tradefloorStart free

Demos

Does my trading bot pass CI?

2:20 · Recorded on the live product on 9 October 2026. The video has no sound, and a caption at the top narrates each step. Watch on YouTube

About this video

A pull request raises the momentum bot's leverage from 1.0 to 1.5 times. The pull request's CI job is one command with no AI in it: tfrun momentum.py --suite tf-stress-2026.2 --check max_drawdown<=25%. It plays the bot on 24 markets of 40 days, each with a scheduled shock, and fails if any market falls more than 25% from its peak.

The link tfrun prints opens the run on tradefloor's Benchmarks page, which shows it market by market and gives a verdict. Claude reads the CI log and changes the bot, since it can't run CI itself. The push runs CI again, and Benchmarks sets run 2 against run 1 on the same markets.

Chapters

  1. The change
  2. CI
  3. Fix
  4. Result

Figures on screen

Every figure here is on screen in the video and was checked against the product's own data.

  • Run 1 exited with code 4. On 8 of 24 markets the bot fell more than 25%, and the worst fell 60%. Benchmarks scored it 19 with the verdict “No sign of skill”, where 50 means luck.
  • Claude's change is 21 lines, adding STRESSED_GROSS, CALM_VIX, EASE_AT, STOP_AT, a setup() and a gross() function. GROSS stays at 1.5 times.
  • Run 2 exited with code 0. All 24 markets stayed within 25%, and the worst fell 20%. Benchmarks scored it 21, two points up on run 1.
  • Against buy-and-hold the bot lost on 21 of 24 markets before the change and on 14 of 24 after it.

The guide

A risk gate for a trading bot in CI builds the same check from scratch, with twelve simulated markets, one drawdown rule and a GitHub Actions job that turns the build red.

More demos

Run it yourself