Museo
Operate the idea.
A small museum of hard ideas you can put your hands on. You don’t read these — you turn the dials until the thing becomes obvious.
Everything here is somebody else’s idea, and nothing here is for sale. That is the whole distinction between this room and the Lab, where the things being operated are mine. An explanation you can manipulate is the same problem as an agent telling a person what it needs — take something opaque and render it so a human can act on it — which is why a portfolio about that problem has a museum attached to it instead of a blog.
Open now
- 01 · ConcurrencyWait, or ApologizeTwo writers, one row. Lock it and everybody waits; don’t lock it and somebody redoes their work. Push the contention up and watch which bill grows faster.Optimistic and pessimistic aren’t safe vs fast — they’re waiting vs waste.
- 02 · AgreementThe Treacherous GeneralsDrag a network past a third of its members lying and watch agreement become mathematically impossible — then see exactly why the threshold is a third, and not a half.Why every consensus protocol you will ever read quotes the same fraction.
- 03 · SuiOwned vs SharedMost Sui transactions never reach consensus at all. Move an object between ownership modes and watch the execution path change underneath it.Consensus as a conditional cost rather than a universal one.
The territory
Three grounds, and the exhibits grow out of them rather than the other way round. Naming the questions is not the same as claiming the answers — what is built is above, and it is two things.
Agreement and concurrency
What a system gives up in order to go fast.
Pessimistic locking buys safety by making everybody wait. Optimistic control bets that collisions are rare, and pays for the bet when it loses. Then CRDTs — the machinery inside Yjs and Automerge — refuse the premise entirely and make conflicting edits merge by construction, which is either the end of the argument or a far more interesting version of it. Byzantine agreement, exhibit 02, is the extreme end of the same question: what does agreement cost when the other participants may be lying rather than merely slow?
The transport layer
What HTTP actually promises, versus what we assume it promises.
The humble queue, which quietly solves more distributed-systems problems than any clever thing ever built. Back-pressure, retries, and why “just retry it” is a decision about correctness rather than about reliability. What serverless genuinely changes about the shape of a request — and what running at the edge changes again, once the compute is in a hundred places and the data is still in one.
The strange data layer
A blockchain is not a transport layer. It is a very exotic database.
It buys agreement between parties with no reason to trust one another, at a price no ordinary database would consider paying — and the entire discipline is judging when that price is worth it, which is almost never and occasionally everything. Sui is the specimen here, because its object model makes the consensus question conditional rather than universal: exhibit 03 is that idea, made operable.