How yafss projections work
A look under the hood at how every projection on the site is built — and how much to trust it.
Every projected score on yafss is built the same way: estimate how many minutes a player will get, estimate how many points they score per minute, and multiply. This page explains how each half of that is worked out, and — just as importantly — what a projection can and cannot tell you.
Step one: projected minutes
Minutes are projected differently depending on where a player lines up:
- Backs — fullback, wing, centre, five-eighth and halfback — are projected to play the full game, because they are rarely rotated.
- Forwards share a fixed pool of minutes across each team's pack. Every forward gets an initial estimate from their own history, then those estimates are scaled so the pack's total comes out right. Heavy rotation reduces everyone; a short bench concentrates minutes into the starters.
Each player's individual starting estimate comes from their recent minutes, weighted so newer games count for more, with the current season weighted above previous ones. Outlier games are trimmed, and short injury-affected cameos are filtered out so one ten-minute night doesn't drag an established starter's projection down. Where a player has no usable history, the model falls back progressively — current season, then all seasons, then players in the same position group — rather than guessing.
Step two: points per minute
Scoring rate is calculated per minute rather than per game, which keeps it comparable across players with very different roles. The model blends three things:
- Recent form — the player's own points per minute, with the last handful of games counted at full weight and older games decaying away.
- Longer-term history — the same player over a much larger sample, so one hot month doesn't fully define them.
- A positional baseline — how players at that position typically score, used as an anchor.
Why small samples get pulled towards average
When a player has few minutes on record, their own numbers are weak evidence, so the model shrinks their rate towards the positional baseline. A player with two enormous games will be projected well below those games; a player with two poor ones will be projected above them. As minutes accumulate, the player's own record progressively takes over.
This is the single most common source of "the projection looks wrong" — and it is usually the projection being appropriately cautious rather than being broken. Genuine debutants are anchored to a rookie-specific baseline rather than to established players at their position.
What a projection is — and isn't
A projection is an expected average, not a prediction of Saturday's score. Real NRL scores scatter widely around any expectation. The projection is a tool for ranking players and for judging whether a breakeven is beatable — not for calling an exact number.
It follows that the right way to use projections is comparatively. "Is A projected above B?" is a question projections answer well. "Will A score exactly 58?" is not.
When projections appear — and when they don't
Projections on yafss are tied to team lists. A player only gets a projection for a round once they've been named in it, which avoids showing a confident number for someone who isn't playing. Players with a recorded absence are handled separately rather than being projected as if available.
Putting projections to work
- Against breakeven. Projection comfortably above breakeven means the player should make money; comfortably below means they should lose it. yafss surfaces this directly as hold / watch / sell signals on player pages.
- Against price. Compare projection to the player's priced-at score (see the magic number) to judge value independently of recent price movement.
- Against their position. Position tiers rank every player by projection within their position, so you can see where the real drop-offs are.
- Across a whole squad. The optimiser and bye planner both run on projections, which is what lets them choose a lineup or plan bye coverage rounds ahead.
Position Tiers Optimiser Player Finder
Where projections struggle
Being honest about the failure modes makes them easier to work around:
- Announced role changes. A player moving to a new position or a new starting job has history that describes their old role. The model catches up over a few weeks; you can see it immediately in the team list.
- Returning from long absences. Minutes are often managed on return, and that intent isn't in the historical data.
- Late changes. A projection reflects the named side. Positional reshuffles after a late withdrawal can change a player's outlook substantially.
In each case the fix is the same: use the projection as the baseline, then adjust for the information you have that the model doesn't.