I had an AI coding agent build a private trading journal from my MT5 history. It reads my trades, never touches an order, and rebuilds itself while I sleep. Explore the real dashboard, test how many trades a number needs, and leave with the prompt for your own.
See how it worksTrading is a numbers game, and nobody can remember every trade. Without a journal, decisions rest on how the last few trades felt, and improving your trading becomes unnecessarily hard.
Claude Code built this journal on my own computer in about 15 minutes. The agent wrote the code once. Every night since, a short Python script does the work, with no model involved and nothing to pay.
Five steps, from the first prompt to tomorrow morning's dashboard.
I described the journal in one prompt. Claude Code wrote, ran and tested the code on my machine: the core in about 15 minutes, then most of an hour of extensions.
The whole session used 168,000 output tokens, about ten US dollars at API prices. On my plan it cost nothing extra.
Windows Task Scheduler starts a short Python script. No model, no tokens and no subscription: the refresh runs without Claude Code being open.
The computer has to be on and MT5 logged in. A missed night is caught up by the next run, because the script reads the full history each time.
Only one file imports the MT5 package, and it may call five read functions. A test fails if any other function appears in the code.
The password is never stored: the script reads from the terminal that is already logged in. Nothing flows back to the account.
Account number, server, broker name and broker-specific symbol suffixes are dropped. Amounts become percentages or R-multiples.
A final check searches the output for those identifiers and refuses to write the file if any of them appear.
The script writes a single file you open in a browser. When a refresh fails, it keeps the last working dashboard and writes the error to a log file.
The set-up described in the article. Arrows show the direction data moves; there is no path from the script back to the account.
This is the journal from the article, built on my own MT5 account and anonymised on import, so amounts appear only as percentages or R-multiples. Switch tabs, hover the charts, sort the trades.
Trading journal
Closed-trade balance, indexed to 100 at the first deposit. Deposits and withdrawals excluded.
Percentage below the previous peak of the equity index.
R and execution
Number of trades per result bucket, in R, net of costs.
Initial risk as a percentage of balance at entry, per trade in time order. Line = median.
Statistical significance
Is the edge distinguishable from zero, or could the result still be luck? Trades are treated as independent draws.
Closed trades only. Times in UTC. Result = trade P&L as a percentage of balance at entry. Select a column to sort; scroll sideways for R, costs and slippage.
Grey rows have fewer than 30 trades: too small a sample to draw conclusions from. EV = mean result per trade. EV (R) = mean R-multiple of the trades in the group that had a stop loss. Sum = simple sum of trade results, not compounded. Slippage: negative = filled worse than requested.
My own trades, from one personal account. History, shown to explain how the journal works: not advice and not an indication of future results. The equity index excludes deposits and withdrawals; Sharpe and Sortino use daily returns annualised with the square root of 252 and a risk-free rate of zero.
Our brains are poor judges of probability. We give too much weight to what happened last: a setup that wins three times in a row starts to feel like it always works, even if it only wins 40% of the time over a hundred trades.
Short-term results are largely luck. Given enough coin flips, five heads in a row is all but certain to happen somewhere. Set a win rate and a number of trades, and see how often a streak shows up anyway.
Exact probability for independent trades that each win with the same fixed chance, computed in your browser. A win rate of 50% is a fair coin.
As a rough guide, treat 30 trades as the absolute minimum, 50 as preferable and 100 or more as the point where the data starts to become reliable.
The trap is that this applies to every slice. A hundred trades spread over five weekdays leaves 20 per day, which is not enough to conclude that Fridays are bad. Dropping Fridays anyway is overfitting: tailoring your rules to noise.
Take your average result per trade, in R, and divide it by its standard error: the standard deviation of your results divided by the square root of the number of trades. That gives a t-statistic. The p-value is the probability of seeing results at least this good if there were no edge.
The usual threshold is a p-value below 0.05, for a single test planned in advance. It assumes your trades are independent and not extremely lopsided. Be stricter for anything you find by scanning many slices of the data.
Student's t with n − 1 degrees of freedom, one-sided. Minimum track record length after Bailey and López de Prado (2012), for a Sharpe ratio above zero at 95% confidence, on per-trade results, so the answer is a number of trades: 1 + [1 − skew · SR + (kurtosis − 1) ÷ 4 · SR²] · (z ÷ SR)², with SR = average ÷ standard deviation.
Read the metrics as a panel rather than one at a time. The Sharpe ratio hides how lopsided your results are, and maximum drawdown hides how long it took to recover. If Sortino sits well above Sharpe, your results lean on occasional large winners; if the two are close, look for large losses hiding in your results.
On a few months of trades, treat these ratios as estimates. Andrew Lo showed in 2002 that serial correlation can overstate an annual Sharpe ratio by as much as 65% in the hedge fund examples he studied.
Gain needed = drawdown ÷ (1 − drawdown).
The Python package that connects to MT5 can also place and close trades. I told the agent to use only the functions that read history, and to add a test that fails if any order function appears in the code.
The agent proved the test works by temporarily adding a call to place an order, watching the test fail, and deleting it again. Try it: add a function to the MT5 module and run the test.
The five read functions are the ones the journal uses. Order functions, and anything else not on the list, make the test fail.
Choose your language, refresh schedule and server time zone. The prompt below is the standard one behind this kind of journal, with your three choices filled in. Paste it into Claude Code in an empty folder, with MT5 open and logged in.
Windows and the MT5 desktop terminal only. The MT5 Python package does not work with the web terminal or on Mac and Linux. On a Mac, you can paste the prompt anyway and ask Claude Code whether it can build the same thing there; that route is untested. Claude Code needs a paid Claude plan or API credits.
Nothing you choose here is stored or sent anywhere. When the dashboard is ready, check one metric by hand on five trades before you trust the rest.
A journal is only progress when you come back to it.
Every 30 trades, or at least once a month, analyse how your trading is doing and where it can improve. Otherwise you are tracking data for the sake of tracking data.
There is no simulation on this page. Every figure is the author's own anonymised data, a formula from the sources below, or computed in your browser from what you enter. Purple, which publishes Hello Purple, builds technology that connects AI agents to trading accounts.
This article is information and commentary. It is not investment advice, not a personal recommendation and not investment research, it takes no account of your circumstances, and it must not be relied on as a reason to trade. Any figure shown is historical or simulated and is not an indication of future results.
Get the next piece the day it is published.
One email when a new report lands, nothing else.