Ok, this is my absolute last try to outsmart the bookies..
Been building a fully automated football betting engine over the last few weeks and finally wiped the test logs to start clean from today. Figured I’d track it live right here.
The whole setup runs in the cloud on GitHub Actions — no manual spreadsheets, no guesswork, and zero "gut feel." Here’s the quick rundown of how it works:
AI reads the news, Python does 100% of the math: Verified injuries, suspensions, and lineup news get parsed into strict JSON. From there, a deterministic Python script on GitHub handles all the Power de-vigging, Poisson grids, and Kelly sizing. Same inputs = exact same output every time.
Market-anchored (no phantom EV): Instead of pretending to outsmart the whole market from scratch, it anchors to the de-vigged market line and only prices the exact delta from verified team news. No draw leakage, no fake underdog bias.
Built-in "Steam Guard": I feed in pre-news odds so the engine checks how much the line already moved. If the market already priced in the injury report, it kills the bet automatically.
Full GitHub audit trail: Every play that clears the gates (+3.5% EV floor, 0.25x Fractional Kelly) gets timestamped and committed straight to a CSV log and Markdown report in the repo before kickoff. After full time, a second GitHub Action settles the bet and auto-updates Kelly ROI, Flat ROI, de-vigged CLV %, and Brier Score.
Filters are strict, so 80–90% of matches I run end up as a . No locks, no chasing action — just seeing if we can consistently beat the closing line over a real sample.
Dropping the first bet right below. Happy to answer any questions about the GitHub setup or the model.
To remove first post, remove entire topic.
Ok, this is my absolute last try to outsmart the bookies..
Been building a fully automated football betting engine over the last few weeks and finally wiped the test logs to start clean from today. Figured I’d track it live right here.
The whole setup runs in the cloud on GitHub Actions — no manual spreadsheets, no guesswork, and zero "gut feel." Here’s the quick rundown of how it works:
AI reads the news, Python does 100% of the math: Verified injuries, suspensions, and lineup news get parsed into strict JSON. From there, a deterministic Python script on GitHub handles all the Power de-vigging, Poisson grids, and Kelly sizing. Same inputs = exact same output every time.
Market-anchored (no phantom EV): Instead of pretending to outsmart the whole market from scratch, it anchors to the de-vigged market line and only prices the exact delta from verified team news. No draw leakage, no fake underdog bias.
Built-in "Steam Guard": I feed in pre-news odds so the engine checks how much the line already moved. If the market already priced in the injury report, it kills the bet automatically.
Full GitHub audit trail: Every play that clears the gates (+3.5% EV floor, 0.25x Fractional Kelly) gets timestamped and committed straight to a CSV log and Markdown report in the repo before kickoff. After full time, a second GitHub Action settles the bet and auto-updates Kelly ROI, Flat ROI, de-vigged CLV %, and Brier Score.
Filters are strict, so 80–90% of matches I run end up as a . No locks, no chasing action — just seeing if we can consistently beat the closing line over a real sample.
Dropping the first bet right below. Happy to answer any questions about the GitHub setup or the model.