OSR V2 rating methodology

1. Eligible statewide network

Completed, rating-eligible varsity games form one season-specific graph. Oregon varsity programs are published nodes. Legitimate out-of-state opponents may remain as boundary nodes, but are never published in Oregon rankings. Boundary nodes with fewer than two distinct opponents are repeatedly removed until stable. Disconnected Oregon teams are omitted rather than assigned an invented rating. Rating margin is capped at seven without changing the canonical score.

2. Recency and venue

Here k is the number of completed-game dates before the snapshot. Pairwise evidence uses symmetric geometric-mean recency. Home advantage is estimated jointly by weighted least squares; neutral games receive no offset, and team parameters are centered to zero.

3. Independent anchor reliability

Each team-to-opponent direction gets its own reliability. Independent and cross-division opponents supply stronger evidence; repeated meetings add only limited information. Directional game weights combine recency and reliability and are normalized per team.

4. Recursive network component

All opponent-dependent values are solved simultaneously as a zero-mean weighted least-squares system using a stable singular-value decomposition. No hidden ridge term or starting Elo is introduced.

5. Pods and partial pooling

Pods are same-classification connected components. Strongly bridged pods retain their earned baseline; weakly connected pods move toward neutral. The ordering and separation of teams within every pod remain intact.

6. Nonlinear result component

Capped differential d is venue-adjusted and fitted by weighted least squares. Common-opponent comparisons add 0.10 direct-game-equivalent strength using geometric-mean recency. Repeated evidence for one team pair is recency-averaged and capped at 0.50 total weight.

7. Opponent-adjusted scoring

An additive model estimates the scoring intercept, home effect, team attack, and team defense. Only winning goals beyond a seven-goal margin are removed for this component. Attack and defense are centered, and higher defense is oriented as better.

8. Final OSR V2 rating

Each component is population-standardized across every analytical node. A rating of 100 is the network average and 15 points is one standard deviation. State and division rank use the same score; division rank limits the ordering to one classification.

Confidence and historical seasons

Confidence is a separate 0–100 evidence-quality indicator using average directional anchor quality, independent-opponent information, and bridge saturation E_p/(E_p+1). It never changes or shrinks Power. Historical ratings are calculated independently by season. Seasons without a valid statewide network are omitted, and reconstructed historical ratings are labeled as reconstructed.

Variable glossary

i, j

The rated team and one of its opponents.

g

A completed, eligible varsity game.

p

A same-division connected-component pod.

k

Completed-game-date age of a game.

d

Venue-adjusted goal differential after the seven-goal cap.

r_g

Recency weight of game g.

n_other

Distinct independent opponents in the directional anchor.

n_cross

Distinct cross-division opponents.

c_opponent

Limited repeat-meeting evidence for an opponent.

E, A

Directional evidence and its bounded reliability transform.

w_ij

Normalized directional game/network weight from i to j.

N_i

Solved recursive network value for team i.

E_p

Crossing-game bridge evidence attached to pod p.

B_p

Pod baseline before or after V2 partial pooling.

q(d)

Nonlinear result transform for capped differential d.

Attack, Defense

Centered opponent-adjusted scoring parameters.

Z(X)

Population z-score: (X − population mean)/population standard deviation.

C

Final weighted standardized component before the 100/15 display scale.