What Ask AI sees in Algo
Algo hands Ask AI the numbers of the view you are on, not a text scrape.
| View | What is sent |
|---|---|
| Backtest report | Strategy, underlying, dates, stats, tearsheet, legs, entry and exit rules, combined stop and target, warnings, and daily P&L — for long tests, the 15 worst days, the 15 best and the last 30 |
| Sweep results | The date range and the top rows of the sweep table |
| Strategy builder | The name, symbol, start and square-off times and every parameter you have set |
| Strategies and portfolios | Each run's status, name, MTM and peak MTM |
| Signals | Each signal's name, on or off, source and route count, plus recent logs |
Because the report arrives as data, the answer can do arithmetic on it — add up the best days, compare win rate with average loss — rather than read numbers off a scraped page. Other views send their visible text. More in how Ask AI reads your screen.
What it can and cannot do
It can explain the stats, locate where the drawdown and the profit came from, translate legs and rules into plain words, compare sweep variants, and say why a signal may not have fired from its logs.
It cannot build or edit a strategy, start a backtest, deploy or stop a run. The builder and the Run button stay yours. Building and running backtests by asking is a separate feature for Claude and ChatGPT — see backtesting by asking.
Questions that work, view by view
| View | Ask |
|---|---|
| Backtest report | “Explain these backtest results simply.” · “Why did this strategy lose money on its worst days?” · “How much of this profit came from the best few days?” |
| Sweep | “Which variants here did best, and why?” · “Is the best variant in this sweep likely overfitted?” |
| Builder | “Explain this strategy in plain words.” · “What are the risks of these legs?” |
| Strategies | “Which of my runs are losing, and by how much?” |
| Signals | “Why did this signal not fire?” |
Every one has a screen word. “Which variants did best and why?” without here or this sweep is treated as general and skips the table.
A worked example: concentration of profit
Say a one-year report shows net P&L of ₹1,20,000 over 240 trading days. Example numbers.
Ask: “How much of this profit came from the best few days?” The report carries the best and worst days, so the answer can add them up. Suppose the five best days total ₹95,000. That is 95,000 ÷ 1,20,000 ≈ 79% of the year's profit from about 2% of the days.
That is the point to stop and ask: “Is this result reliable, or could it be luck?” The answer explains that a result carried by a handful of days is fragile, that past results do not predict future ones, and what to check next — for example whether those days share an event. The report's own tools are covered in the backtest report and parameter sweeps.
Limits
- Explains, does not run. No building, running, deploying or stopping.
- Long reports are summarised to the worst, best and recent days before sending.
- Not a forecast. It will not say a strategy will make money.
- Credits. Data-heavy reports use more AI credits than short questions; see /pricing.
How to open it
- Open Algo & Backtesting and go to the report, sweep, builder or Signals view.
- Click the Ask AI button at the bottom-right; each view offers its own suggested questions.
- Edit a suggestion before sending if you want a different angle, keeping its screen word.
Common questions
Can Ask AI read my backtest report?
Yes. On the report view it receives the stats, legs, rules, warnings and daily P&L directly from the page.
Can Ask AI tell if a backtest is overfitted?
It can point out warning signs in the result on screen, such as a few days carrying the profit or the best sweep variant standing alone. It cannot prove future performance.
Can Ask AI build a strategy in Algo?
Not from the in-app panel. It explains strategies; building and running backtests by asking is done from Claude or ChatGPT through the MCP connection.
Does Ask AI see every day of a long backtest?
For long tests it receives the worst 15 days, the best 15 and the last 30, alongside the full stats and tearsheet.
Why did my sweep question ignore the table?
It probably had no screen word. Add 'here' or 'in this sweep' so the table goes with the question.
More on Algo & Backtesting
More in Algo & Backtesting · App: Algo & Backtesting · Definitions: glossary · All: every article