Is it fair?
Every lucky mode is built to average 105.6% of the price. That 5.6% is Luckout's only revenue. Here is what lucky buyers have actually paid so far, read from the contract within the last minute.
Live Average
87.5%
of the price, over 10 settled lucky buys
Promised
105.6%
in the long run, for every mode and every size
Over a few dozen Draws the average swings a lot; over thousands it settles near 105.6%. Every Draw links to a Draw Proof you can recompute in your browser, like this one.
The Cashback Reserve
Free
$293.08
Locked for purchases in progress
$0.00
Most cashback one purchase can get now
$58.62
Each lucky buy locks what its biggest possible cashback would cost before it starts, so every cashback is already covered. One purchase gets at most the lower of a fixed amount and a share of what's free, so the reserve always outlasts a run of luck.
What the contract guarantees
- Every Game Version is immutable, and the contract refuses any table that doesn't average exactly 105.6%, or whose cashback could leave you above the price.
- The Draw runs a block after delivery is confirmed, so nobody, including us, can retry it or cancel a purchase after seeing the result.
- Your per-purchase and daily limits apply to every purchase, whoever submits it. The daily limit resets every day at .
- A purchase we neither deliver nor cancel can be reclaimed in full after 24 hours.
What it can't prove: that a code was delivered (Bitrefill isn't on chain), and that we buy in the mode you chose. Both are bounded by the most you confirmed and by your limits. And Luckout can upgrade the contract: every upgrade is a public transaction, and it changes the code hash below.
Code hash NM61Tqbof2ncPBzVTwiAHKMfsr6g5YBTKtj2e8gqPZe. Build the contract from its source with cargo near build reproducible-wasm and you get the same hash.
The assistant's AI runs in a verified enclave
The Telegram assistant's model, deepseek-ai/DeepSeek-V4.1-Flash-1017, runs on NEAR AI Cloud inside a secure enclave. Every hour our server checks that with a fresh nonce: NVIDIA verifies the GPUs, Intel verifies the CPU enclave, and the enclave's signing key must be bound into both. Every reply must carry that key's signature over exactly what we sent and got back, or the assistant doesn't send it. Without a passing check, the assistant stays offline.
Latest check
Passed
Replies verified since the server started
0
Replies withheld
0
- Passed:The model runs on at least one attested TEE node
- Passed:Gateway: answered with our fresh nonce
- Passed:Gateway: Intel TDX quote verified
- Passed:Gateway: its signing key and our nonce are in the TDX report
- Passed:Gateway: runs the container config it attests
- Passed:Model node 1: answered with our fresh nonce
- Passed:Model node 1: Intel TDX quote verified
- Passed:Model node 1: its signing key and our nonce are in the TDX report
- Passed:Model node 1: runs the container config it attests
- Passed:Model node 1: NVIDIA verified its GPUs
Checked Fri, 13 Nov 2026 12:25:00 GMT with nonce 673942ac8adc7d3718e677941b46cd7efc492de4ac792b2d37f22779af0e71a3. Model signing key 0xfe0071895500a9755c18044948ab4f4e1a7f337e. How NEAR AI attestation works