Philo Park
Casual Punter
Posting this as a running archive rather than a tip sheet: match previews and analytical notes generated from my own model, which I'll keep adding to as I go through the World Cup and beyond.
Quick background on the approach
I run OddsLine, a football analytics workspace built around a probability model rather than picks. The core is a Poisson-based scoreline projection blended with Elo ratings, with an ML layer on top that accounts for the known weaknesses of a pure Poisson approach and refines the baseline further. That output gets compared against live market odds to flag where the market's price and the model's fair value diverge.
One thing worth being upfront about: the model isn't built to chase a "perfect" prediction. It's deliberately constructed from objective, pre-match information only, so it functions as a clean baseline rather than something trying to overfit to every possible signal. That's a design choice, not a limitation I'm apologising for; a baseline is only useful if it stays honest about what it does and doesn't account for.
On top of that sits a RAG-based AI layer that pulls in current team news so the numbers aren't read in isolation. More on why I split it this way (model does the stats, AI handles the context) here: Why Most AI Betting Tools Fail (and How to Actually Use AI for Betting)
What I'll be posting in this thread
Quick background on the approach
I run OddsLine, a football analytics workspace built around a probability model rather than picks. The core is a Poisson-based scoreline projection blended with Elo ratings, with an ML layer on top that accounts for the known weaknesses of a pure Poisson approach and refines the baseline further. That output gets compared against live market odds to flag where the market's price and the model's fair value diverge.
One thing worth being upfront about: the model isn't built to chase a "perfect" prediction. It's deliberately constructed from objective, pre-match information only, so it functions as a clean baseline rather than something trying to overfit to every possible signal. That's a design choice, not a limitation I'm apologising for; a baseline is only useful if it stays honest about what it does and doesn't account for.
On top of that sits a RAG-based AI layer that pulls in current team news so the numbers aren't read in isolation. More on why I split it this way (model does the stats, AI handles the context) here: Why Most AI Betting Tools Fail (and How to Actually Use AI for Betting)
What I'll be posting in this thread
- Match previews starting with the current World Cup, expanding out to cover most of Europe's major domestic leagues as well as the Champions League, Europa League, and Conference League once the new season gets underway. Each write-up walks through the statistical case, the counter-case, the likeliest match pattern, and where that leaves the betting markets
- These are analytical reads, not picks. Where the model and market disagree, I'll say so, but what anyone does with that is their own call
- I won't be discussing personal bet positions in this thread. I build the model, so posting my own stakes would be a conflict of interest
18+, for research purposes; usual disclaimers apply. Happy to take questions on the modelling side (Poisson/Elo/feature engineering) if anyone's interested in that.
Last edited:
