pga methodology

Golf Methodology (PGA + LPGA)

PGA layers 14+ custom metrics on strokes-gained, joined with weather and pin difficulty. LPGA runs 10k 4-round Monte Carlo tournament simulations.

PGA uses 14+ custom metrics on top of strokes-gained, joined with weather and pin difficulty at the round level. LPGA uses L04 forward (player-mean) + L06 tournament Monte Carlo (10k sims, RMSE 2.85 strokes per round). Both face-validity tested — if a 'clutch putter' metric ranked someone nobody would call clutch, it got cut.

engine

PGA: strokes-gained + 14+ custom metrics. LPGA: L04 player-mean + L06 Monte Carlo tournament sim

PGA projections blend strokes-gained data with 14+ custom Resolve metrics (course fit, signed-by-role, etc.), joined with round-level weather and pin difficulty. LPGA uses an L04 forward model fitting per-player round-mean from 5+ years of LPGA round data, then L06 Monte Carlo simulates 10,000 4-round tournaments (cut is top 65 + ties after round 2).

data

Where the inputs come from

sources

PGA Tour shot-level (strokes-gained inputs), LPGA round-level scores, weather + pin difficulty per-round

training

PGA: rolling multi-season. LPGA: 5+ years of round-level scores

holdout

Each custom metric face-validity tested before shipping. LPGA L06 uses L04 test-set RMSE (2.85 strokes/round) as residual

calibration

Out-of-sample performance

metric

PGA per-tournament Brier on top-10 finish. LPGA L04 RMSE 2.85 strokes/round; L06 cut-line accuracy

value

L04 test-set RMSE: 2.85 strokes per round

PGA custom metrics each pass a face-validity gate before joining the model — if a 'clutch putter' metric ranked someone nobody would call clutch, it got cut. Equity-first ship means LPGA is the same fidelity tier as PGA.

key levers

What controls the projection

Face-validity gates

Every custom metric has to rank players in a way that passes a face-validity check before it ships. Several metrics that looked promising in fit got cut at this gate.

Weather + pin joins

Round-level weather and pin-difficulty data joined to per-round projections. Helps separate course-fit signal from raw player skill.

Course fit (PGA)

Signed-by-role + course-fit metrics give per-player tournament adjustments based on course characteristics that play to their strengths.

L04 per-player mean (LPGA)

Fit on 5+ years of LPGA round data. The per-player mean is the simulation input for L06.

L06 tournament Monte Carlo (LPGA)

10,000 simulated 4-round tournaments per event. Each sim draws round scores around the L04 player-mean with 2.85-stroke residual. Cut is top 65 + ties after round 2.

faq

Common questions

Why two different engines for PGA and LPGA?

PGA has rich shot-level data (strokes-gained); LPGA does not at the same fidelity. We use round-level for LPGA because that's what the data supports. Both engines are face-validity tested and equity-first — neither is a junior partner.

What's strokes-gained?

A per-shot metric that compares a player's outcome to the field average given the same starting position. Industry-standard for separating player skill from course difficulty. We build 14+ metrics on top of it.

What's the L04 forward model?

LPGA forward projection — fits a per-player round-mean from 5+ years of LPGA round-level scoring data. The mean is the simulation input for L06.

How does the LPGA L06 simulation work?

10,000 four-round tournaments. Each round, each player's score is drawn around their L04 mean with a 2.85-stroke residual (RMSE from L04 test set). Cut is top 65 + ties after round 2. Final standings come from sim distribution.

Do you do golf live during tournaments?

PGA Live (/pga/live) shows in-tournament leaderboard + projections updating during play. LPGA tournament simulation re-runs as rounds complete.

apex framework For the platform-wide methodology framework — pre-registration policy, data philosophy, bias controls, and honesty notes — see the apex methodology page.