Sample model · ASCEND-QX 4.2

How We Actually Do It

No black box theatre. Here is the exact pipeline every Ascend selection passes through, the weights behind it, and a live worked example from today's board.

The Pipeline

Ingest

Team and player logs, pace, rotation reports, injury designations, travel and rest, weather, and multi-book line movement are normalised into one feature frame per market.

Standardise

Every feature is converted to a z-score against its own league-season distribution, so a rebound rate and a line move are comparable on the same scale.

Weight

The eight feature families are combined with fitted weights that sum to 1.0, producing a single composite signal in standard-deviation units.

Calibrate

A logistic link maps the composite signal to a win probability. The slope is fit against historical closing lines so probabilities are honest, not inflated.

Compare

The posted American odds are de-vigged into an implied probability. Model probability minus implied probability is the edge — no edge, no play.

Stake & tier

Quarter-Kelly sets a unit size capped at 3u. Signals above the strong threshold become the 5 strong plays; the next band becomes the 3 value plays; individual stat lines become the 3 props.

Feature Weights

Rolling Form (L10)18%

Exponentially-weighted performance over the last ten games.

Matchup Rating16%

Opponent positional defense / pace-adjusted allowance.

Usage & Role15%

Snap share, touches, minutes projection and role stability.

Pace & Game Script12%

Projected possessions and expected game flow.

Injury / Rotation Impact12%

Teammate absences that shift volume toward the subject.

Line Movement11%

Open-to-current move and steam detection across books.

Situational Factors9%

Rest, travel, altitude, back-to-backs, weather.

Market Efficiency Gap7%

Consensus vs sharp book divergence.

Worked Example

A real selection from today's slate with every intermediate value exposed.

#1 · Strong Play · NCAAB / NCAAF

Ball State — Total Points

Buffalo @ Ball State · College · 00:30 UTC

Under 51.5

+100 · 0.25u · BetRivers

Sportsbooks · 10 books

Model Proj.

134.3

Win Prob.

51.9%

Edge vs Market

+2.1%

Confidence

54/100

Line Movement and Rolling Form (L10) are the dominant drivers (Line z=-1.86, Rolling z=+1.06, Injury z=-1.12). Composite signal -1.2σ against a market implying 60.3%. Priced at +100 (best of 10 books, BetRivers) against a de-vigged consensus of 49.8%.

Why this prediction exists

Model probability: 51.9%
Price implied: 49.8%
Market consensus: 49.8%
Edge vs stricter reference: +2.1 pts
Edge needed to publish: +5.2 pts
Data quality: 92/100

books disagree sharply on the price (wide dispersion)

Model explanation

Speculative · 54/100

Model win prob.: 51.9%
Market implied: 49.8%
Edge: +2.1%
Kelly stake: 0.25u

Key statistical drivers

Line Movementz -1.86
Rolling Form (L10)z +1.06
Injury / Rotation Impactz -1.12

Why this is a Strong play, not a Value play

Strong plays require both a calibrated win probability at or above the model's conviction floor and agreement across the top-weighted signals. Here the model lands at 51.9% against an implied 49.8%, giving a meaningful mispricing versus the closing-line forecast. Line Movement, Rolling Form (L10), Injury / Rotation Impact all push the same direction, so there is no internal disagreement to discount — which is why the Kelly allocator sizes it at 0.25u rather than a reduced stake.

Why Under 51.5

The projection sits 82.8 below the posted line of 51.5. The model takes the Under because the projected distribution clears the number 51.9% of the time, while the price only demands 49.8%. A gap this size is large enough to survive a normal usage or pace swing.

Probabilities are isotonic-calibrated outputs of ASCEND-QX; stakes use quarter-Kelly. Model output is analytical information, not a guarantee of outcome.

signal   = Σ (weight_i × z_i)            // composite, in σ
p_model  = 1 / (1 + e^(-1.65 × signal × 2.6))
p_market = devig(american_odds)
edge     = p_model - p_market
stake    = clamp(0.25 × kelly(p_model, odds) × 4, 0.5u, 3u)

→ p_model  = 51.9%
→ p_market = 49.8%
→ edge     = +2.1%
→ stake    = 0.25u  (confidence 54/100)

Coverage

The system knows every major sport and its prop markets.

Football

NFL

Game markets: Spread, Total Points, Moneyline

Props: Passing Yards, Rushing Yards, Receptions

Basketball

NBA

Game markets: Spread, Total Points, First Half Spread

Props: Points, Rebounds, Assists, Pts+Reb+Ast, Three Pointers Made

Baseball

MLB

Game markets: Run Line, Total Runs

Props: Strikeouts, Total Bases, Hits + Runs + RBIs

Hockey

NHL

Game markets: Puck Line, Total Goals

Props: Shots On Goal, Points, Goalie Saves

Soccer

EPL / UCL

Game markets: Asian Handicap, Total Goals

Props: Shots On Target, Player Shots

College

NCAAB / NCAAF

Game markets: Spread, Total Points

Props: Points, Rebounds

Combat / Tennis

UFC / ATP

Game markets: Moneyline, Total Rounds / Games

Props: Significant Strikes, Total Aces