 Trade-Up Contract Guide: Odds, Float Math and a Real Profit Check

A trade-up contract turns ten skins of one rarity into a single skin of the next rarity up. It is the only crafting mechanic in CS2, and unlike case openings, every part of it can be calculated before you commit: the list of possible outcomes, the probability of each one, and even the exact float of every possible result. That also means every mistake is a math mistake, and most contracts people run are losing ones for reasons they could have computed in thirty seconds.

This guide covers the rules as they stand in mid-2026 (two major updates changed them since late 2025), the probability and float formulas, and a full worked contract from The Breakout Collection priced with real asks from our July 2026 market snapshot.

The rules as of 2026

The core mechanic is unchanged since CS:GO: you hand over ten items of identical rarity, and the game returns one item of the next rarity, drawn from the collections your inputs belong to. Around that core, the current ruleset looks like this:

Contract	Inputs	Output
Standard	10x Industrial Grade	1x Mil-Spec
Standard	10x Mil-Spec	1x Restricted
Standard	10x Restricted	1x Classified
Standard	10x Classified	1x Covert
Knife/glove	5x Covert	1x Exceedingly Rare (knife or gloves)

The details that decide whether your contract is even legal:

Consumer Grade (white) skins and Contraband items cannot be used. Industrial Grade is the lowest input tier.
Inputs can come from any mix of collections, but every input’s collection must contain at least one item of the next rarity, otherwise the game will not accept it.
StatTrak is all or nothing. Ten StatTrak inputs produce a StatTrak output; you cannot mix StatTrak and normal skins in one contract.
Souvenir skins are now valid inputs. Since the May 22, 2026 update, souvenir items can be entered alongside normal ones. The souvenir attributes (gold name, tournament stickers) are stripped, and the output is always a normal-quality item.
Knife and glove trade-ups exist. Since the October 22, 2025 update, five Covert skins can be exchanged for one Exceedingly Rare item from the collection of one of the inputs. Five StatTrak Coverts return a StatTrak knife.

If you want to check what a specific collection contains at each tier, every collection has its own page under CS2 collections with the full item list and live prices.

Which skin comes out: the probability model

The outcome is picked in two steps. First the game selects one of your ten inputs at random, each with a 1 in 10 chance. That input’s collection becomes the output collection. Then the output skin is drawn uniformly from that collection’s next-tier items.

So the probability of any specific outcome is:

P(outcome) = (inputs from its collection / 10) x (1 / next-tier items in that collection)

Example: you enter 6 Restricted skins from The Breakout Collection (3 Classified outcomes) and 4 Restricted skins from The Dreams & Nightmares Collection (also 3 Classified outcomes). Each Breakout Classified has a 6/10 x 1/3 = 20% chance. Each Dreams & Nightmares Classified has a 4/10 x 1/3 = 13.3% chance. The six probabilities sum to 100%, and you can verify all of it before clicking anything.

This is the lever that separates trade-ups from gambling on cases: you choose the odds by choosing the input mix. Collections with few next-tier items concentrate probability; collections with five or six outcomes dilute it.

The float formula, and why FN inputs do not mean an FN output

The output’s float is fully deterministic. The game does three things:

Normalizes each input’s float within that skin’s own float range: (float - min) / (max - min).
Averages the ten normalized values.
Maps the average onto the output skin’s range: output = outMin + avg x (outMax - outMin).

The result then falls into a wear bracket:

Wear	Float range
Factory New	0.00 to 0.07
Minimal Wear	0.07 to 0.15
Field-Tested	0.15 to 0.38
Well-Worn	0.38 to 0.45
Battle-Scarred	0.45 to 1.00

The trap is step one. Skins have individual float ranges, and normalization stretches a capped range onto the full 0 to 1 scale. A P250 | Supernova has a range of 0.00 to 0.40, so a Factory New copy at 0.05 float normalizes to 0.125, not 0.05. Feed ten of those into a contract and any output with a full 0 to 1 range lands at 0.125: Minimal Wear, not Factory New. If you have never dealt with float ranges before, read the float and wear guide first; trade-ups are the one place in CS2 where a 0.02 difference in input float moves real money.

A worked contract: P250 Supernova into the Breakout Classifieds

The Breakout Collection comes from the Operation Breakout Weapon Case and is a textbook single-collection setup: four Restricted skins feeding exactly three Classified outcomes. All prices below are the lowest CSFloat ask in our July 2026 snapshot (the snapshot covers 33 marketplaces; CSFloat is used here because it is liquid and float-searchable).

The relevant part of the collection:

Tier	Skin	Float range	Snapshot price
Restricted (input)	P250 \| Supernova	0.00 to 0.40	FN $4.37, MW $1.78, FT $1.72
Classified (outcome)	Desert Eagle \| Conspiracy	0.00 to 0.30	FN $26.99, MW $17.86
Classified (outcome)	Five-SeveN \| Fowl Play	0.00 to 1.00	FN $99.99, MW $18.21, FT $17.42
Classified (outcome)	Glock-18 \| Water Elemental	0.00 to 0.70	FN $65.00, MW $18.20, FT $17.28

Each outcome has a 1/3 probability. Now run three versions of the same contract:

Scenario	Inputs (10x Supernova)	Avg input float	Outputs you get	Cost	Expected value	EV / cost
A: cheap MW	$1.78 each	~0.11 (norm 0.275)	Conspiracy MW, Fowl Play FT, Water Elemental FT	$17.80	$17.52	0.98
B: ordinary FN	$4.37 each	~0.05 (norm 0.125)	Conspiracy FN, Fowl Play MW, Water Elemental MW	$43.70	$21.13	0.48
C: low-float FN	premium each	under 0.028 (norm under 0.07)	all three Factory New	see below	$63.99	depends

Scenario A is what the market gives you for free: buy the cheapest liquid inputs, get mid-tier outputs, and land at 0.98 of your money before fees. After even a 2% sale fee you are underwater, and on Steam (roughly 13% between the platform fee and the CS2 fee) the same contract returns about $15.24 on $17.80 spent. Efficient markets have already priced this contract to roughly break-even.

Scenario B is the classic blunder. Factory New sounds premium, so people pay 2.5x more per input. But an average float of 0.05 on a 0.40-capped skin normalizes to 0.125, so two of the three outcomes come out Minimal Wear. You paid $43.70 for $21.13 of expected value. Half your money evaporated because of one normalization step.

Scenario C is where actual craft lives. Fowl Play has a full 0 to 1 range, so it is the binding constraint: to keep all three outcomes Factory New, the average input float must stay below 0.07 x 0.40 = 0.028. Do that and the expected value jumps to $63.99, because you unlock the $99.99 FN Fowl Play and the $65.00 FN Water Elemental. The break-even input price becomes $6.40 per Supernova before fees (about $6.27 if you sell on a 2% fee market). Sub-0.028 Supernovas cost more than the $4.37 wear-floor ask, and exactly how much more decides everything. Price each copy individually on a float-searchable market; if you can fill all ten slots below roughly $6, the contract is positive expected value. At $5.50 average, you pay $55 for $63.99 of EV, about +16% before fees.

One more thing scenario C does not remove: variance. A third of the time you win a $99.99 skin, a third you roughly break even at $65.00, and a third you take a real loss on the $26.99 Conspiracy. Positive EV pays off across many contracts, not one. If a single loss changes what you can afford to do next, size down, the same rule as in the investment guide.

Fees eat the margin

Every EV number above is a listing price, not money in your pocket. What actually lands depends on where you sell:

Steam Market: roughly 13% of the buyer price is fees, and proceeds stay locked in your Steam wallet.
Third-party cash markets: typically 2 to 8% commission, real money out.

A contract that looks +10% on listing prices is negative on Steam and marginal on a cash market. Always compute EV net of the fee on the platform where you will actually sell, and if the plan ends in real money, the cash-out guide compares the routes and their real costs.

Knife and glove contracts: what the October 2025 update actually did

The five-Covert knife contract sounds like free money and mostly is not, because the market repriced instantly. From our snapshot: the cheapest Breakout Coverts are P90 | Asiimov at $169.48 (BS) and M4A1-S | Cyrex at $165.67 (BS), so five inputs cost around $830 to $850. The output pool is the case’s 13-item Exceedingly Rare pool, and the floor outcome, a Battle-Scarred Butterfly Knife in Forest DDPAT, is worth about $422. You need well above-median luck (a Fade FN sits at $2,334) just to clear cost.

Two second-order effects are worth knowing even if you never run one:

Covert wear curves flattened. Asiimov is $170.00 in FT and $169.48 in BS in our snapshot. Battle-Scarred Coverts stopped being cheap because a contract does not care how the input looks, only what collection it is from. Cheap beater Coverts are largely gone.
The market shrank. Skin market trackers estimated the total market lost over $1.75 billion in value after the update, mostly from knife and glove prices adjusting to the new supply channel.

If your knife contract hits a Doppler, phase determines the price, and the phase is random; the Doppler phases guide explains which ones matter.

Souvenir inputs: the May 2026 change, measured

Souvenir trade-ups produced the cleanest natural experiment the CS2 economy has seen. The Cobblestone Collection has exactly one Classified, M4A1-S | Knight, and exactly one Covert, AWP | Dragon Lore. Ten souvenir Knights therefore produce a normal Dragon Lore with 100% certainty. A guaranteed Dragon Lore printer, in a game where souvenir drops for this collection no longer exist.

The market did the math within days. In our July 2026 snapshot a souvenir FN Knight asks about $2,600 on CSFloat, so ten inputs run roughly $26,000. And the float formula bites again: Knight’s range is 0.00 to 0.10, so typical FN Knights around 0.04 float normalize to 0.40, which maps to a 0.28 float Dragon Lore. That is Field-Tested, worth about $6,193. Even with impossibly perfect 0.00 floats, the FN Dragon Lore output at $11,136 is worth well under half the input cost. The “free” Dragon Lore costs about four of them. Guaranteed outcomes get priced in fastest of all.

Common mistakes, ranked by how much they cost

Averaging raw floats instead of normalized ones. Scenario B above: the single most expensive misunderstanding in trade-ups.
Valuing outcomes at lowest ask and ignoring fees. Your sale nets ask minus fee at best, and undercutting is normal. Haircut every outcome by your platform’s fee before computing EV.
Diluting a good pool with one stray input. One leftover skin from a collection with six next-tier items adds six low-probability outcomes; if they are cheap ones, it drags the whole contract’s EV down for a 10% slice of probability.
Buying inputs at Steam retail. Input sourcing is half the edge. The same skin is routinely 10 to 20% cheaper on cash markets, and our price pages show the spread across every market per wear.
Ignoring outcome liquidity. An outcome with three listings and no sales history is not worth its ask. Check offer depth on the outcome’s page before counting it as value.
Expecting to choose the result. You get one random item under the model above. Ten inputs never guarantee the specific skin you want unless the collection has exactly one next-tier item.
Quality mix-ups. StatTrak cannot mix with normal. Souvenir now can, but the souvenir attributes are destroyed, which matters because some souvenir skins carry sticker value worth more than the trade-up output.

Is running contracts worth it in 2026?

As a way to print money blindly, no. Our snapshot shows liquid single-collection contracts clustering just under break-even before fees, which is what an arbitraged market looks like. As a skill game, yes, within limits: the edges that remain are float crafting (scenario C), sourcing inputs below reference price across 33 markets, and reacting faster than repricing after Valve changes rules, which has now happened twice in nine months. All three edges come from doing arithmetic your counterparty skipped. Collection pages with per-tier float ranges and live per-wear prices across markets are the raw material for that arithmetic; the collections index is where to start digging.