The mean of that distribution is the golfer's ability read through the course, and the spread is his scoring variance. The engine runs the field forward round by round, applies the cut, preserves the missed-cut zero and the fat tails that make golf high-variance, and reduces the many tournaments to win, top-ten, and head-to-head distributions checked against the market. Every output is a bet, and none of it writes back to a rating.
The unit of the simulation is the round score, modeled as a near-normal draw. The mean is the golfer's KR read through the course, his ability priced as an expected score against the field. The spread is his outcome-variance dial, his scoring standard deviation around a tour average of roughly 2.75 strokes per round, combined with the course's own variance. This is the two-dial doctrine made mechanical. Move the course variance and watch the spread, not the mean.
The KR sets the mean and it does not move as you drag the course variance; the variance dial and the course set the spread, and they do. A long, penal, high-variance course widens every golfer's round distribution and raises the upset rate, and a forgiving one narrows it. The engine is reading a fixed number through a changing test, never re-rating the golfer. The KR sets the mean, the variance dial sets the spread, and only the spread moves with the course.
Illustrative engine read on the real per-round scoring model (near-normal round score, mean from the KR through the course, spread from the outcome-variance dial plus course variance). Composite golfer, demonstration figures, scoring parameters v0.
The tournament is the round distribution played out over the field. The engine runs the field forward round by round, ranks it, and applies the cut, and the cut is load-bearing, because a missed cut is a zero. Make-cut is a first-class output and the finish distribution is bimodal: a mass at missed-cut and a spread across the weekend. Carry the within-week round correlation and the bounded closing pressure and watch the tails fatten.
A golfer trending low or high tends to continue, so within-week correlation widens the tails and makes hot and cold weeks more common than four independent rounds would. Closing pressure shifts the read near the lead, bounded and confidence-gated, and weather enters as a bounded environment factor, never a rating change. The missed-cut zero and the fat tails are preserved because that is how golf actually resolves. Preserve the missed-cut zero and the fat tails, because that is what makes golf high-variance.
Illustrative engine read on the real tournament model (round-by-round play, the bimodal cut with make-cut first-class, within-week correlation, bounded closing pressure and weather). Composite field, demonstration figures, cut parameters v0.
The many simulated tournaments reduce to the outcome distribution and its derived probabilities: win, top-five, top-ten, make-cut, and head-to-head, each with its own variance. Every output is a distribution and a bet, never a single predicted line. Where a live market exists, the engine's odds sit beside it as a check, and match play resolves as a hole-by-hole variant that is higher-variance by construction.
A cold reader can watch the engine reason against the market on the same field and see where they agree and where they part. Match play compresses every edge toward the coin, because a single hole can turn it and a blow-up costs one hole rather than many strokes, so a favorite who is a strong stroke-play bet is a much weaker match-play one. Every number is a distribution, and none of it writes back to a rating. Every output is a distribution checked against the market, and the arrow never runs back to the rating.
Illustrative engine read on the real output structure (win, top-five, top-ten, make-cut, and head-to-head as distributions, the market check, the higher-variance match-play variant). Composite field, illustrative market odds, all figures v0.
The tournament is the two dials played out over the field. The engine simulates golf the way golf resolves, round score by round score, preserves the missed-cut zero and the fat tails that make the sport high-variance, prices every outcome as a distribution checked against the market, and never writes back to a rating. The read flows into the outcomes; the outcomes never flow back into the read.
The sim runs on the finalized reads and the field and venue around them, and it feeds the week's strategy and the value model. It is where the ratings become odds, and the odds never become ratings.