Home

Research. Test. Trade. Know why.

Arvo is a platform for systematic trading that treats evidence as part of the product. Research an idea, test whether it holds up, paper trade it against a real broker, run it live — and keep the record that explains how every decision was made.

Most platforms help you find a strategy that worked. Arvo is built on the assumption that most things that worked did not work — you just looked at enough of them.

From idea to execution, in one place

flowchart LR
    H["**1 · Hypothesis**<br/><small>describe what should work</small>"]
    E["**2 · Experiment**<br/><small>search parameters,<br/>chosen out of sample</small>"]
    V["**3 · Evidence**<br/><small>search bias, costs,<br/>market conditions</small>"]
    P["**4 · Paper**<br/><small>real fills,<br/>real latency</small>"]
    L["**5 · Live**<br/><small>same rule,<br/>same risk policy</small>"]
    X["**6 · Explain**<br/><small>every position,<br/>walked back</small>"]

    H --> E --> V --> P --> L --> X
    X -.->|"what you learned"| H

    style V fill:#1f6feb,stroke:#1f6feb,color:#fff

1 · Hypothesis. Say what you think should work, as a rule with entries and exits.

2 · Experiment. Test it over a range of parameters. Arvo chooses on one slice of history and judges on a slice it never saw.

3 · Evidence. The result is weighed against how many things you tried, re-run at higher costs, and read against the market conditions that produced it. You get a verdict and advice, not just a number.

4 · Paper. Run it on a broker's paper endpoint and compare the price the rule decided at with the price it actually got.

5 · Live. The same rule, under the same risk policy that sized every backtested trade. Not a reimplementation — the same code.

6 · Explain. Any position walks back to the fill, the order, the risk decision, the signal, the rule and the bar.

What you can do with Arvo → Why it's different →

What makes Arvo different

• It can tell you it doesn't know

  Three verdicts, not two: Supported, NotSupported, and Inconclusive. A rule with eleven trades over twenty years has not been shown to work or to fail — and saying so is more useful than a ratio computed from eleven trades. Research →

• It counts how much you searched

  Try a hundred configurations and the best one looks good whether or not the rule has any edge. Arvo works out what the best of that many would score by luck, and makes the winner beat it. Why Arvo is different →

• One risk policy, research and live

  The code that sized every backtested trade sizes the real ones. A backtest cannot assume a position your live account would have refused, and a halt stays in force across a restart. The risk gate →

• Fills are measured, not assumed

  A backtest assumes a fill. Paper trading on a real broker gives you an actual one, at an actual spread — so the gap between them is a measurement. Sessions →

• Every position can be walked back

  Bar → signal → risk decision → order → fill → position, as one chain, with the rule and the market conditions on every trade. The audit chain →

• An agent that cannot reach your money

  Attach Claude Code and it writes rules, runs studies and reads findings. It holds no credential and has no tool that places an order — enforced by the build, not by a prompt. The Agent panel →

Who it's for

Traders with a systematic idea	You have a strategy and want to know whether it survives honest testing before it sees money
Developers	Rules are data, the API ships as Rust crates and a Python package, and the engine runs headless with no window at all
Quantitative researchers	Out-of-sample selection, deflation against search size, walk-forward, cost sensitivity and regime analysis, with replayable findings
People arriving from TradingView	Bring a Pine script in, and find out what it does under evaluation it has never had

What it costs you

Stated plainly, because it's the other side of the design:

• Fewer results will look good. That is the product working.
• A genuine but weak edge may come back Inconclusive. The bar is deliberately conservative.
• It is slower to a conclusion than a tool that reports the best of a sweep — because it refuses to report the best of a sweep.

Start

Two routes, depending on why you're here.

• I want to test and trade my ideas

  Install, connect an account, write a ruleset, run a study, read the verdict, paper trade what passed. The trading path →

• I want to build on it

  Run the engine headless, drive it from the command line, Python or an agent, then extend it. The developer path →

Under the hood

The engine runs studies, keeps the data library, holds credentials, hosts sessions and serves a local API. The window is one client of it; so are the Python package, an MCP agent, and the command line. The engine keeps running when the window closes, so a schedule runs and a session trades with nothing open.

Architecture → · Where this is going → · Run it headless →