Stake Mines predictor: what a seed checker can tell you, and what nothing can

No tool can show you the next Stake Mines board, ours included. What a seed checker can do is reproduce every board you have already played, exactly. This post explains where the mines come from, why that rules out prediction, and what the Mines tools actually compute.

Where the mines come from

Every Stake Original draws its result from three values: the server seed, which Stake holds, the client seed, which you set, and the nonce, which counts up by one each bet. They go into HMAC-SHA256, and the hash bytes become a list of numbers between 0 and 1.

hash   = HMAC-SHA256(server seed, "client seed:nonce:round")
floats = hash bytes, four at a time, each scaled to [0, 1)

Mines uses one float per mine. The first float picks one of the 25 tiles. That tile is removed, the second float picks one of the remaining 24, and so on until every mine is placed. The whole board exists before you click anything. Clicking only reveals it.

tiles = [0 … 24]
for each mine:
  index = floor(float × tiles left)
  mine  = tiles[index], then remove it

The provably fair guide works through the hashing for one bet by hand.

What is hidden before the round

Two of the three inputs are public. You chose the client seed, and the nonce is shown on every bet. The server seed is not. Before you play, Stake shows only its SHA-256 hash, which commits Stake to the seed without revealing it. The seed itself is revealed when you rotate to a new pair.

That is the whole design. While a seed pair is active, nobody outside Stake can compute the floats, because the one input that matters is missing. After rotation, everybody can, which is what makes past bets verifiable.

What a predictor would need

A tool that showed safe tiles on a live board would need one of three things:

• The active server seed. Stake does not publish it, and a hash cannot be reversed. The seed cracking post (opens in new tab) puts numbers on what a brute-force search would take.
• A pattern across nonces. HMAC-SHA256 output has none. Board 1,000 says nothing about board 1,001, which is why the streaks guide finds the same independence in real bet archives.
• A bias in tile positions.Each float selects uniformly among the tiles left, so every tile is equally likely to hold a mine. The tracker’s heatmap, below, shows this directly over large samples.

A “scanner” that reads the board from your screen has the same information you do: nothing. It cannot do better than picking tiles at random, because that is what it is doing.

The multiplier is fixed in advance

Because the board is uniform, the multiplier for revealing n safe tiles depends only on n and the number of mines b. It is the inverse of the chance of surviving those reveals, less the 1% house edge:

chance(n safe)  = C(25 − b, n) / C(25, n)
multiplier(n)   = 0.99 × C(25, n) / C(25 − b, n)

Three mines and five tiles pays 2.00x and succeeds 49.57% of the time. Ten mines and ten tiles pays 1077.61x and succeeds 0.09% of the time. Which tiles you choose, and in what order, changes neither figure. The Mines payout table (opens in new tab) lists every combination from 1 to 24 mines.

What you can check

Verification runs the generator above on a revealed seed. The steps:

1. Rotate your seed pair in Stake’s fairness settings. Stake shows the server seed that was active.
2. Open the Mines seed checker (opens in new tab), enter that server seed, your client seed and the nonce range you played.
3. Compare each recomputed board with the bet in your history. The mines are where Stake said they were, or the hash of the revealed seed does not match the hash you were shown, which would be visible too.

That proves the boards you already played were not changed after the fact. It says nothing about the boards on your new seed pair, which are hidden again behind a new hash.

What the tracker shows

The address /stake-mines-predictor on Stake Analyzer now opens the Mines tracker (opens in new tab), and the rename is deliberate. The tracker records your bets as they land and draws a heatmap of where mines fell, with cash-out stats by tile count. It describes what happened. Over a large sample every tile converges on the same frequency, which is the uniform placement above showing up in your own data.

If you want to see the generator run on a board you control, the Stake simulator (opens in new tab) plays Mines on a seed pair it holds for you and lets you replay any nonce afterwards, with the server seed shown.