The Selection Bias Index

Why raw points lie about young prospects

Two prospects in the same league can post identical point totals and be radically different players. One was the team's first-line centre with the top defenceman on the ice, double the offensive-zone starts, and 90 seconds of power-play time every night. The other was a third-line winger with checking-role assignments, defensive-zone starts against opposing first lines, and no special-teams role. Both scored 50 points. Their season-on-season trajectories, their NHL translation odds, and their actual on-ice impact are completely different — but a raw points-per-game ranking treats them identically.

This is a selection-bias problem: the production we observe is conditional on the opportunity the coach selected the player for, not on the player's underlying talent alone. SBI is the layer of the ProdigyChain algorithm that decomposes the opportunity from the output so the ranking reflects the player, not the deployment.

How the bias correction works

For every player-season SBI estimates a deployment vector — what kind of usage the player received — then re-weights production against that vector to produce an opportunity-adjusted score. The output is a scalar that downstream factors (including F17 and F19) consume so league-level and draft-year scoring already incorporate the opportunity correction by the time they reach the published rank.

SBI does not try to reconstruct shift-level NHL-style microstats from junior or European data that often doesn't track them at the same granularity. Instead it uses the deployment proxies the source leagues do publish reliably — minutes, role tier, zone-start ratio where available, special-teams time, on-ice goal differential, and line-context inferred from the team's scoring distribution — and compresses them into a single deployment score per player-season.

The five deployment inputs

SBI reads five deployment signals per player-season and folds them into the opportunity vector:

• Ice time and role tier.Total minutes, but more importantly the role bucket — first-line / second-line / third-line / fourth-line for forwards; top-pair / second-pair / third-pair for defencemen. Role bucket is inferred from minutes relative to teammates when the league doesn't publish it directly.
• Zone-start ratio.Where the player's shifts begin — offensive zone, defensive zone, or neutral. CHL, NCAA, USHL, and the top European leagues track this with varying granularity; SBI uses the published figure where available and falls back to a team-context prior when not.
• Quality of competition and quality of teammates. Whether the player is matched against opposing top lines and whether he's deployed alongside the team's best linemates. SBI uses on-ice teammate scoring rates and head-to-head shift overlap to estimate both — a 70-point season with the league's best centre as a linemate is read differently than 70 points with third-line teammates.
• Special-teams deployment. Power-play time per game and penalty-kill time per game. A prospect with 4:00 of power-play time per night has a structurally inflated point baseline; SBI discounts power-play points against power-play opportunity and surfaces even-strength production as the most predictive signal.
• On-ice goal differential at even strength.The team's goal differential when the player is on the ice, relative to the team's overall rate. A signed-impact estimate when shift-level microstats are missing — useful for two-way forwards and defencemen whose value isn't fully captured by the box score.

These five inputs are not equally weighted across all leagues — the algorithm learns league-specific weights from historical NHL-translation outcomes because the same nominal deployment in the WHL and the SHL has different downstream meaning. The output is a single deployment-adjusted production score per player-season that downstream factors read directly.

Why this matters for rankings

Without SBI a prospect ranking inherits the coach's deployment decisions as if they were the prospect's own talent. A second-line centre with strong underlying play but a top-line teammate sitting ahead of him on the depth chart looks like a worse prospect than he is; a power-play specialist with limited five-on-five impact looks like a better prospect than he is. Both cases are recurring blind spots in pure points-per-game rankings — and both are exactly what SBI is built to detect.

SBI also explains why two prospects who finish neck-and-neck in raw scoring can be ranked far apart on ProdigyChain: the one with first-line minutes, offensive-zone starts, and a star linemate is scoring inside a deployment greenhouse; the one with checking-role minutes against opposing top lines is scoring against the grain of the assignment. SBI surfaces that difference and the published rank reflects it.

Known limitations

• Sparse-deployment leagues.Some European development leagues and second-tier junior loops don't publish zone starts, head-to-head shift data, or special-teams breakdowns at all. SBI falls back to a team-context prior that uses team scoring distribution to estimate role, but the confidence indicator on the player profile is dropped to reflect the imputed-rather-than-measured deployment signal.
• Single-tournament samples.IIHF U18 and U20 tournaments have rotating linemates and rotating opposition so deployment varies shift to shift in ways that don't aggregate cleanly to a season-level vector. SBI shrinks tournament results toward the player's season-long deployment baseline rather than overwriting it.
• Mid-season role changes. A prospect promoted from the third line to the first line mid-season has two distinct deployment vectors. SBI emits a weighted blend by games and surfaces the split on the player profile so readers can see the underlying role change rather than just the blended SBI score.
• Goaltender deployment.SBI is a skater factor — goaltender opportunity adjustment (workload, team save percentage, shot-quality faced) is handled by a separate goaltender-specific model and isn't represented in the SBI score.

Each player profile shows a per-factor confidence indicator next to SBI so readers know when the opportunity adjustment is operating on robust deployment data and when it's working from imputed proxies in a less-instrumented league.