Skip to content

What you can do with Arvo

Seven things, in roughly the order you would do them. Each one links to the page that walks you through it.

Test whether a trading idea holds up

"Does this actually work, or did I just find a number I liked?"

Describe an idea as a rule, choose what to test it on, and run a study. You get back a finding: a verdict, plain-language advice, and the evidence behind both.

The verdict is one of three — Supported, NotSupported, or Inconclusive. The third one matters: it means there isn't enough evidence to say, which is a real answer and one most tools cannot give you.

Run a study and read the verdict →

Search parameters without fooling yourself

"Which settings work best?"

Run a rule over a grid of parameter values. Arvo picks a winner on one slice of history and judges it on a slice it never saw — then checks whether the winner beats what the best of that many tries would score by luck alone.

So a parameter search gives you a result you can weigh, rather than the best of a hundred guesses wearing a Sharpe ratio.

Write a ruleset →

Test one idea across many instruments

"Does this work generally, or just on the one chart I looked at?"

A panel runs one rule across a whole list of instruments, with one configuration for all of them. Tuning per instrument would be a second search; holding the configuration fixed means the rule has to work broadly to pass at all.

It also pools evidence, so a rule that trades rarely can accumulate enough trades to say something.

Universes →

Bring in a strategy you already have

"I have this in TradingView."

Point Arvo at a Pine v5 script and it reads what it can express as a rule — and tells you by name what it could not, rather than translating approximately and leaving you with a rule that quietly does something else.

The result is then a rule like any other, and goes through the same evaluation.

Import from Pine →

Paper trade it against a real broker

"Would this behave the same way with real market conditions?"

Take a finding and run it on a broker's paper endpoint: real API, real fills, real latency, simulated money. Arvo compares the price the rule decided at against the price it actually got, every time.

That difference is the thing a backtest assumes and cannot measure. It is why five paper days are required before a rule can touch real money.

Start a paper session →

Let it run

"Can I stop watching it?"

A session keeps trading with nothing open — no window, no terminal, no agent attached. It checks its own positions against the broker on every poll, and if the two disagree it stops taking new positions and waits for you rather than trading on a position it is not sure it holds.

One kill switch stops everything, and it stays stopped until a person clears it.

Sessions → · Halt trading →

Find out why you own something

"Why did it buy this?"

Walk backwards from any position: the fill, the order, the risk decision that sized it, the signal, the rule that fired, and the bar it fired on — with the market conditions at the time.

Not a log you search through. A record built to answer that question.

Explain a position →

Work with an AI agent

"Can Claude help me with this?"

Attach an agent and it can write rules, run studies, and read every finding. It cannot place an order, reach a broker, or touch a credential — not by policy, but because the tools that would let it do those things do not exist on the surface it is given.

Starting a session stays a person's decision.

Ask the agent → · With Claude Code →

What Arvo will not do for you

Worth saying plainly, because it is the other half of the product:

  • It will not tell you what to trade. No recommendations, no signals, no shared list of winning strategies.
  • It will not make a weak idea look good. Fewer results will pass here than elsewhere. That is the point, not a limitation.
  • It will not trade on its own initiative, and neither will an agent you attach to it.

Next