TuttleMath / No mystery score
Every number on the board, in plain English.
Early in a season the samples are tiny, so more players show large variance. That is the data being honest, not the model breaking. As more games are added, role stability and repeated usage provide a stronger base for comparison.
No single column should make the decision alone. Projection estimates the center, WCS and BCS frame the range, value measures cap efficiency, and plausibility asks whether the underlying role supports the result. The clearest decisions usually appear when several signals agree.
Every projection carries a confidence weight based on how many games sit behind it. Two games of data produce a wide, unreliable estimate; eight games of consistent usage produce something you can lean on. The stepper lets you change that window, and everything downstream — variance, plausibility and the blended sort — recalculates immediately so you can see how sensitive a player is to the sample you chose.
DraftKings pricing encodes an expectation of production. Converting salary into an implied point total gives a second opinion that was produced independently of the model. When the model and the market agree, confidence rises. When they disagree sharply, the player is flagged rather than promoted, because one of the two views is going to be wrong.
Offensive players priced under roughly $4,800 are usually backups, and their per-game averages come from garbage time or a single unusual week. Those players receive a heavy plausibility discount that ramps back to full credit at position-specific starter salaries. Team defenses are exempt, since cheap defenses are frequently legitimate plays rather than bench pieces.
Questionable, doubtful and out designations apply graded penalties. A player who may not play cannot be as plausible as one who is certain to start, no matter how attractive the raw value looks on the board.
Because TuttleMath runs conservative by design, a projection landing 12 to 28 percent below the season average is treated as normal and coloured accordingly. Anything outside that band is highlighted so you can investigate the cause instead of assuming the model made a mistake.
The blend sort normalises each chosen category into percentiles so that salary, points and efficiency can be compared on one scale. Your first preference carries the most weight, the second slightly less, and so on down to five. That produces a single ordering that reflects your priorities rather than a generic ranking.
Start with position filters to narrow the pool, then sort by plausibility to see who the model actually trusts. From that shortlist, switch to DSPP or points per $1,000 to find which trusted players are also efficiently priced. Only then look at BCS, because ceiling is the tiebreaker between similar options rather than the opening question.
Watch for players who appear near the top under several different sorts. Consistency across independent measures is the closest thing to a strong signal that this kind of modelling can offer. Conversely, a player who tops exactly one column and disappears from the rest is usually an artefact of that single metric, not a hidden edge.
Finally, remember the limits. These figures describe expectations, not outcomes. A perfectly reasoned lineup can lose and a careless one can win; the point of working through the math is to make good decisions repeatedly, not to control any single Sunday.