Autonomous research that shows its work.
Assayer Research is building an autonomous research engine that organisations can give a mandate and hold accountable for its work. It selects questions, runs empirical studies, challenges its own methods and records what the evidence supports. People set its boundaries and keep the decisions about authority and publication. It is starting with financial markets.
Loading the live view…
This panel reads a filtered activity file written by the research system. If the file cannot be reached, nothing is shown rather than old or invented activity.
The engine's latest recorded steps, newest first. Each line is a fixed sentence for that kind of step.
Operates within a defined mandate. Scope, budget, access and publication remain owner decisions.
- Working on its own
- Waiting on a person
- A person stepped in
Reading the research record
When a person stepped in
No person stepped in during these hours.
The last 7 days
The record so far
A prototype under calibration. Nothing has been tested on locked data or reviewed by an independent expert yet, and there is no real-money trading. What is not yet shown →
Go deeper
How the engine works, every study it has run, what we are learning, and what is not yet shown.
How it works →
The agents and checks, how it runs itself, the loop every study follows and the statistical toolkit.
02Research →
Every study, newest first, each with its own page: question, rules, method, result and numbers.
03Notebook →
What we are learning from building it, the good and the bad, in dated notes.
04Limits →
What is not yet shown, what is outside the mandate, and the corrections record.
Why “Assayer”
An assayer tests ore to find out what it contains. Assayer Research brings that spirit to financial ideas, and to the engine itself.
We are building a small research operation that works mostly on its own, and shows every time a person steps in. Within a defined mandate it chooses questions, builds and runs investigations, sends its work through review, and records the results.
It is also a practical test of autonomous AI: can a system recover when something breaks, repair its tools, improve its methods, and learn from earlier work? It has already repaired faults and revised methods after finding weaknesses. Running real investigations shows where these ideas hold, where they fail, and what to improve next.
Finance is the first field, with the aim of finding insights that hold up well enough to act on. Over time we want to see how much of this carries into other kinds of research.
The method comes first.
Each study records its methods and decision rules before computing its results.
The work is challenged.
Separate AI review roles examine the design, analysis and conclusions.
Uncertainty stays visible.
Findings, failed tests and inconclusive results remain in the research record.
Talk about the research
For research discussion, independent review of a study, or institutional enquiries. Outside claims are not being taken for testing yet; the limits page explains why.