Getting started

Welcome to AlphaProve. Here's the short version: sign in, describe a strategy in plain English, and let the assistant write, backtest, and score it for you. No coding and no API calls: everything on this page happens by clicking in the app.

What you need

Just an AlphaProve account. New accounts start on the free tier, which is enough to describe strategies and run backtests within the quota and lookback caps.

1. Sign in

Open the app and sign in with your email and password. Don't have an account yet? Request private-beta access. New requests need approval before you can sign in and open the workspace.

2. Find your way around

The sidebar on the left is your map. You'll spend most of your time in the first few items:

• AI Chat: describe a strategy or research your saved strategies and results.
• Chat History: every conversation, so you can pick up where you left off. Filter by phase, or export a conversation as a Markdown file. Conversations from the former AI Trader open here too, and old AI Trader links redirect to the same chat.
• Library: your saved strategies, ready to re-run or tweak.
• Results: the tearsheet from every backtest you've run.
• Forward Walk and Risk: deeper tuning once you have a strategy you like.
• Editor: write the strategy code yourself (optional: see the end).

3. Describe your strategy

On AI Chat, tell the assistant what you want to test in plain English. Not sure where to start? The greyed-out example shows the kind of thing that works, and the starter chips below the box give you one-click ideas.

Type your own idea (an entry rule, a symbol, a timeframe) and press Send.

4. Answer a couple of questions

Before it writes any code, the assistant fills in the gaps. Here it asks how you want to exit the trade and manage risk, and offers sensible presets you can click, or you can type your own answer.

Answer questions in their question card. You can use the main composer for a separate research question without answering a pending strategy question. Once the specification is ready, select Generate code, then review the strategy before you run a backtest. Suggestions never place trades or start paper trading.

5. Backtest and read the results

Run the backtest straight from the chat. Before it starts, a Risk & sizing panel lets you review the market, date window, capital, and leverage the assistant picked. Adjust anything, or just run with the defaults.

For a saved strategy you can also ask the assistant for a backtest plan: you review the exact settings on a card and accept it. The same chat answers market questions and researches the web with cited sources too; see the AI Chat guide.

Results stream in live and settle into a full tearsheet: an equity curve, the headline numbers (total return, Sharpe, max drawdown, win rate, trade count), and the list of trades. Every run stays available later under Results in the sidebar.

The app, page by page

The walkthrough above is the fast path. This section is the reference: every item in the sidebar, what it's for, and every choice you can make on it, in plain terms. Some pages are part of the Trader or Quant plans; that's noted where it applies.

New Strategy

This is the home screen and where most work starts. You describe a trading idea in plain English and the assistant turns it into a working strategy.

• The prompt box: type your idea, then press Send (or the Enter key). A greyed-out example inside the box shows the kind of thing that works.
• Starter chips: one-click ideas below the box (a moving-average "golden cross", an RSI oversold bounce, a MACD crossover). Clicking one fills the box so you can tweak it before sending.
• More examples: for more ideas than the starter chips, the Examples page collects complete strategies to learn from.
• Clarifying questions: before writing anything, the assistant asks about anything you left out (which market to trade, how to exit, how much to risk) and offers quick-pick answers, or you can type your own.
• "I understood this as": the assistant shows its understanding of your idea for you to approve with Looks good → Generate, or correct by replying in the chat. If part of your idea can't be built, it tells you which part.
• Run backtest: once the strategy is written, this button opens the Run settings panel (below) and tests it.

The Run settings panel

Whenever you start a backtest (from here, from Library, or from Risk), a panel lets you review and adjust everything about the test. The assistant fills in sensible defaults; change what you like, or just press Run.

• Market & window
  • Symbol: which market to trade, such as BTCUSDT. (Default: BTCUSDT.)
  • Timeframe: the candle size the strategy trades on, from 1 minute up to 1 day. Smaller candles mean more, faster trades. (Default: 1h.)
  • Start and end date: the slice of history to test over. Both days are included in full (UTC); today is left out because its candles are still forming. You can only pick dates real market data exists for; the panel shows the available range. (Default: roughly the last 90 days up to today.) Your plan limits how far back you can go.
• Capital & leverage
  • Initial capital: the pretend starting balance for the simulation. (Default: 10,000.)
  • Leverage: how much borrowing the account is allowed. Higher leverage means bigger swings and an earlier "liquidation" (forced close) point. It does not decide how big each trade is. Your plan sets the ceiling: Free 10×, Trader 50×, Quant 100×. (Default: 1×, i.e. no borrowing.)
• Risk & sizing
  • Position sizing sets how the size of each trade is decided:
    • Risk per trade: every trade risks the same small slice of your balance, worked out from how far away your stop is. (The default.)
    • Percent of equity: each trade uses a set percentage of your balance.
    • Fixed notional: every trade is the same dollar amount.
    • Fixed contracts: every trade is the same number of units.
    • Leverage-based options: the two above, scaled up by your leverage.
  • Dynamic risk: optionally shrink your risk automatically when you're losing (as losses deepen, or after a losing streak). Off by default; when on, a risk floor sets how low it's allowed to go.
• Strategy parameters: if the strategy has adjustable knobs (moving-average lengths, an RSI threshold, a stop style, and so on) they appear here with the strategy's own defaults, so you can tune it without touching any code.
• Fill model: fills are simulated on 1-minute candles. A market order fills at the next bar's open, plus slippage. For crypto symbols with recorded order books (BTCUSDT from 2023, four more from April 2025; candles go back further), market orders and stops are priced by walking the book instead, with no slippage added, and stops and limits are timed on recorded trades, on every plan. Slippage then applies only where no book exists. Outside the books, on Trader and Quant you can turn on Tick fills to time stops and limits on recorded trades (see Fill model). Limits fill at their own price and have no queue-position modelling.
• Advanced
  • Margin mode: cross means your whole balance backs every position; isolated gives each position its own walled-off margin. Cross survives volatility longer; isolated caps the damage from a single bad trade. (Default: cross.)
  • Max open positions: the most trades allowed open at once. (Default: 2.)
  • Max drawdown stop: an automatic circuit-breaker that halts the run if the account falls this far from its peak. Off by default.

Library

The Library lists the strategies you have pinned: a strategy you add with Add to library in the chat, Save to library in the Editor, or create by saving a new strategy in the Editor. Other strategies, such as AI drafts you never added, are still reachable from the Editor's Load list but do not appear here. Pinned strategies are ready to re-run, tune, forward-walk, or publish.

• Views and filters: quick tabs (All, Recently tested, Best Sharpe) plus a Filters panel to narrow the list by symbol, timeframe, quality (Sharpe range), worst drawdown, age, whether it's been tested, and whether it's editable. Once your filters match no tab or saved view, a + Save as view button appears next to the tabs: name it and it becomes your own tab (× deletes it). You can also sort by most recent, best Sharpe, or name.
• What each row shows: the strategy's name, its status (never tested, running, completed, failed), timeframe, return, worst drawdown, and a small equity sparkline of the latest run.
• Per-strategy actions
  • Test / Test again: run a backtest (opens the Run settings panel).
  • Cancel / Retry: stop a run in progress, or retry one that failed.
  • Forward walk: stress-test it on unseen data. (Trader and Quant.)
  • Edit code / View code: open it in the Editor. Editing needs a paid plan; AI-written strategies may be view-only.
  • Publish: list it on the marketplace. (Quant, once it has a completed backtest.) Publication shares the code or configuration and parameter settings used by the selected run, listing title and description, and displayed verified results. Later edits do not change that published copy. Run and publish again to share a changed strategy. Your account email and original chat prompt are not included automatically. Review the text and code you choose to share before publishing.
  • Rename / Delete.
• Bulk actions: tick several strategies to backtest or delete them all at once. How many backtests can run at the same time depends on your plan (Free 1, Trader 3, Quant 5).
• Strategy detail: clicking a name opens its own page: a summary of its last backtest, the equity curve, a link to the full tearsheet, the original prompt that created it, and a copy of its configuration you can copy to the clipboard.

Marketplace listings marked Author update needed need a new run and publication before anyone can import them again. Their older results are hidden because the tested version cannot be confirmed. Copies already in your library remain available. New imports keep the published code and tested parameter settings. Choose the market, date range, and other run settings for your own tests; you can also change the parameters.

Results

Every backtest you run is kept here, along with the full report, the tearsheet, for each one. The list can be filtered by status and sorted by any column, and it refreshes itself while runs are still working. From the list you can open a run's tearsheet, or cancel and delete runs.

The tearsheet explains how a strategy did. The headline numbers, in plain terms:

• Net PnL: the bottom line in dollars, how much the strategy made or lost.
• PnL %: the same result as a percentage of your starting balance.
• Sharpe: a quality score that rewards steady gains and penalizes a bumpy ride. Higher is better; above about 1 is decent, above 2 is strong.
• Win Rate: the share of trades that made money. On its own it doesn't guarantee profit: a few big losses can outweigh many small wins.
• Max Drawdown: the worst drop from a high point before recovering. Smaller is safer.
• Trades: how many trades the strategy made.
• Extras: a row of smaller stats covering profit factor (dollars won divided by dollars lost), expectancy (the average result per trade), the average win and loss, and the best and worst single trades.

It also draws an equity curve (your balance over time, green above your starting point and red below) and a price & trades chart (the market's candles with your entries and exits marked: green for winning exits, red for losing). Runs with enough trades get more cards: best & worst trades (from 2 trades), an exit reasons breakdown (from 5 trades: how trades closed: hit their profit target, hit their stop, closed on a signal, flipped, or were liquidated), and a P&L distribution (from 8 trades: the shape of your results, many small wins versus a few big ones). A run with fewer trades simply leaves those cards out. Below the charts, the trade log lists every trade with its entry and exit price, size, profit, fees, and how long it was held. The equity curve and the trade log can both be downloaded as spreadsheets.

Forward Walk

Trader and Quant

Forward Walk is part of the Trader and Quant plans.

A backtest can look great simply because you kept tuning it until it fit the past. That's curve fitting, and it rarely survives real markets. Forward Walk guards against it: the app repeatedly tunes the strategy on one stretch of history, then tests those settings on the next, unseen stretch, and joins the unseen results end to end. If the strategy still makes money on data it never saw, its edge is more likely real.

• Setup
  • Pick a strategy: it must have adjustable parameters to tune; the app shows which ones will be tried.
  • Market, dates, capital, leverage, sizing: the same choices as a normal backtest (see the Run settings panel above).
  • In-sample window (months): how much past data to tune on at each step. (Default: 4.)
  • Out-of-sample window (months): how much later, unseen data to then test on. This is the honesty check. (Default: 2.)
  • Anchored vs rolling: rolling keeps the tuning window a fixed length that slides forward; anchored always starts at the same point and grows, so each step learns from all the history so far. (Default: rolling.)
  • Parameter grid: a read-only preview of which knobs will be tried and over what values (these are set inside the strategy itself), plus how many combinations that adds up to and whether it's within your plan's limit.
• Reading the results
  • WFO efficiency: the headline score, how much of the tuned-in performance survived on unseen data. Roughly, 0.70 or higher is strong and below 0.40 suggests the strategy was curve-fit.
  • Out-of-sample totals: the profit, number of trades, win rate, and worst drawdown across all the unseen windows.
  • Stitched equity curve: a single line of only the unseen results, the closest thing to "how would I actually have done".
  • Per-step table: for each window, which settings were picked and how they performed on unseen data, so you can see whether the best settings stay steady over time or jump around.

Risk

Trader and Quant

The Risk page is part of the Trader and Quant plans.

This page is where you set how much you're willing to risk and stress-test a strategy's results before trusting them.

• Leverage and margin: your plan sets the maximum leverage (Free 10×, Trader 50×, Quant 100×), and you choose cross or isolated margin. The risk settings themselves (leverage, position sizing, dynamic risk, and stops) are the same ones in the Run settings panel described under AI Chat above.
• Monte Carlo: takes one backtest and replays it thousands of times with the trades shuffled, some skipped, or the settings nudged, to show the whole range of outcomes luck could have produced, including the chance of wiping out the account and a typical worst-case drawdown. It answers "was my result skill, or a lucky order of trades?"
• Prop Firm: simulates how often the strategy would pass a funded-trader evaluation (reaching a profit target without breaking a loss limit) and shows the pass/fail odds.

Editor

Trader

Writing and editing code in the Editor is part of the Trader plan and above. Free users can view an AI-written strategy's code but not change it.

The Editor is the optional, advanced path. Instead of describing your idea to the assistant, you write the strategy yourself, with full control over every detail. Most people never need it. The assistant handles the same job in plain English.

It's a proper code workspace with live hints for the building blocks, automatic error-checking, and a few actions:

• Load a strategy: open one of your own to keep working on it.
• Save: checks your code for problems, then saves it (asking you to name it if it's new). The app refuses to save code it judges unsafe to run.
• Save to library: pins the strategy to your Library so it's easy to find.
• Test: sends the strategy to the run screen to backtest it.

If the strategy changed somewhere else after you opened it (in another tab, or by Apply to original in the AI Chat), Save stops and says so. Choose Reload to see the latest version, or Save as new to keep yours as a separate strategy.

A brand-new strategy starts from a small working example rather than a blank page, so you always have a shape to build on.

When you want the details of writing a strategy by hand, these pages go deeper:

• Strategy authoring: the rules for writing, validating, and submitting a custom strategy.
• SDK reference: the full toolkit your code works with.
• Examples: complete strategies to learn from.

Saving limits

Chat messages and strategy prompts can contain up to 100,000 characters. Strategy names and marketplace titles allow 200 characters, descriptions 5,000, and strategy code 200,000. If a request is too large, shorten the text or reduce the settings included; the app rejects the request without saving a shortened version.

Customized card layouts are saved separately for each of your results, strategies and Lab campaigns. You can save 500 layouts in total and 20 named templates per page type. Existing saved layouts remain editable at the limit. Layouts belonging to deleted items may be cleared automatically when space is needed. Resetting the card order to standard changes the arrangement; it does not remove its saved-layout entry.

Settings

Everything about your account lives here.

• Profile: your email address, your current plan, and Sign out.
• Backtest defaults: a default market, timeframe, starting capital, and leverage that pre-fill every new backtest, so you don't set them each time.
• Usage: the AI credits you have left this month and when they reset. Every plan has a monthly allowance (Free 20, Trader 300, Quant 1,500); none is unlimited.
• Agent access: the AI agents (MCP connections) you have authorised, and revoking them. See MCP.
• Plan: a side-by-side comparison of Free, Trader, and Quant and what each includes. Online upgrades are not open yet, so the Upgrade buttons are disabled for now. When paid plans open, Upgrade will go to secure Stripe checkout and changing or cancelling a paid plan to the Stripe billing portal.