RRABBY 138
PredictionsPlayersRecord BookRecordsSchedulesSchedule StrengthRabby vs VegasResultsDashboardDailyCompareMy TeamsArchiveData
HomeGamesLiveRankingsStatsTeamsConferencesPredictionsPlayersRecord BookRecordsSchedulesSchedule StrengthRabby vs VegasResultsDashboardDailyCompareMy TeamsArchiveData
Methodology & documentation

How Rabby works

This is the behind-the-scenes home for the ranking system, projection engine, data rules, release process and research program.

What Rabby measures

  • Rabby ranks current team strength, not résumé, poll reputation, brand value or betting-market opinion.
  • All 138 FBS teams receive an ordinal rank and a separate continuous/cardinal strength value.
  • The Power 25 is simply teams ranked 1–25.

Evidence used

  • Opponent-adjusted results, location and scoring margin with blowout/garbage-time damping.
  • Offensive and defensive efficiency, yards per play, turnovers, special teams and situational information where available.
  • Roster construction, returning production, quarterback state, transfers, injuries/availability, coaching continuity and uncertainty.

Evidence states

  • Confirmed Signal
  • New Signal
  • Mixed Evidence
  • Unresolved

A win can be negative evidence and a loss can be positive evidence when the underlying performance warrants it.

Projection Engine v0.3

  • Uses continuous team strength rather than ordinal-rank differences.
  • Includes historically calibrated home field and separate offensive/defensive scoring identities.
  • Produces expected margin, expected total, projected score and win probability for FBS-FBS games.
  • 2025 holdout: 13.33 margin MAE, 16.94 RMSE, 70.7% straight-up accuracy, 12.82 total MAE and 15.91 total RMSE.

FBS/FCS common-DI bridge

  • Crossovers use common-DI cardinal ratings, never ordinal-rank subtraction.
  • The current bridge contains 138 FBS and 128 FCS teams on one calibrated scale.
  • Home field is approximately 2.498 points and crossover residual uncertainty is approximately 12.06 points.
  • Cross-division totals and projected scores are not published unless separately supported.

Market independence

  • Sportsbook lines are comparison data, not a Rabby input.
  • Rabby does not move merely because Vegas disagrees.
  • Market snapshots and Rabby predictions are preserved before games so the comparison can be graded honestly.

Release governance

  • Historical releases are immutable.
  • Every official ranking release must contain exactly 138 teams and ranks 1–138 exactly once.
  • Large methodology-only moves trigger migration review.
  • Public rank trajectories use official weekly or formal midweek releases, not every individual game-state rerate.

Weekly production cycle

  • Lock predictions before kickoff.
  • Ingest results and available statistics.
  • Grade predictions and market comparisons.
  • Update all 138 teams and propagate opponent-network effects.
  • Run migration/integrity checks, freeze the new release and generate the next slate.

Advanced-stat research

  • Play-by-play, EPA/play, success rate, explosiveness, drive efficiency, havoc and related measures are challenger inputs.
  • They are not added merely because they exist; they must improve out-of-sample prediction performance.
  • The test sequence is historical backtest → holdout validation → live shadow test → promote or reject.
  • Missing fields remain unavailable, never zero.

Source & data architecture

  • The project maintains a canonical ranking state, game/stat ledger, immutable historical releases, source/data manifest and work queue.
  • Core game data are cross-checked across authoritative sources when possible.
  • 2026 PBP acquisition routes are mapped through ESPN/SportsDataverse infrastructure; advanced fields remain unpublished until content-level QA is complete.

Public website rules

  • The private research workbook is not the public website data source directly; approved publication exports feed the site.
  • Public spreads display to the nearest 0.5, totals to the nearest whole number, projected scores as whole numbers and win probability as a whole percent.
  • Full-precision values remain in the internal prediction ledger for grading.
  • Future schedules can be shown before future Rabby predictions are published.

Current data limits

  • Core score, opponent, location and broad team-stat coverage is strong through the current release.
  • Universal advanced play-level metrics and complete personnel/availability information are still being expanded.
  • The website does not fill missing advanced fields with estimates simply to make tables look complete.
Documentation principle

Public pages show football. This page shows the machinery.

Rankings, games, stats and team pages are intentionally kept focused on useful football information. Model development, data coverage, calibration, QA, release rules and research status live here.

Live scores, live stats & video

Rabby's live center uses a same-origin Cloudflare proxy to retrieve the current ESPN college-football scoreboard and game-summary feeds. Live scores and box-score fields are supplemental display data; they do not alter a frozen Rabby prediction after kickoff.

Rabby does not host, restream, mirror, or bypass access controls for television broadcasts. Game pages show the scheduled network and may expose an official broadcaster watch link when the upstream event feed provides one. Viewing still depends on the broadcaster, geography, and the user's subscription.

Public discovery metrics

Chaos Meter

The public Chaos Meter is a presentation/discovery score, not a production prediction input. It combines projected game closeness, Rabby-market disagreement and projected scoring environment to help visitors find potentially entertaining games. It never changes Rabby team strength, projected margins or win probabilities.

Game badges

Labels such as SIDE FLIP, RABBY EDGE, UPSET WATCH, COIN FLIP and FCS DANGER are derived from certified prediction and market fields. They are navigation/storytelling labels only.

Live ticker and Game Center

The ticker and live Game Center are presentation layers built on current scoreboard/game-summary data. They never modify frozen Rabby pregame predictions. Before kickoff, the ticker shows schedule and prediction context; during games it shows current score/status; after finals it shows completed scores. Upset Alert means the pregame Rabby favorite is currently trailing, not that Rabby has issued a new live prediction.

Game ticker presentation

The public ticker rotates through three presentation states: current game status, the frozen Rabby pregame forecast, and a sourced or data-derived game note. Injury language is shown only when supported by the verified game-intelligence layer; otherwise the note uses a matchup, series-history, or Rabby-versus-market fact. The ticker does not create a live model or alter the frozen prediction.

Editorial reporting standard

Game-day intelligence uses official school athletics pages, game notes, press conferences and current player interviews first. Availability notes are dated and sourced. Short quotations are used selectively; most reporting is paraphrased. Editorial images are displayed from official athletics sources with attribution and a source link. No unverified injury, personnel or series-history item is published merely to fill a card.