Everything we have published, newest first, from every desk.
The new consensus protocol targets about 150 ms finality. On the morning of 26 September the Foundation's own dashboard showed devnet finality at 322 ms, and vote transactions gone from blocks.
BatchV1_1 briefly fell below 80% validator support, which wipes the accumulated time. The ledger shows the new majority from 25 September; Permission Delegation reset too.
Eight files, 453 lines added and 620 removed. A standalone draft becomes nested patches against the specs it modifies, with Test Plan sections added.
SIMD-0376 now specifies the cofactored equation from Taming the many EdDSAs plus explicit rejection of non-canonical encodings and small-order A and R - because ZIP-215 alone accepts the all-zero pubkey.
A three-line change to bip-0375/deps/dleq.py applies % GE.ORDER to the challenge and adds two lines to proof verification. Small diff, cryptographic surface.
Referendum #498 made Kusama an experiment-focused peer of Polkadot. The open questions are where it runs, who secures it, and whether the Vision Programme should be capped at two years.
Attackers spoofed transaction data so that legitimate-looking transfer approvals drained hot and warm wallets. The haul includes about 103 million XRP; the CEO points at North Korea.
An ethresear.ch post reframes the question as the Trusted Computing Base: everything trusted rather than proven, which is what humans still have to check.
Twenty-one CFD firms have closed since 2025 and three are cancelling permissions. The conduct was little UK business plus an authorisation that made linked overseas companies look regulated.
Bitget's approvals trusted their own request generator. Solana's consensus trusted a library. XRPL trusted a duplicated spec. And 21 CFD firms sold a licence they barely used.
