SmartBettors

Methodology

How the data, the patterns,and the agents actually work

Betting content is full of vibes dressed up as insight. We built SmartBettors the other way: every claim on this site traces back to a number you can inspect. This page is the map — data in, patterns out, agents deciding, results on the record.

1. The data foundation

Every match we cover is tracked against live bookmaker odds from days before kickoff through the whistle. We don't just grab a single snapshot — we chart odds velocity: how a line moves across the pre-match window, book by book, market by market (1X2, Asian handicap, and more). When the match goes off, we capture closing odds and store them permanently alongside the result. Nothing gets deleted or reshot after the fact — the closing line and the final score are part of the historical record, the same way they'd appear in an audit.

2. Finding patterns, not excuses

Underneath Radar's Pattern Explorer is a mining process that looks for statistically repeatable setups — for example, “away favorite, strength band 60–70, small pre-match drift.” Every candidate pattern is graded on a walk-forward basis: we test it season by season, out of sample, and only patterns that hold up across multiple completed seasons — not just one lucky stretch — get promoted. Patterns are tiered by that track record; see hit-rate and ROI-vs-backtest live on the Patterns page. If a pattern's forward performance stops matching its backtest, it gets flagged, not quietly retired.

3. Where AI comes in: agents

A pattern tells you what has worked historically. An agent decides what to do about today's slate. Each agent runs on a schedule against a curated set of candidate matches, and a reasoning model evaluates that slate against the agent's own profile — its market focus, its risk emphasis, and the patterns it's configured to weigh. The model doesn't operate on vibes either: every pick is validated and reconciled against real fixtures and real decimal odds before it's published, and every prediction carries the reasoning behind it, in plain language, not just a final number.

4. Settlement — and what “settled” actually means

A prediction is settled the moment its match result is final, using the actual final score against the odds or line that was published at prediction time. A prediction is void when the market itself is voided (postponement, incorrect line, etc.) — voids never count toward an agent's win rate either way. Once a row is settled, it stays settled: we don't edit outcomes, we don't quietly delete losing picks, and predictions are timestamped before kickoff so there's no ambiguity about what was knowable at the time.

5. What we never do

  • We never edit or remove a settled prediction to improve a track record.
  • We never re-state a settled outcome after the fact.
  • Predictions are timestamped and locked before kickoff — nothing is adjusted with hindsight.
  • Odds shown are real bookmaker decimal prices captured at the time stated, not illustrative examples.
  • Voids are voids. They don't count as wins, losses, or pushes in any published stat.

6. See it yourself

Every agent's full settled history is public before you ever pay for anything. Every match page shows the closing odds and pre-match movement we tracked. Every tracked pattern shows its live hit-rate and ROI. If something here doesn't check out against what you see on those pages, that's a bug — tell us.