Every few weeks someone in a CS2 Discord posts a crash system that "never loses." Auto-cashout at 1.5x, double the bet after every bust, sit out until you've seen three red rounds in a row. It looks clever for an evening or two. Then a long losing streak shows up, as it always does, and the screenshots stop.
These systems don't fail because the player got unlucky. They fail because of one small number, the house edge, and because a crash round has no memory. This post goes through how a crash point is actually produced, the myths we see most often, and the math behind each one, with tables you can check on a phone calculator. Most examples use a 5% house edge because it keeps the numbers clean. The real figure depends on the site and should be listed on its fairness page.
How a crash point is actually decided
Crash looks like a live event. The line climbs, chat panics, someone bails at 1.2x and someone else rides it to 40x. In reality the crash point was fixed before the round started.
On a provably fair crash game the result comes from a hash. A lot of sites still use a design descended from the old Bustabit script: the operator generates a long chain of hashes in advance (often millions), where each hash is the SHA-256 of the next one, and then plays the chain backwards. A public value that nobody could know when the chain was created, such as a future Bitcoin block hash, gets mixed in as a salt so the operator can't hand-pick a friendly chain. Each round's hash is converted into a number, and that number is the crash point.
Two things follow from this design, and they matter for every myth below:
- The crash point doesn't depend on how many people bet or how much they bet. It existed before anyone clicked.
- The next crash point can't be worked out from previous ones. Doing that would mean reversing SHA-256.
If you want the full mechanics of server seeds, client seeds and verification, our provably fair explainer covers it step by step.
The one formula worth memorizing
For a crash game with house edge e, the chance that a round reaches at least multiplier m is roughly:
Chance of reaching m ≈ (1 − e) / m
With a 5% edge, the chance of reaching 2x is 0.95 / 2 = 47.5%. A game with no edge would give you exactly 50%. That 2.5-point gap is the entire business model.
Implementations differ. Some sites force a small share of rounds to bust instantly at 1.00x, others shade the whole curve a little. The outcome has the same shape either way: every target you could pick is slightly worse than fair, and by the same proportion.
| Cashout target | Chance at 3% edge | Chance at 5% edge | Chance at 8% edge |
|---|---|---|---|
| 1.10x | 88.18% | 86.36% | 83.64% |
| 1.20x | 80.83% | 79.17% | 76.67% |
| 1.50x | 64.67% | 63.33% | 61.33% |
| 2.00x | 48.50% | 47.50% | 46.00% |
| 3.00x | 32.33% | 31.67% | 30.67% |
| 5.00x | 19.40% | 19.00% | 18.40% |
| 10x | 9.70% | 9.50% | 9.20% |
| 25x | 3.88% | 3.80% | 3.68% |
| 100x | 0.97% | 0.95% | 0.92% |
| 1,000x | 0.097% | 0.095% | 0.092% |
Your return is the same at every target
Multiply the chance by the payout and you get the average return per dollar bet. At 2x: 0.475 × 2 = 0.95. At 10x: 0.095 × 10 = 0.95. At 1.1x: 0.8636 × 1.1 = 0.95. With a 5% edge, every target hands back 95 cents per dollar over time.
So the cashout point doesn't change how much you lose in the long run. It only changes how the loss arrives. Low targets give you lots of small wins and the occasional ugly loss. High targets give you long dry spells and the odd big hit. Almost every myth below is an attempt to find a target or betting pattern that escapes that 0.95. None of them can.
Myth 1: "Auto-cashout at 2x is the safe play"
A 2x auto-cashout feels sensible because it's close to a coin flip. And it is close to a coin flip. It's a slightly bent coin, and the bend never goes away.
The table shows what that 47.5% does across different session lengths if you bet $1 every round. "Ahead" means you finish with strictly more than you started.
| Rounds played at 2x | Expected result | Chance you finish ahead |
|---|---|---|
| 10 | −$0.50 | 31.7% |
| 100 | −$5.00 | 27.4% |
| 500 | −$25.00 | 12.2% |
| 1,000 | −$50.00 | 5.3% |
Short sessions are where people get fooled. About three in ten 10-round sessions end in profit, so plenty of players run a 2x system for a few nights, come out ahead, and decide it works. Stretch the same habit over a month and the chance of being up shrinks toward nothing, while the expected loss grows in a straight line with the amount you wager.
Lower targets aren't safer, they're lumpier
The 1.1x crowd has its own version: "I win almost nine times out of ten." True enough. At a 5% edge you hit 1.1x about 86.4% of the time. But a win pays 10% of your stake and a loss costs all of it, so one bust wipes out ten wins.
Over ten rounds at 1.1x you need all ten to land just to finish ahead. The chance of that is 0.8636 to the power of 10, about 23%. Over 100 rounds the chance of being ahead drops to roughly 11%, and over 500 rounds it's basically zero. You'll have a lot of pleasant, boring sessions, then one or two that take back weeks of tiny wins.
Myth 2: Martingale will grind out a steady profit
Martingale is the oldest system around. Bet $1 at 2x, double after every loss, go back to $1 after a win. Each win at 2x recovers all the previous losses plus $1, so the theory says you can't lose as long as you keep doubling.
The flaw is that the doubling has to stop somewhere. Your balance runs out, or the site's maximum bet does. This is how fast a streak escalates with a $1 starting bet at 2x and a 52.5% chance of busting each round:
| Losses in a row | Total lost so far | Next bet needed | Chance of this streak |
|---|---|---|---|
| 1 | $1 | $2 | 52.5% |
| 2 | $3 | $4 | 27.6% |
| 3 | $7 | $8 | 14.5% |
| 4 | $15 | $16 | 7.6% |
| 5 | $31 | $32 | 4.0% |
| 6 | $63 | $64 | 2.1% |
| 7 | $127 | $128 | 1.1% |
| 8 | $255 | $256 | 0.58% |
| 10 | $1,023 | $1,024 | 0.16% |
Say you bring $255. That covers eight losing bets in a row ($1 + $2 + $4 and so on up to $128). Eight straight busts below 2x happen about 0.58% of the time, which sounds tiny. But a Martingale player doesn't play one sequence. They play hundreds. Each finished sequence wins exactly $1, and the 0.58% wall shows up on average once every 173 sequences.
Put those together. On average you win about $173 in $1 steps, then lose $255 in one go. The expected result works out to roughly −$0.48 per sequence. There's a 44% chance the wipeout arrives within your first 100 sequences, and since a sequence averages about two rounds, that's only around 210 rounds of crash.
A bigger bankroll just moves the cliff further away. With $1,023 you survive ten losses, the bust chance per sequence drops to about 0.16%, and the average run before disaster stretches to around 630 sequences. Then the $1,023 goes in one hit, against roughly $630 of accumulated winnings. And many sites cap the maximum bet, which can stop the doubling long before your balance would.
Myth 3: "It's due for a big one"
After a run of instant busts and 1.1x crashes, chat fills up with people convinced a 50x is coming. This is the gambler's fallacy at its purest.
Go back to how the round is made: a hash generated long before tonight. The previous ten crash points have no influence on it. If the chance of reaching 10x is 9.5%, it's 9.5% after ten low rounds, after ten high rounds, and after a 500x. Low rounds don't build pressure. There's nothing in the system that could store pressure.
The reverse belief exists too. Some players skip rounds after a big multiplier because "it just paid out." Skipping costs you nothing (sitting out is the only strategy with a 0% house edge), but it isn't protecting you from anything either.
What normal streaks look like
Streaks feel suspicious because people underestimate how common they are. At a 5% edge, each round has a 52.5% chance of busting below 2x. Over a 300-round session:
- A run of six or more sub-2x rounds in a row happens about 96% of the time.
- A run of eight or more happens about 56% of the time.
- Going 20 rounds without a single 10x happens about 14% of the time.
Seeing any of these isn't evidence of a rigged game or a coming payout. It's what randomness looks like when you stare at it for an hour.
Myth 4: You can read patterns in the history bar
The colored strip of recent results is there to keep you engaged. It looks like data, so people treat it like data: three greens then a red, wait for two reds, bet on the third.
Nobody has a pattern that survives a few thousand rounds, because the numbers come from SHA-256 output, and that output is random for every practical purpose. If you could predict the next hash from the last one, you'd have broken the hash function that Bitcoin relies on. You'd have better things to do than play crash for skins.
Crash predictors are a scam, every time
Telegram channels and video descriptions sell "crash predictor" bots and scripts. Some want a fee. Some want you to log in through their own page, which is exactly how people lose their Steam inventories (our guide to the Steam API key scam shows how that works). Some are plain malware. A tool that actually predicted a provably fair crash game would be worth more than a site's whole bankroll. Nobody would sell it for $30 in a group chat.
Myth 5: Splitting your bet covers the risk
A popular variation: put half the bet on a low auto-cashout to "cover" the round and half on a high one for the upside. Say $10 in total, $5 cashing out at 1.5x and $5 at 10x, at a 5% edge.
| What happens | Chance | You get back | Net result |
|---|---|---|---|
| Crashes before 1.5x | 36.7% | $0 | −$10.00 |
| Crashes between 1.5x and 10x | 53.8% | $7.50 | −$2.50 |
| Reaches 10x or higher | 9.5% | $57.50 | +$47.50 |
| Average per round | $9.50 | −$0.50 |
Look at the middle row. The "cover" bet means you lose $2.50 in the most common outcome, and the combined return is still exactly 95%. Splitting changes how the results feel. The edge doesn't move.
Myth 6: Manual cashouts beat auto-cashout
Some players think watching the curve and clicking at the right moment gives them an edge. Your reflexes don't change the distribution of crash points, so the best manual play can do is match what an auto-cashout at the same target would have done.
In practice it does worse. Lag, a slow click, a phone that stutters, and the very human urge to hold for "just a bit more" all push you toward later cashouts. Later cashouts don't pay better on average. They just bust more often. Set an auto-cashout before the round and leave it alone.
Myth 7: The site rigs crash against big bettors
This one deserves a fair hearing, because shady sites do exist. On a properly provably fair crash game, though, the crash point is committed through the hash chain before bets are placed, so the site can't see a big bet coming and cut the round short. You can check this yourself: after a round, take the revealed hash, run it through the site's published formula (most fairness pages have a verifier), and confirm the crash point matches.
The real "rig" is out in the open. It's the house edge. If a site doesn't state its edge or doesn't let you verify past rounds, don't play crash there. That's a better filter than any theory about whales.
What you actually control
| Factor | Can you control it? | Does it change your expected loss? |
|---|---|---|
| Cashout target | Yes | No, only the size of the swings |
| Bet sizing pattern (Martingale and friends) | Yes | No |
| Which site, and its house edge | Yes | Yes |
| Total amount wagered | Yes | Yes, the loss scales with volume |
| Rakeback and rewards | Partly | Yes, a little |
| When you stop | Yes | Yes |
The bottom half of that table matters more than any system. Your expected loss is roughly the house edge times everything you wager. Wager $2,000 over a month of $10 crash bets at a 5% edge and the expected cost is $100. Rakeback can hand back a slice of that (our guide to bonus codes and wagering requirements shows how to work out what rakeback is really worth), but nothing turns the number positive.
Site choice is the only lever on the edge itself. Clash.gg, CSGORoll, CSGOLuck and SkinRave all run crash with provably fair verification, and our reviews of Clash.gg and CSGORoll go into how their versions play day to day. If you're new to the mode, the crash game guide covers the basic rules and the variants you'll run into.
How we'd play crash if we were going to play it
- Set a session budget in real money before depositing, not in site coins or skins. Treat it as already spent.
- Pick one auto-cashout target that matches the swings you can stomach, then leave it. Low targets mean frequent small wins and rare big losses. High targets mean long droughts and rare big wins.
- Keep each bet to 1-2% of the session budget, so a normal bad streak doesn't end the night in ten minutes.
- Never raise your stake to chase a loss. That's Martingale with extra steps.
- Verify a few rounds now and then on the site's fairness page. It takes two minutes.
- Stop at a time limit, not at "once I'm back to even." Back to even is where a lot of long nights start.
- Ignore predictors, signal groups and anyone selling a system.
Crash is fine as paid entertainment with a known price, about 5 cents per dollar wagered in our examples. It turns into a problem when you start believing a system has lowered that price to zero. If it already feels that way, our responsible gambling guide for CS2 players is a better use of the next ten minutes than another strategy video. Every site we review is 18+ only, and none of them accept players from the US or UK.
