Two different questions
Performance and projected strength
Efficiency Index describes this season’s performance. It uses the same published formula as the Ravens dashboard: opponent- and venue-adjusted offense (45%), defense (45%) and special teams (10%), with game-state weighting and a frozen prior-season calibration. Higher is better on a 0–100 scale, with 50 representing average unit strength; the score is neither a percentile nor a win probability. Read the published Efficiency methodology.
Team Strength Index (TSI) estimates strength in scoring-margin points relative to an average team on a neutral field. A +5 team projects five points better than average. Subtract the opponent’s TSI, then add the fitted home advantage when applicable, to obtain a model margin. The points are neither a win probability nor a betting guarantee.
What goes into TSI?
- Passing and designed-rushing EPA, adjusted jointly for opponents and venue.
- Passing and rushing success rate: the frequency of plays that improve expected points.
- Special-teams value above context expectations.
- Historical team performance, with more carryover for offense than defense, and an eight-week recency half-life.
- Quarterback EPA and designed-rushing value, with estimates pooled toward historical role/draft cohorts when samples are small.
- Fumble-recovery luck stabilization: fumbling remains costly in expectation, while fortunate recoveries receive less credit.
The relative contribution of each feature is learned from historical scoring margins. Subjective rankings, DVOA and betting lines are outside the formula. Margin coefficients were fitted using seasons through 2023.
Historical weighting and team changes
Team offense starts with a stronger historical prior than defense. Historical carry is 80% for offense and 40% for defense before current plays compete with those priors. The beta reduces direct prior influence by 10% relative and transfers that weight toward current-season evidence. New coaches reduce both historical carries by 35%; a new primary quarterback reduces offensive carry by 40%. As games accumulate, current performance has greater influence.
Reducing a negative historical rating can move a team upward toward average. The transition rules are applied across the league; they do not supply a team-specific bonus. Early-season ratings remain sensitive to these assumptions.
Reading the contributions
Offense + defense + special teams + lineup adjustment equals TSI points. Positive defensive contributions indicate a stronger defense. Quarterback performance is already part of offense; the lineup adjustment replaces the embedded quarterback contribution with the expected starter’s estimate, avoiding double counting. It is not a standalone quarterback ranking.
Closely rated teams are effectively grouped. A small change can move a team several places without materially changing its expected strength. One-decimal display rounding can make neighboring teams look tied; sorting uses the full underlying values.
Quarterback choices
Default rankings assume the listed depth-chart starter. Each projection shows its assumed quarterbacks and forecast date. Depth order does not confirm availability. The weekly backup toggle changes upcoming projected spreads. The ranking checkbox also applies that choice to the displayed team rankings when you want to explore a longer absence. Scenarios are saved within your browser session and can be reset.
Current-week injury designations appear when available. A missing designation does not confirm that a player will start. QB estimates are particularly uncertain for players with limited NFL samples.
What the results page proves
All displayed results are for 2026. Saved forecasts are timestamped before kickoff; the first entries were saved during private research. Earlier games are retrospective replays using prior-week performance, with the model selected later. Each game identifies its forecast type. If a different QB started, the main spread is adjusted using only the QB statistics available before that week; the original forecast remains visible. Completed-game QB identities are verified using the first recorded known-quarterback dropback in play-by-play, with the schedule as a fallback. This avoids substituting a later relief QB or a trick-play passer. This lineup correction is retrospective, not a saved prediction of the QB change. Closing lines may contain later injury information, so that comparison has an information-timing limitation.
Game error measures how far the displayed scoring margin was from the actual margin, including any labeled actual-starter adjustment. Future saved results will provide a stronger evaluation than repeatedly revisiting historical games.
Data and update policy
Scheduled rankings advance after every game in a week has finished and the corresponding play-by-play is available. Manual interim updates may include a partial week; they are explicitly labeled and exclude pending games. Current-week scores and market snapshots can refresh independently. Statistical corrections may revise descriptive ratings; already saved forecasts retain their original values. We refresh after the Wednesday correction window before the next Thursday game.
Public, keyless data comes from nflverse. Their update schedule explains availability and statistical corrections. International games receive no home-field advantage. Weather, rest, non-QB injuries and explicit player-matchup effects are outside this beta.