Autonomous market infrastructure
Aurifern builds autonomous systems for on-chain markets, and measures every one of them long before it touches money.
Method
Save the input and the time it arrived. A feed that went quiet is not evidence that nothing happened.
Replay the same opportunity against a known baseline. If the comparison isn’t like for like, call it inconclusive.
Passing a test and being allowed to act are different things. Each step further needs its own decision.
Work
A few engines and the ground they share. Small on purpose.
Flagship
A batch auction wants one settlement that works for every order in it. This solver hunts for better ones and checks each before it answers. No claims about wins or money yet.
Live view The solver in CoW Protocol’s shadow competition, where answers are scored but never settled.
Every forecast is written down the moment it’s made, then checked against what actually happened, not against what we wish we’d said. The part that does the research can’t spend a cent. It can only suggest, and a strict set of risk rules decides whether anything happens at all.
Markets move too fast for “it looked good at the time” to count as a record. This is the record-keeping under every engine: any result can be checked again later, from the same inputs.
Detailed evidence is shared with protocol teams on request. Ask for it
Architecture
Market and chain data, each record with its source and time.
Past moments rebuilt and replayed. Nothing here trades.
Limits and permissions, walled off from everything else.
One per protocol. Each keeps its own risk book.
Status and incidents from what actually ran. No forecasts.
About
Automation does the repetitive work. Judgement stays with a person. Nothing touches capital before it has been measured.
Dan Z brings a mathematical background and a researcher’s habits to markets: state the hypothesis precisely, then try to break it. AI is built into the whole workflow, from reading protocols and papers to simulation, code and review. That is how a founder-led shop works at research depth without a large team.
Track record
Say hello
If you’re working on an auction, a data source or a failure mode worth examining, write with specifics. This is a small research-and-build operation, not a trading product for clients.