FPL Intelligence Engine

🔥 The "Haul Alert"

When you see a flame icon next to a player, it means the model thinks they are "due" for a double-digit score (9+ points).

Example: Neto vs Szoboszlai

Imagine Neto has blanked for 3 games straight, while Szoboszlai scored a lucky deflection last week.

The Human View: "Szoboszlai is in better form; I'll pick him."
The Model View: "Neto is getting 3 shots per game and taking all corners (High xGI). He's playing against a 'Leaky' defense. He gets the 🔥 icon because his luck is mathematically likely to turn."

Scenario: The Salah "9+ Point" Trigger

Salah misses 2 big chances but has 8 touches in the box.

The Model: Even if he blanks today, the simulation sees a 9+ point haul in 25% of its runs. It triggers the 🔥 icon because the "clumpiness" of his stats suggests an explosion is imminent.

The model often highlights out-of-form players who have "Elite underlying stats" because it bets on the Process rather than the Result.

⚡ Performance-First Philosophy

As of v3.1, the engine has removed all reliance on Effective Ownership (EO) and market sentiment. The model no longer cares about who is "template" or "popular".

The Logic

Process Over Popularity: Ownership is a lagging indicator. By the time a player is highly owned, their value is often priced in. The engine identifies players based on underlying stats (xG, xA, CS Prob) before they become "obvious".
Injury "Wait and See" Penalty: Any player returning from a 2+ gameweek absence receives a 20% reduction in projected minutes. This accounts for physical rustiness, rotation risk, and the need to "prove fitness" before trusting them fully.

Example: The Foden Return

If Phil Foden misses 3 games with a hamstring injury and is declared "fit" for the next GW, most managers rush to bring him in. The model applies the 0.8x multiplier, reducing his expected minutes from 90 to 72. This prevents over-valuing a player who may be eased back in or subbed off early.

🎖️ Captaincy Tiers

1. The "Top Pick"

This is the reliable choice. Best balance of safety (expected points floor) and ceiling (explosivity).

Example: Haaland. Even if he's not predicted for a massive haul, his "Floor" is so high (90+ minutes) that he usually stays here.
The Twist: If Haaland plays Arsenal (A) and Saka plays Burnley (H), the model might swap them. It knows Saka gets extra points for a clean sheet and has a wider "path" to points.

2. The "Differential"

The high-risk, high-reward pick for managers looking to gain ground in their mini-leagues. This pick has low ownership but explosive potential.

Example: Szoboszlai or Enzo. They are "Talismanic" for their teams but have low ownership (< 10%). The model picks them when their xGI Share is massive for an easy upcoming fixture.

3. The "Defensive Pick"

Prioritized for strong Clean Sheet probability and defensive floor. Often a high-scoring defender or goalkeeper.

Example: Gabriel or Tarkowski.
Scenario: Gabriel plays Liverpool (H).
- The Human View: "Liverpool score every game; skip Gabriel."
- The Model View: "Gabriel has a 4.2 DefCon average. Even without a clean sheet, he's likely to get bonus points from clearances and has a scoring threat. His ACS (Actual Clean Sheets) shows he's a reliable top-tier asset."

🛡️ DefCon

"DefCon" is our proprietary metric for defensive reliability. It rewards players who "do the dirty work."

The Tarkowski Rule

Even if Everton concede a goal, Tarkowski might score 5-6 points because he clears the ball 15 times and blocks 3 shots. The model values this "Floor" because it's predictable, whereas goals are random.

🧤 The Goalkeeper Paradox

The model uses a specialized Poisson-derived penalty for Goalkeepers to find the "Sweet Spot" between save volume and clean sheet potential.

Tier 1 (The Wall): GKs with low expected goals against (xGC < 1.0) get a boost. The model trusts them to keep a clean sheet.
Tier 2 (The Save Magnet): GKs in mid-table teams. The model weighs their save potential against the risk of conceding.
The Cap: To prevent "Shooting Gallery" picks, save points are capped relative to clean sheet probability. If a GK is likely to concede 3+ goals, the model will avoid them even if they might make 10 saves.

⚙️ The Engine Room

For the managers who want to know the "Why" behind the "Who." The model processes over 100 data points per player every week.

Core Metrics

xP (Expected Points): The mathematical average of potential outcomes based on historical conversions. It helps you see the "statistical expectation" for a player in a given gameweek.
xGI (Expected Goal Involvement): The sum of Expected Goals (xG) and Expected Assists (xA), adjusted for opposition defense quality. This tells you who is getting into the most dangerous positions.
FDR (Fixture Difficulty Rating): A numerical scale (1-5) representing the difficulty of the upcoming fixture. Lower numbers mean easier opponents.
xGI / AGI: (Attackers Only) "Expected Goal Involvement" vs "Actual Goal Involvement". This compares the statistical quality of chances with real-world output.
CSP / ACS: (Defenders & GKs Only) "Clean Sheet Probability" vs "Actual Clean Sheets". This shows the likelihood of a shutout vs historical defensive results.
Form / TSB: Combined view of recent trend and overall ownership.
DefCon: A consistency score for defenders. It tracks blocks, clearances, and interceptions—the "hidden" points that guarantee a high floor even without a clean sheet.

The Intelligence Logic

We use an XGBoost (Extreme Gradient Boosting) model. It doesn't just look at who is playing well; it look at 5 years of historical patterns to identify "Regression to the Mean"—predicting when a hot streak will end or a dry spell will break.

Composites & Weights

The output you see is a Composite Score. We weigh Recent Form (70%), Fixture Difficulty (50%), and ICT Index (30%). This prevents the model from being "blinded" by a single massive haul from a mediocre player.

📡 "Floor Mode"

In some gameweeks, all the "Big Six" teams play each other. This is a low-ceiling week.

The Signal: The model suggests Pickford or a safe defender as a high-value pick and stops chasing "Haul Alerts."
What it's doing: It has entered Floor Mode. It's telling you: "This week is a minefield. Don't chase hauls that aren't there. Pick the players who are guaranteed to play 90 minutes."