Spent the last few months building a Layer-1 blockchain from scratch as a way to go deep on post-quantum cryptography in a real system rather than a toy. A few things that might interest this crowd:
No elliptic curves anywhere. Every signature — transactions, peer identity, admin auth — is ML-DSA-87 (NIST PQC standard). That means 2592-byte public keys and 4627-byte signatures flowing through everything, which forced some interesting stack-management work: ML-DSA-87 signing needs ~4 MiB of stack, so all crypto ops run on an explicit 16 MiB-stack thread to avoid STATUS_STACK_OVERFLOW on Windows (and silent corruption on Linux under load).
No HTTP. P2P transport is QUIC + TCP with a custom Noise_XX handshake that binds the ephemeral session key to the peer's long-term ML-DSA-87 identity before any application data moves.
Merkle-Patricia Trie state root with compact (~2 KB) inclusion/exclusion proofs for light clients — went through two iterations to get proof size down from an initial ~25 KB.
WASM smart contracts via Wasmtime under a strict 16 MiB memory / 512 KiB stack mandate, gas-metered host functions.
This is a solo research project, not under active development going forward — publishing it mainly for anyone curious how a PQ-native L1 fits together end to end, or who wants to pick it apart. Happy to answer questions about any of the design decisions above.
If AI could auto-generate a production-ready L1 with post-quantum crypto, a custom Noise_XX P2P transport, and a packed Merkle-Patricia Trie without a single runtime bug, we’d already be living in the singularity. Until then, sadly, it all has to be hard-coded by hand
Has the em-dash definitely been eliminated from English language? I remember a previous-century book on English writing had finely written sentences that used them, and I am sure that I used quite a few in my PHD thesis more than 15 years ago.
I wouldn't say "eliminated", but from what I have seen, it is not as common as it used to be. In the context of coding specifically, the em-dash is a sign of AI writing most of the time. It isn't on keyboards. Unless someone intentionally finds it, I see no reason why they would use it.
Stack jumping. Since the RSP jumped past the Guard Page directly into this active, kernel-valid memory, the processor calmly allows the write. The cryptographic stream begins dumping megabytes of its matrices right on top of someone else’s data. To Linux, this looks like a completely legitimate operation the process is writing to memory that belongs to it. And the fact that one thread has reached into another’s “pocket” and overwritten Arc pointers or raw transaction bytes there the kernel doesn’t care
