Course-fit model
Who this week's course sets up for: a blend of course-neutral skill, course-specific fit, and current form — every number a percentile within this week's field (higher is better).
True Skill blends each player's strokes-gained categories across recent seasons; Course Fit weights proximity bands, driving precision, short game, and venue history to match this course; Form is a difficulty-adjusted strokes-gained measure — each recent round scored against how hard it actually played, weighted toward the most recent starts and shrunk toward the field for small samples. Method details on the methodology page.
How the course-fit model works
The model asks one question: which players in this week's field suit this week's course. It combines three inputs into a single 0-100 score. The first is baseline skill — how good the player is, measured from their strokes-gained record. The second is fit — how well the specific things this venue rewards line up with the specific things the player does well, using the course profile shown on every venue page. The third is recent form. Each component is shown beside the total, so a high score is always decomposable: you can see whether a player ranks highly because they are excellent everywhere, because the course suits them unusually well, or because they are hot right now. This is the one place on OTIS where published numbers are combined into a score of our own, and it is labelled as a model wherever it appears.
What it does and doesn't predict
It does not predict a winner. Golf tournaments are won from a field of well over a hundred professionals separated by very little, and any honest model of them produces a distribution rather than an answer. A fit score says that a player's game matches what this course has historically rewarded — nothing more. It has no knowledge of injuries, travel, weather on the day, or how someone slept. Players with little or no measured tour data are flagged rather than quietly averaged into the middle, because an unmeasured player is an unknown, not an average one. If you want explicit probabilities with a published accuracy record instead of a ranking, the probabilities page states them and grades itself afterwards.
How to read a fit score
Read the score as a ranking device, not a rating with physical meaning: it is useful for comparing players within this week's field and meaningless across weeks. Small gaps are noise. A two-point difference between adjacent players should not change anyone's mind; a twenty-point gap is worth a look, and the component columns will usually tell you where it came from. Because the fit component depends on the venue profile, a course the tour has played many times produces a better-supported score than a first-time host with little history to profile. The venue page for each course shows what that course rewards and how much data sits behind it, which is the honest way to judge how much weight this week's numbers deserve.
Common questions
What is a golf course-fit model?
- It is a way of ranking a tournament field by how well each player's strengths match what a particular course rewards. OTIS combines baseline skill, course fit and recent form into a 0-100 score, and shows each component separately so the total can be taken apart.
Does the course-fit model predict who will win?
- No. It ranks how well players suit the venue, which is a different and much narrower claim than predicting a winner. It knows nothing about injuries, weather on the day or travel. For explicit probabilities with a public accuracy record, see the probabilities page.
How is course fit measured?
- From the profile of what each venue has historically rewarded — its scoring characteristics and demands — matched against each player's own strokes-gained record. Venues with a long hosting history produce better-supported profiles than first-time hosts.
What happens to players with little tour data?
- They are flagged rather than hidden or averaged into the middle of the field. A player without measured tour rounds is an unknown quantity, and labelling them as such is more honest than assigning them a confident-looking score.
Sourcing is set out on the methodology page; every stat name is defined in the glossary.