Quant Mode
Quant Mode is an AI research loop that hunts for trading strategies, tests the evidence and learns from every result. You create an alpha hunt, save it and run it whenever you want.
- Hunt for alpha. Give it as much or as little direction as you want. Your quant builds the idea, backtests it, stress-tests it and grades what survives.
- Keep the best. Learn from the rest. Approve promising prospects and send failed ideas to the graveyard with a record of why they died.
- Get smarter with every hunt. Your quant reads the graveyard before it searches again, then rewrites its own plain-English playbook as it levels up.
Words we use
| Word | What it means |
|---|---|
| Alpha hunt | One automated research run: pick an idea, build it, test it, stress-test it, grade it. |
| Prospect | A strategy an alpha hunt found and handed to you for review. It waits under Prospects until you approve or kill it. |
| Graveyard | Killed strategies, kept with a note on why they died. The next hunt reads those notes. |
| Learn | A playbook rewrite. Every few finished hunts, your quant re-reads the graveyard and levels up. |
| Grade (F–S) | A letter on the simulated evidence, not a recommendation to go live. S is only awarded after every stress check passes. |
| Pin | On a backtest: "this is the run this strategy should be known for." The Optimizer only uses pinned runs. |
| Portfolio | A live line-up made only from Approved strategies. Hunts never add to one automatically. |
Run your first hunt
- Open Quant Mode in the sidebar of the app.
- Click Run an alpha hunt. A builder opens.
- Give it a title. That's the only required field. Leave the rest blank and we pick the idea, the markets and the timeframe.
- Click Save and run.
That's your part done. The hunt runs on its own. If it finds something worth your review, the result appears under Prospects with its final grade.
Want to point it? In the same builder, pick a universe, set a timeframe range, write a base idea, or drop in one of your Horizon scripts. See Alpha Hunts. Once a hunt is saved, Run an alpha hunt becomes a list — click a title to run it again.
What happens after you click
An alpha hunt takes one trading idea from nothing to a graded, stress-tested strategy. Six steps, all of them ours:
| # | Step | What we do |
|---|---|---|
| 01 | Pick the idea | We start from the hunt you created. If you left its optional fields blank, we choose something worth trying using what has worked before and what has already failed. |
| 02 | Build it | We write the strategy: entries, exits, stops, the lot. |
| 03 | Test it | We backtest it against real market history. |
| 04 | Fix it and test it again | While it's losing, we change it and re-test. Round after round, until it makes money or it's clear it never will. |
| 05 | Prove it isn't luck | Anything profitable goes through the stress test: eight hard checks on whether the result was a real edge or a fluke. |
| 06 | Grade it | The whole thing comes back as a single letter, F to S. |
Most ideas die somewhere in the middle. That's the point of doing it here instead of in your account.
The grade is the answer
You never have to read a number. We read all of it for you (return, drawdown, how many trades, whether it survived the stress test) and turn it into one letter:
| Grade | What it means |
|---|---|
| S | An A-quality backtest that passed every stress check. |
| A | Strong simulated result, including its stress evidence. |
| B | Profitable, but with caveats. |
| C | Around breakeven or not yet convincing. |
| D | Losing or structurally invalid. |
| F | Heavily losing or too thin to grade reliably. |
The initial backtest earns a provisional F–A grade. The hunt then runs all eight stress checks and folds that evidence into the final letter. An S can only come from an A-quality backtest that passes every check.
The grade is calculated from the results, not decided by the AI, so a strategy can't be talked into a better letter. A suspiciously perfect backtest (huge returns off a handful of trades) is caught and marked rather than rewarded. See Backtesting in Horizon for the full grading rules.
Keep it or kill it
When a hunt finds something worth reviewing, it creates a prospect with a final grade. That prospect waits under Prospects for one of two decisions:
- Approve: it moves to your Approved shelf, sorted best grade first.
- Kill: it goes to the graveyard.
Nothing is deleted. A killed strategy keeps its chat, its code and its backtests; you can open it any time. Killing is a filing decision, not a delete.
We learn from the dead
This is the part that compounds. Every killed strategy is buried with a short note on why it died: the costs ate it, it only worked in one stretch of history, it fell apart when the settings were nudged.
Before every new hunt, we read those notes. So the second hunt is better informed than the first, and the fiftieth is better informed than the second. That's why it gets better every time you run one.
Your quant levels up
Every hunt you finish (approved or killed) counts as a finished idea. Your first finished idea unlocks a Learn straight away, then every 3 after that.
A Learn re-reads the whole graveyard, rewrites your quant's playbook, and gains a level. The playbook is on the Quant Mode page under "What your quant believes", in plain English, so you can always see what it has worked out.
From Approved to a portfolio
An alpha hunt researches one strategy. It does not target, join or change a portfolio.
After you approve the prospects you trust, open My Scripts → Approved and click Build best portfolio. HorizonAI chooses a complementary line-up from their pinned backtests, then opens the Optimizer so you can compare Steadier and Faster sizing for the same strategies.
You can also create a portfolio manually and add Approved strategies yourself. Approval always happens first; a prospect waiting for review is never placed into a live line-up.
Take it to your platform
An approved strategy is a normal HorizonAI strategy. Open it and it behaves like anything else you've built: you can chat to it, change it, re-test it. When you're happy, export it to TradingView, MetaTrader 5 or NinjaTrader and trade it where you already trade.
Or don't click the button at all
Agents run alpha hunts for you on a schedule, overnight, while you're doing something else. Same hunt, same grading, same graveyard. You just come back to the results instead of starting them.
Fine print
- A flat cost of 20,000 premium tokens per hunt, taken when the hunt starts, no matter how much work it takes. On Pro and Elite it comes from the paid premium allowance first, then bonus tokens. On the free trial it comes from the trial's one-time pool. See Models, tokens and usage.
- Your free trial can run hunts. The one-off 300,000-token trial pool funds roughly 14 if none of that pool has been used by ordinary chat, inside the usual 7 day trial window.
- Pro or Elite for the ongoing monthly allowance. The trial's one-off pool funds hunts during the 7 days; after that you need a paid plan.
- Server failures are refunded. If Horizon stops the hunt before it produces a verdict, the flat charge is returned. Stopping it yourself is not a server failure.
- Agents are Pro and Elite only. Hunts you start yourself work on the trial; putting them on a schedule needs a paid plan. See Agents.
- Everything a hunt produces is a real strategy, in your chat list and in My Scripts, with its full history. There is no separate walled garden.
- Backtests are not promises. They're trades worked out against past prices, not trades that happened, and not a prediction of live results. A grade (F through S) is a score on that simulation — not a recommendation to trade, and not us going live for you. HorizonAI never connects to a broker or places orders. The stress test exists to make the number less of a lie, not to make it true. See the Risk Disclosure.
HorizonAI