Start with the player, then add the game
A useful projection begins with what the player has demonstrated over a meaningful period. Season production supplies stability, while recent games can identify role changes, injuries, lineup movement and changes in opportunity. Neither period should automatically replace the other.
Bosh Roid combines longer-term production with more recent information and then adds game-specific context. The exact weights and internal transformation rules are proprietary, but the major information categories and the evaluation process are published.
Season baseline
A larger sample provides a stable estimate of the player’s established production and role.
Recent form
Recent games help identify changes in playing time, batting order, minutes, usage and health.
Role and opportunity
Confirmed lineups, expected starters, rotation position and available teammates shape the opportunity available to the player.
Opponent and environment
Handedness, probable pitchers, opponent profiles, venue, roof, temperature and wind refine the game-specific expectation.
Store the forecast before the result
Model evaluation is only trustworthy when the forecast is preserved before the outcome is known. Bosh Roid stores forecast snapshots and uses the newest eligible pregame snapshot for each date, game, player and metric. Repeated runs are deduplicated so the same underlying forecast opportunity is not counted multiple times.
After the game, the saved projection is matched with the final player statistic. The platform calculates absolute error, squared error, signed error and directional results where the required comparison data is available.
Why projections should remain adjustable
Playing-time assumptions are often the most important input in fantasy sports. A projection can change meaningfully when a player moves into the starting lineup, loses minutes, changes batting-order position or inherits a larger role after a trade or injury.
Bosh Roid’s scenario tools allow users to adjust those assumptions and observe how the rest of a roster changes. The goal is not to pretend that every input is known with certainty. It is to make the assumptions visible and allow the user to test alternatives.