A small lab for markets that are still being built.
PRLX researches, builds and tests trading strategies on alternative markets. We are a small team, and our systems run close to the venues they trade.
Each question we ask becomes a research programme, with its own data, tools and written record.
PRLX began with a handful of questions about new markets. How is a Solana block really ordered, and where does a transaction land inside it? How are short-dated prediction markets quoted, and what happens to those quotes when a faster market moves first? Could an AI agent learn to trade, rather than be told how?
We are deliberately small. Every strategy is designed, tested and run in-house, and we trade only our own capital. We don't manage money for others, take deposits or give advice, which leaves us free to drop an idea the moment the evidence turns against it.
We publish once we have finished with an idea: a design we are ready to discuss, or a study whose outcome is settled. Live strategies stay private.
We're named for an error.
Parallax error is the wrong reading you get when you look at a gauge from the side. The needle hasn't moved; your viewpoint has. Markets have the same problem. Every venue, validator and feed shows the price from a different position, and a little late.
In a market the needle is the price, the scale is the moment it is observed, and the gap between them is latency. Two observers in two places read two prices. Arbitrage is what that difference is worth, and speed is how small you can make your own.
Eight rules, applied to people and machines alike.
They began as the constitution for our agent research and became the way the whole lab works.
Process over outcome
A trade is judged on the quality of the decision, not on whether it happened to make money.
Pre-register or it didn't happen
A test counts only if its hypothesis, data split and pass mark were written down before the data was read.
Mechanism before money
We need a reason an edge should exist, and something else it predicts, before capital goes near it.
Capital preservation is absolute
Limits and kill switches are set before a system runs and are never renegotiated mid-session.
Dissent is an asset
The case against an idea gets the same hearing as the case for it, especially when the room agrees.
Truth in the ledger
Nothing is deleted. Every hypothesis, decision and outcome is numbered and kept, including the wrong ones.
No lookahead, ever
A model may only use information that existed at the moment it was asked to decide.
Trust is earned in increments
New systems start on paper, then at small size, and grow only as the evidence does.
How we trade.
- We never sandwich users, on any venue.
- We trade our own capital only. We don't take outside money, accept deposits or give investment advice.
- Trading keys never sit on research machines. Live systems sign only the transactions they are permitted to send, within fixed limits.
- Live systems run inside fixed limits, including a rolling cap on what they can spend in any 24 hours.