BOA Methodology
How Burnt Orange Analytics builds, validates, interprets, and publishes predictions, rankings, player grades, and matchup analytics.
Data inputs
BOA uses structured college-football data, historical results, recruiting information, roster-talent indicators, schedule context, player production, and BOA-derived ratings. The primary structured data source is CollegeFootballData.com.
Prediction architecture
Preseason and farther-out forecasts use preseason-safe information. The immediate in-season matchup can use completed-game information through the forecast's stated as-of week. This separation prevents future games from being scored with information that was not yet available.
Margin-first modeling
BOA models expected scoring margin and game total, then converts projected margin to win probability using centered monotonic calibration. Winner, probability, projected score, margin, and spread are checked for directional agreement before publication.
Verified publishing safeguards
BOA preserves prior forecasts and checks each update before it becomes public. If required data or validation checks fail, BOA keeps the last verified forecast rather than publish an unverified replacement.
Reading a BOA forecast
What the game numbers mean
Game pages emphasize the football forecast itself. Technical model lineage is retained internally for reproducibility, while the public view focuses on the numbers describing the matchup.
Win Probability
The model-estimated chance of each team winning the game. The two probabilities are complementary and sum to approximately 100%.
Projected Score
BOA's expected scoring output for each team. Displayed scores preserve the model's margin and total while preferring football-realistic combinations built mostly from touchdowns with extra points and field goals. Projected scores are estimates, not exact-score guarantees.
Projected Margin
The expected scoring difference between the teams. It expresses how many points BOA projects the favored team to win by on average.
BOA Spread
The projected margin expressed in conventional spread notation. For example, Texas -6.5 means BOA projects Texas as a 6.5-point favorite.
Projected Total
The sum of the two projected scores. It represents the model's expected combined scoring environment for the matchup.
BOA Pick Confidence
The model-estimated win probability of the team BOA selected as the winner. It answers: “How likely does BOA think its pick is to win?”
How confidence is calculated
BOA Pick Confidence does not use a second hidden model. It is the published win probability of BOA's selected winner. If Texas is projected at 63.3% to win, BOA Pick Confidence is 63.3%. If the opponent is projected at 72.0%, confidence in the opponent pick is 72.0%.
BOA Pick Confidence is still a probabilistic estimate, not a guarantee. A 70% confidence pick can lose; calibration is evaluated across many games rather than judged from a single result.
How BOA rankings are displayed
BOA team rankings are ordered from the active production Team Rating. The rating is the underlying model value; the displayed rank is the team's position when active teams are sorted from highest to lowest rating. It is a model ranking, not a poll vote.
Preseason ratings can use recruiting, roster talent, historical program strength, and other preseason-safe information. In-season ratings can evolve as approved current-season information becomes available.
Grades and player context
A published BOA Player Grade is the approved overall grading output for that player and season. Once a season's grades are finalized, the published overall grade is not silently recalculated by the website.
Player profiles can also show position-specific production metrics. Those metrics provide context and should not be interpreted as hidden grade sub-scores unless BOA explicitly labels them as grading components.
Different positions require different statistical context
BOA does not force every player into the same generic metric boxes. When the source statistics are available, player profiles use measures appropriate to the player's position group.
If a required statistic is unavailable, BOA prefers a blank or unavailable state rather than displaying an invented zero.
How BOA builds matchup-specific player expectations
BOA player stat projections begin with verified 2025 production and the active 2026 roster. The model then adjusts each eligible player's per-game baseline using the matchup's projected Texas scoring, expected game script, opponent offensive and defensive strength, and the player's published BOA grade when one is available.
Outputs are position-specific: quarterbacks receive passing and rushing expectations, skill players receive usage and yardage expectations, defenders receive tackles and disruption metrics, and specialists receive kicking or punting metrics. Transfer production is stored as a separate prior-school baseline and never overwrites canonical Texas statistics.
BOA requires at least four verified games and meaningful prior usage before publishing a projection. New or low-usage players remain unavailable until a defensible baseline exists. These numbers are model expectations for informational use, not betting props or guarantees.
Pregame and postgame information serve different purposes
Before kickoff
BOA can present matchup information using team ratings, win probabilities, projected score, projected margin, spread, projected total, BOA Pick Confidence, current player grades, and available season production. Pregame sections should be useful before a game is played.
After the game
Final game statistics, player game logs, and result-based analytics populate only after verified game data is available. BOA does not fabricate postgame statistics for an unplayed matchup.
Advanced fields such as expected points, success rate, efficiency splits, or player-matchup projections are shown only when BOA has an approved data source and calculation for that metric. A placeholder is not treated as a published analytic.
How forecasts change during the season
BOA separates near-term in-season forecasting from farther-out forecasting. The immediate next game can use approved information from completed games, while later games continue to use the appropriate long-horizon information state until they move into the near-term window.
New publications do not erase the prior pregame forecast. BOA preserves forecast versions so historical predictions can later be compared with actual results without rewriting history.
Locked historical test
Current long-horizon production model
The current rich-preseason long-horizon architecture was selected without using the locked 2023–2025 test period for training or tuning. On 2,264 locked games, it improved on the BOA preseason-Elo baseline across the core winner, margin, and probability metrics used for promotion.
Historical validation measures behavior over many games and does not guarantee any individual outcome. BOA also publishes projected totals, but the long-horizon total model did not outperform every simple locked-test total baseline and should not be interpreted as a uniquely validated betting edge.
External benchmarks are diagnostic only
BOA may compare outputs with public prediction systems or market lines as a reasonableness check. External probabilities and spreads are not hard-coded targets or direct inputs used to force BOA toward a particular answer.
Data sources
CollegeFootballData.com (CFBD) — primary structured college-football data source used by BOA.
College Football Reference (Sports Reference) — supplemental historical/reference source where applicable.
BOA may also use other publicly available sources for verification and context. Source use does not imply sponsorship, partnership, or endorsement.