The engineering notebook behind a post-quantum, shielded proof-of-work chain.
Security analyses, consensus notes, and honest negative results from building TSN — a CPU-mined, ZK-shielded L1. We publish what held up under scrutiny, and what didn't.
Selected work
pinned
The present that doesn't prove itself
We were tempted to reach for recursive "the present proves itself" proofs to kill a 16-minute miner warmup. An adversarial review corrected us: you don't prove your own disk to yourself. A note on caches versus proofs — and where each one actually pays.
All research
6 entries№ 004ArchitectureThe present that doesn't prove itself
Recursive proofs don't speed up a restart — a restart re-reads a disk you already trust. A precise line between a trust cache and a proof-carried checkpoint, and what each is actually for.
№ 003SecurityStealing a shielded note across a fork
When you change the spend-authorization scheme, the fork boundary is a theft window unless every legacy note is re-bound or abandoned. The MigrateToAuth cutoff, and the invariant we should have had from day one.
№ 002EngineeringTilting proof-of-work toward the CPU
Sizing the PoW working set to live in CPU cache turns a GPU's throughput advantage into a memory-latency penalty. Measured: 26–40× CPU over GPU. And why "cache-hard" is a truce, not a victory.
№ 001ConsensusForging weight by lying about height
Fork-choice trusts cumulative work, and a block that inflates its own height can inflate its weight. The one-line admission rule — height == parent + 1 — that shuts the door.
Why fair block distribution between miners is subtle
A field diary of a wrong diagnosis: variance → pacing → propagation delay → a note-accumulator livelock. What actually makes two equal miners take turns in long streaks.
Two nodes on one LAN that refuse to talk directly
The reachability policy binds miners to loopback — which quietly routes same-LAN peers through a distant relay. The private-address exception we're adding, without re-exposing miners to the internet.