same attack, on a rule we're testing in the lab: local agreement, where a block goes in once 2 of 3 proposers attest it
today's rule skipped 16 slots, under agreement none, every cblock final in 383 to 387 ms
acrux.computer/lab/c38d527b-…
proposer 3 sent its commitment to one co-proposer and not the other, trying to get two different views of the slot
they attested what they actually saw, the vectors didn't match, nothing was agreed and the cblock got skipped. nobody gets a slot out of a split
acrux.computer/lab/dd837333-…
slot 4,366,946: every transaction in it was sent to all three proposers, so 24 came in across three blocks and 8 came out, duplicates dropped
then it's sorted by fee per CU, compute limit, hash, and packed onto four tracks, the same way on every validator
proposer 3 sent its commitment to one co-proposer and not the other, trying to get two different views of the slot
they attested what they actually saw, the vectors didn't match, nothing was agreed and the cblock got skipped. nobody gets a slot out of a split
acrux.computer/lab/dd837333-…
the 20 ms commit window runs on each node's own clock, counted from when it saw the slot start. nothing in the protocol needs synced clocks
line 24 of the paper's Algorithm 1 next to the function that does it in acrux-node: acrux.computer/spec
send a transaction 2-of-3 on Acrux and it goes out as three encrypted pieces, one to each proposer. any two put it back together, none of them can read it alone
measured: 8,000 lamports and 339 ms to executed, the same as sending it to all three in the clear
acrux.computer/send
three hubs today, FRA, NYC, TYO. a fourth one can be run by outside operators now
four of them register and join a hub through the registry with 1% of stake between them, and it goes active at the next bridge-epoch boundary. windows land on it from then
acrux.computer/docs/node/joi…
every transaction on Acrux now shows what its payer paid and where it went
one from the merged order: base 5,000, priority 38, inclusion 3,000, 8,038 lamports total. half the base fee, 2,500, is burned
after slot 4,752,000 those lamports are ACRUX
ran an outside node into the FRA hub on the live net, from a desktop about 290 ms away
direct links to all four FRA co-proposers, 3,172 finalized slots re-executed, every state hash matching, and it kept up. proposing comes once it's admitted
acrux.computer/docs/node/joi…
pulled the plug on one of the three proposers before its window even started
the other two marked it missing, slot 64 agreed on just their two blocks and the window kept going. that's why it's three and not two
acrux.computer/lab/91b654be-…
ran a whole Solana mainnet epoch through Acrux. 431,775 blocks, 156M transactions
6,585 sandwiches. the traders took 77.82 SOL net, the victims we could price lost 85.85 SOL to front legs. sent 2-of-3 on Acrux there's no front leg
acrux.computer/replay
Replayed 2,012 real Solana mainnet slots through Acrux's merge.
48 sandwiches, 66 people caught in them. Sent 2-of-3 on Acrux, the 54 we could price would've kept 0.153 SOL of the 0.192 the traders took.
acrux.computer/replay
nitter.gq/AcruxComputer/status/2…
sign two blocks for one slot on Acrux and you've signed your own slashing proof.
ran it in the Fault Lab: proposer 3 sent two commitments for slot 64, both signatures sit side by side, cblock skipped. the other two don't get blamed for anything.
acrux.computer/lab/81ad5fe5-…
here it is running. acrux replay on slot 470,916,233: the trader's buy sits right in front of the victim in the leader's order, 2 and 4. On Acrux they land at 33 and 35.
nitter.gq/AcruxComputer/status/2…
Replayed 2,012 real Solana mainnet slots through Acrux's merge.
48 sandwiches, 66 people caught in them. Sent 2-of-3 on Acrux, the 54 we could price would've kept 0.153 SOL of the 0.192 the traders took.
acrux.computer/replay
nitter.gq/AcruxComputer/status/2…
Build 1 is done: Solana programs run on Acrux unchanged.
Raydium's own SDK made a CPMM pool on Acrux at the program's mainnet address and swapped 1,000 A for 3,950.59 B. SPL Token, Token-2022, ATAs and Memo the same way, only the URL changed.
If you've got a Solana app, this is how it moves: acrux.computer/docs/sdk/move…











