About these numbers

Every result here comes from publicly available NJSIAA High School wrestling data, which publishes New Jersey high school wrestling meets. The site reads those results and, every night, recomputes a rating for each wrestler and a percentage for each matchup. Nothing here is entered by hand or by a coach — it is all worked out from who beat whom.

There are only four steps, and none of them is complicated:

1. Rate each wrestler. A wrestler who beats strong opponents is rated higher than one with the same record against weak ones. Beating someone good counts for more than beating someone who loses a lot — so a 20–2 record in a tough conference and a 20–2 in an easy one do not come out equal.
2. Compare the two. A matchup is the gap between the two ratings. A big gap is a lopsided bout; a small gap is a close one.
3. Turn the gap into a percentage. The bigger the gap, the higher the favorite’s chance. An even gap is 50–50.
4. Pull thin cases toward the middle. When one wrestler has very few bouts on record, the number is deliberately moved toward 50–50, because the model genuinely knows less (see the data labels).

When the site says a wrestler has a 74% chance, it means: in matchups that look like this one, the favorite wins about three times out of four — and loses the fourth. 74% is not a promise. It is a statement that the underdog wins here about a quarter of the time.

The number is meant to be honest, not just confident. Every season the site checks itself: of all the bouts it called around 70%, did close to 70% of them actually go that way? On matchups where it has real data the answer is yes, within a point or two. That check is the whole point — a number you can trust is one that is right as often as it claims to be.

Confidence — how sure the model is about who wins. It follows the favorite’s chance:

Data — how much the model has to go on. It looks at the wrestler in the matchup with the fewer bouts, because a prediction is only as good as its thinner side:

• Solid — both wrestlers have a full season or more of bouts. No badge is shown; this is the ordinary case.
• Limited data — one wrestler has a modest record. The number is pulled part-way toward 50–50.
• Thin data — one wrestler has very few bouts on record. The number is pulled halfway to 50–50, and the bout is flagged, because it is a weak estimate and should be read as one.

The model only sees published results. Several of the things that decide a real bout are not in there at all, so they are not in the number:

• Injuries and illness. The single biggest reason a prediction is wrong. If a wrestler is hurt, the model does not know until it shows up in results.
• Who actually made weight. Weight certifications are not published, so a wrestler’s real weight range is only estimated.
• This season’s improvement. A wrestler who has come on sharply since December is still partly rated on December.
• The practice room. Who is moving up, who is being rested, what a coach has planned. This is the real advantage a coach has over this tool.

The site is good at picking the winner and much weaker at the margin. In New Jersey high school wrestling, closely matched wrestlers pin each other about as often as mismatched ones do, so the bonus-point and pin percentages are the least certain thing on a card. Read them as a rough guide, not a forecast.