<<<<<<< Updated upstream

What the AI predicts

• Match result (1X2) — home win, draw, away win, with three calibrated probabilities that sum to 100%.
• Totals (over/under goals) — typically 2.5 and 3.5 goal lines.
• Both Teams to Score (BTTS) — calibrated yes/no probability per match.
• Asian handicap — for closer matches where the 1X2 line lacks edge.
• Live in-game — minute-by-minute win probability factoring in goals, red cards, and expected goals (xG) accumulated.

What makes EPL different

Soccer is a low-scoring, draw-heavy sport — which makes it harder to predict at the match level than NBA or NFL but easier to find edge in markets that exist because of the sport's structure (BTTS, total goals, draw lines).

1. Draws are real outcomes. ~25% of EPL matches end level. Models that ignore the draw probability bleed value on every match.
2. Expected goals (xG) is a strong feature. EPL has dense xG data going back years. The model uses team-level xG, opponent-adjusted xG, and player-level xG contributions.
3. Schedule density matters. Champions League midweek + EPL weekend creates fatigue effects the model picks up on.
4. Manager change signal. A new manager hire causes a measurable bounce in the first 4–6 matches that the model captures.

How accuracy looks

30-day rolling, EPL + English-club European competition:

Confidence band	Hit rate	Avg CLV
50–60%	54.2%	+0.6%
60–70%	64.0%	+1.9%
70–80%	73.5%	+3.4%
80%+	82.7%	+5.1%

Common questions

What about other leagues?

Pick1's soccer model also covers La Liga, Serie A, Bundesliga, Ligue 1, MLS, plus the Champions League and Europa League knockout rounds. EPL is the most-covered league because it has the deepest training data and the most market volume.

Does the AI predict draws?

Yes — explicitly. The model outputs a probability for each of the three outcomes (home, draw, away). The draw probability is calibrated separately and is one of the more profitable markets for the model in tight matches.

Live predictions during a match?

Yes. The live model updates per significant event (goal, red card, substitution, penalty awarded). Live in-app shows a probability bar with the model's current read.

What about player props (goalscorer, assists)?

Top scorer and anytime-goalscorer markets are supported for the higher-volume players where shot data is dense. Lower-volume players are skipped — too noisy at the player-prop level.

=======