← THE CHARTSPOD LEDGEROPEN TERMINAL →
AI · BACKTEST RESEARCH

AI Crypto Backtest Research: A Prompt That Requests Evidence and Invalidation

An AI backtest prompt should specify everything that makes backtests unrealistic, so the agent can't quietly assume the optimistic version.

By ChartsPod Editorial Team · Reviewed by ChartsPod Research
Published · Updated · 5 min read

Short answer

An AI backtest-research prompt should define the rule precisely, the market, timeframe and date range, signal timing (closed bars), fill model (next-bar open plus slippage), fees per side, funding at settlement times for perpetuals, and an out-of-sample period. Require trade count, net return, max drawdown, profit factor, expectancy after costs, the trade list, and the number of variants tried. Ask the agent to state anything it could not model.

The prompt

PartText to include
Rule'Long when 20-EMA > 50-EMA and close > 20-EMA; exit on close < 50-EMA.'
Data'Venue A ETHUSDT perp, 4h, 2023–2025; 2025 out of sample.'
Execution'Closed-bar signals, next-open fills, 5 bps slippage.'
Costs'0.05% per side; settled funding at fundingTime.'
Report'Trades, net return, max DD, PF, expectancy, trade list, variants tried.'
Illustrative: in-sample vs out-of-sample net return (%)Illustrative: in-sample vs out-of-sample net return (%): In-sample +38 %; Out-of-sample +6 %.ILLUSTRATIVE: IN-SAMPLE VS OUT-OF-SAMPLE NET RETURN (%)In-sample+38 %Out-of-sample+6 %
The out-of-sample number is the one to judge. Original ChartsPod editorial graphic.

Evidence to require

The answer must includeThe answer must include: Exact rule; In- and out-of-sample split; Fill model; Fees and funding; Full metric set; Variants tried; Unmodelled items listed.THE ANSWER MUST INCLUDEExact ruleIn- and out-of-sample splitFill modelFees and fundingFull metric setVariants triedUnmodelled items listed
For backtest research, reject answers that skip these. Original ChartsPod editorial graphic.

Red flags in the output

  • Only equity curve shown
  • No variant count
  • Fills at signal close
  • No out-of-sample period

Related: AI Crypto Backtests: Check Fills, Fees and the Number of Trials, How to Backtest Crypto Strategies Without Lookahead Bias.

Paste the backtest research prompt into ChartsPod's agent and open each tool call it makes.

Open the ChartsPod terminal

Ask the agent to state anything it could not model.

Compare AI Crypto Backtest Research across Binance, Bybit, OKX and Hyperliquid markets.

Open markets in ChartsPod

Questions readers ask

What should an AI backtest prompt include?

Rule, data, timing, fills, costs, OOS period and full reporting.

Why ask for variants tried?

The best of many is biased.

What if the agent can't model something?

It should say so.

Sources and methodology

ChartsPod's AI agent (69 tools, visible tool calls, bring-your-own-key, no order placement) was checked on its public pages on 17 October 2026. Prompts are illustrative templates; outputs vary by model and data.

  1. ChartsPod: product overview, pricing and FAQ — free terminal, venues, modules, no order placement; checked 17 Oct 2026.
  2. ChartsPod documentation — guides for order flow, derivatives, alerts, backtests, journal and workspaces.

This article explains research tools and workflows. It is not financial advice, does not recommend buying or selling any asset and does not predict prices. Crypto derivatives carry a risk of rapid and total loss of margin.

CHARTSPOD EDITORIAL EDITION · RESEARCH AND EDUCATION ONLY · NOT FINANCIAL ADVICE.