Solaris / build log
Pot: £674.07 (+34.8%)

Building Solaris: My £500 Live Trading Bot Log

I am building an autonomous desktop trading app in Python that uses a small committee of AI agents and hard math rules to trade 17 global markets. Rather than testing in a vacuum forever, I put £500 of my own money into an IG spread betting account to see what actually happens on live spreads. Each week it runs on its own, I fix whatever breaks, and each weekend I post the honest update here.

Started: £500.00 (15 Nov) Current Cash: £674.07 Closed Deals: 26 Bot Status: Sleeping for weekend

Week 1: £500 to £674, two painful index losses, and rebuilding the bot midweek

Last Sunday night I topped my IG account up to an even £500.00 and switched Solaris onto live autopilot before going to bed. Demo accounts are great for checking whether orders fire, though you learn ten times faster when real money is on the line and you have to pay real bid and ask spreads plus overnight funding.

The basic idea behind Solaris was to stop a single language model from guessing trades or burning API tokens every five minutes. A plain Python loop watches 17 indices, commodities, and currency pairs for free in the background. Only when something meaningful moves does it wake up four separate specialist prompts (one looking at chart trends, one checking IG retail crowd positioning, one reading breaking news, and one watching bond yields and the economic calendar). A fifth prompt acts as a risk critic whose sole job is to poke holes in their ideas, and a sixth makes the final call and writes a review after every trade closes.

A 60 second capture of the Solaris v2 desktop app showing the specialist reports, the RSI momentum check, and the Week 1 trade log.

How the first week actually played out

On paper, finishing Week 1 at £674.07 (+34.81% after £3.42 in overnight financing and dividend adjustments) looks like a smooth ride. Watching the logs day by day was anything but smooth.

Monday and Tuesday started steadily on Gold, Nasdaq, and Silver, and by Wednesday morning a clean short on Japan 225 (+£88.67) pushed the account all the way up to £753.88. Right after that peak, three glaring flaws showed up back to back and forced me to build Solaris 2 midweek:

There were plenty of bright spots too. My favourite moment of the week happened at 12:57am on Friday morning on EUR/JPY. A sudden Tokyo flush pushed open profit past 80% of target and crushed hourly RSI down to 13.0 right above the previous day's low. Instead of holding out for the last few points and riding a snapback up, the bot closed early at market and banked £26.22, capturing 99% of the move's peak.

What I patched before shutting down on Friday

On Friday lunchtime I retired Solaris 1 completely, moved Solaris 2 onto the live account, and pushed a batch of fixes based on everything that broke during the week:

Week 1 Curve and Full IG Trade History

Every execution from the live account between Sunday 15 November and Friday 20 November is listed below. Gross profit across the 26 closed deals was £177.49, less £3.42 in overnight swap interest and a Nasdaq dividend adjustment, leaving £674.07 in cash for the weekend.

£500.00 → £674.07 (+£174.07 net) Midweek High: £753.88 · Thursday Low: £583.46
£500 £753.88
19 wins · 1 breakeven · 6 losses (Deals: +£177.49 | Swaps & Divs: -£3.42)
Date Market Side Stake P&L What Happened

Plan for Week 2

Solaris 2 wakes back up tonight when markets open for the new week, running solo on the £674.07 balance. I am leaving the strategy code alone all week so I can get a clean five day read on the new upgrades, with one eye on thinner US market hours around Thanksgiving on Thursday and Friday. Next update goes up next weekend, win or lose.

Follow the weekend updates

If you want an email when I post next weekend's numbers and code notes, drop your address below.