@Filiprogrammer Would you mind pointing your GTX 1060 one more time?

The representation of keys/second has been tweaked a little (mostly to recovery from a long pause)

sime

Another heads up to @OT, Krackpot now uses two private pools. https://stacker.news/items/1545473

Shout out to @OT for the push to [not public boardcast the transaction](https://stacker.news/items/1529472/r/sime?commentId=1529858).

, a browser WebGPU solver for Bitcoin Puzzle 71. In that post I said the app broadcasts the claim transaction to mempool.space. That is no longer true, and this is the correction.

The reason is front-running. A puzzle spend reveals the public key in the input scriptSig.  The key sits in a known 2^70 window, so Pollard's kangaroo recovers it from that public key in about 2^35 group operations, which is minutes on one GPU. A bot watching the mempool re-derives the key and rebroadcasts at a higher fee. This is not hypothetical: Puzzle 66 and Puzzle 69 were both taken this way. Bumping my hardcoded fee does not help, because the attacker simply bids higher.

So on a hit the browser now posts the signed transaction to Rebar Shield's JSON-RPC endpoint, which routes to partner pools' private mempools.

Two more tiers behind that. The same signed transaction is encrypted to a dedicated Nostr key (NIP-44, gift-wrapped per NIP-59, published to five relays) so I can submit it privately too. It is the identical transaction with the finder's 6 BTC output already signed in, so I cannot redirect it. If the Shield call errors, the UI gives manual Slipstream instructions. In every case the recovered key and the full signed hex appear first, so nobody depends on me or on Rebar to claim.

Write-up, updated for the new claim path:

It still ships unminified with no build step, so you can read the claim path in devtools before you run it. If you see a hole in it, I want to hear about it.

bitcoin

Last week I [posted Krackpot here](https://stacker.news/items/1529472), a browser WebGPU solver for Bitcoin Puzzle 71. In that post I said the app broadcasts the claim transaction to mempool.space. That is no longer true, and this is the correction.

The reason is front-running. A puzzle spend reveals the public key in the input scriptSig.  The key sits in a known 2^70 window, so Pollard's kangaroo recovers it from that public key in about 2^35 group operations, which is minutes on one GPU. A bot watching the mempool re-derives the key and rebroadcasts at a higher fee. This is not hypothetical: Puzzle 66 and Puzzle 69 were both taken this way. Bumping my hardcoded fee does not help, because the attacker simply bids higher.

So on a hit the browser now posts the signed transaction to Rebar Shield's JSON-RPC endpoint, which routes to partner pools' private mempools.

Two more tiers behind that. The same signed transaction is encrypted to a dedicated Nostr key (NIP-44, gift-wrapped per NIP-59, published to five relays) so I can submit it privately too. It is the identical transaction with the finder's 6 BTC output already signed in, so I cannot redirect it. If the Shield call errors, the UI gives manual Slipstream instructions. In every case the recovered key and the full signed hex appear first, so nobody depends on me or on Rebar to claim.

Write-up, updated for the new claim path:
[https://krackpot.io/blog/how-it-works](https://krackpot.io/blog/how-it-works)

App: [https://krackpot.io](https://krackpot.io)

It still ships unminified with no build step, so you can read the claim path in devtools before you run it. If you see a hole in it, I want to hear about it.