# FantasyOmatic — Methodology and Published Evidence

Fantasy football ratings calculated from football data—not generated by a chatbot.

Our offensive ratings use statistical machine-learning models. AI Coach and Podcast use generative AI to explain published results and help you apply them; they do not calculate the underlying ratings.

Week 2, 2026 incorporates 1 completed current-season weeks. Training seasons: 2026. Defensive evidence is unavailable.

Model published: 2026-09-15T23:35:21.751817+00:00
Availability observed: 2026-09-16T10:00:13.153869+00:00
Publication updated: 2026-09-16T10:00:43.593113+00:00

Offensive Overall, This Week, ROS and Playoffs each use a separate frozen conversion of their native model output. Their difference does not isolate an opponent effect. Kicker and D/ST models include schedule and market inputs and use within-position scales. Expert consensus is a comparison benchmark where verified.

The normal model run is scheduled Tuesday at 2 a.m. Pacific after Monday's games. Publication waits for verified source readiness and retries when required. Availability observations follow official reporting and inactive-list deadlines. Missing coverage does not establish healthy or active status.

## Published data
- [Original model receipt](https://fantasyomatic.com/evidence/ratings/20260915T232826Z)
- [Exact revision receipt](https://fantasyomatic.com/evidence/ratings/20260915T232826Z/revisions/356249)

- [Rankings](https://fantasyomatic.com/data/rankings): Complete published player ratings, provenance and current availability.
- [Matchups](https://fantasyomatic.com/data/matchups): Schedule and supported native opponent effects. Unavailable evidence is explicitly labeled.
- [Methodology](https://fantasyomatic.com/why): Statistical approach, validation and model limitations.
- [Newsstand](https://fantasyomatic.com/news): Published analysis; verify each article's date and supporting release.

## Does ChatGPT generate FantasyOmatic ratings?
Our offensive ratings use statistical machine-learning models. AI Coach and Podcast use generative AI to explain published results and help you apply them; they do not calculate the underlying ratings.

## How are the offensive ratings calculated?
Ridge regression fits statistical relationships in verified football performance data. Versioned calibration converts each native offensive output to its public rating. Offensive Overall, This Week, ROS and Playoffs each use a separate frozen conversion of their native model output. Their difference does not isolate an opponent effect. Kicker and D/ST models include schedule and market inputs and use within-position scales. Expert consensus is a comparison benchmark where verified.

## Are the ratings copied from expert consensus?
The offensive model calculates its own outputs. Dated expert consensus can be used as a comparison benchmark. Kicker and D/ST models separately use schedule and market inputs; they are not described as market-independent.

## What does historical bootstrap mean?
Before qualifying current-season evidence exists, the published release can use verified historical player-model outputs and current roster, schedule and availability inputs. A provisional floor of 40 is a rule for insufficient evidence, not earned performance. The release provenance identifies the actual source seasons.

## How is model quality evaluated?
Historical walk-forward tests train on eligible earlier evidence and evaluate later outcomes. Validation reports identify the model version, seasons, sample size and error metric; historical results do not guarantee future accuracy.

FOM_PUBLICATION {"version":1,"season":2026,"week":2,"runTs":"20260915T232826Z","revision":"356249","ratingBasis":"current_season","defenseEvidence":"unavailable","modelPublishedAt":"2026-09-15T23:35:21.751817+00:00","availabilityObservedAt":"2026-09-16T10:00:13.153869+00:00","revisionPublishedAt":"2026-09-16T10:00:43.593113+00:00"}