Forensic Hockey
Methods

Research provenance

A durable record of every external source we've read - what it claims, whether it held up when we reproduced it on our own data, and what shipped. We never adopt a claim on authority alone. That includes the ideas we tested and rejected - the honest part most sites never show.

27

Sources read

17

Ideas adopted

16

Tested, not adopted

23

Set aside

RYDER-04Shot QualityAlan Ryder · Hockey Analytics, Jan 2004
AdoptedTested, not adoptedAlready hadSet aside
  • Already had: per-shot xG; team Expected Goals; GSAx = his SQNSV idea
  • Adopted: the three-index Goal-Prevention Decomposition
  • Tested, not adopted: rebound danger as an explicit defensive sub-metric
BERNIER-18Forecasting Real-Time Win Probability in NHL GamesChristophe Bernier · Boston College, 2018
AdoptedTested, not adoptedAlready hadSet aside
  • Adopted: a pre-game prior with a time-decaying weight, blended onto our live WP curve
  • Tested, not adopted: Tested / worth pursuing later
SCHUCKERS-16Statistical Evaluation of Ice Hockey GoaltendingMichael Schuckers · book chapter, 2016
AdoptedTested, not adoptedAlready hadSet aside
  • Adopted: the goalie-rating uncertainty treatment
  • Tested, not adopted: Tested, worth pursuing (logged, not yet built)
RABER-21Defensive Efficiency Metrics (DEMs)Matthew Raber & Daniel Eisenberg · Big Data Cup, ~2021
Tested, not adoptedAlready hadSet aside
  • Already had: our /goal-prevention decomposition IS Raber's "keep the components separate" design
  • Tested, not adopted: Tested, worth pursuing (logged, not built)
SIIVONEN-24Predicting Player Performance in the NHL using Data Analytics and Advanced MetricsOnni Siivonen · LUT bachelor's thesis, 2024
AdoptedTested, not adoptedAlready hadSet aside
  • Adopted: the skater finishing-reliability treatment
  • Tested, not adopted: Tested, worth pursuing (logged, not built)
CAMPBELL-22Predicting Productivity of Hockey Players via Mixture Models (Empirical Bayes Methodology)Connor Campbell · U. Alberta MSc thesis, 2022
AdoptedAlready hadSet aside
  • Adopted: the scoring-rate empirical-Bayes "true rate" treatment
ADDONA-10A Closer Look at the Relative Age Effect in the NHLVittorio Addona & Philip A. Yates · JQAS, 2010
AdoptedAlready had
  • Adopted: an evergreen Forensic Take
  • Already had: Already had / ⚪ not adopting
MATSUZAWA-17Using Machine Learning to Predict Future Points in the NHLTakehiro Matsuzawa · Harvard bachelor's thesis, 2017
Tested, not adoptedAlready hadSet aside
  • Already had: His findings reproduce (and we already embody them)
  • Tested, not adopted: The headline is inflated (two ways)
  • Set aside: Not adopting - no feature shipped (the honest outcome)
STIHL-23The Impact of Winning Faceoffs on Expected Goals (xG) in Power playsElias Stihl · KTH / LINHAC, 2023
AdoptedSet aside
  • Adopted: an evergreen Forensic Take
  • Adopted: Power-Play Forensics - the faceoff as the front of a PP luck chain
  • Adopted: the mirror, and which players control the draw
DAYARATNA-12The Pythagorean Won-Loss Formula and HockeyKevin Dayaratna & Steven J. Miller · Hockey Research Journal, 2012/13
Adopted
  • Adopted: the goals→wins lens that completes the luck chain
WEISSBOCK-14Forecasting Success in the National Hockey League using In-Game Statistics and Textual DataJoshua Weissbock · M.Sc. thesis, U. Ottawa, 2014
AdoptedTested, not adoptedAlready hadSet aside
  • Adopted: two evergreen Forensic Takes
  • Adopted: the Series Forecast (pre-series win probability)
  • Adopted: a compact bracket forecast chip
THOMAS-07Inter-arrival Times of Goals in Ice HockeyAndrew C. Thomas
AdoptedTested, not adoptedAlready hadSet aside
  • Adopted: two evergreen Forensic Takes
  • Tested, not adopted: Tested, worth pursuing (logged, not built)
SILVEY-20National Hockey League Data: Analysis and PredictionHoward Silvey · Carleton honours project, 2020
Tested, not adoptedAlready hadSet aside
  • Tested, not adopted: The valid kernel (correctly attributed; logged, not a feature)
FOUND-16Goal-based Metrics Better Than Shot-based Metrics at Predicting Hockey SuccessRob Found · The Sport Journal, 2016
Tested, not adoptedAlready hadSet aside
  • Tested, not adopted: Tested, worth noting (logged)
XU-23Risk, Reward, and Reinforcement Learning in Ice Hockey AnalyticsSheng Xu, Oliver Schulte, Yudong Luo, Pascal Poupart, Guiliang Liu · book chapter, ~2023
AdoptedTested, not adoptedAlready hadSet aside
  • Adopted: Game Score CONSISTENCY (the Sharpe-ratio risk-adjusted lens)
  • Tested, not adopted: Tested / worth pursuing (logged, not built)
SADEGHKHANI-21Predicting the Scoring Time in HockeyAbdolnasser Sadeghkhani & S. Ejaz Ahmed · J. Statistical Theory and Practice, 2021
Already hadSet aside
OLSSON-26Modeling Player Evaluation in Ice Hockey Using Machine LearningMathias Olsson · KTH MSc thesis, 2026
Already hadSet aside
PILEGGI-12SnapShot: Visualization to Propel Ice Hockey AnalyticsHannah Pileggi, Charles D. Stolper, J. Michael Boyle & John T. Stasko · IEEE TVCG / InfoVis, 2012
Already hadSet aside
LI-04Description, Analysis and Prediction of Player Actions in Selected Hockey Game SituationsFahong Li · UBC MSc thesis, 2004
Already hadSet aside
ROUTLEY-15A Markov Game Model for Valuing Player Actions in Ice HockeyKurt Routley · SFU MSc thesis, 2015
AdoptedAlready hadSet aside
  • Already had: Already had (the overlaps)
KASKENMAA-23Using Data Analytics In Hockey Player Talent IdentificationMarkku Kaskenmaa · Oulu UAS MSc thesis, 2023
Tested, not adoptedAlready hadSet aside
  • Already had: The survey is a near-checklist of what we already build/processed
  • Tested, not adopted: The one novel-to-us testable kernel - streak MEMORYLESSNESS - scripts/analyze_kaskenmaa_streaks.py
  • Tested, not adopted: Two more kernels, FRESHLY TESTED (not asserted) - analyze_kaskenmaa_important_goals.py, ..._dcorsi.py
PISCHEDDA-14Predicting NHL Match Outcomes with Machine Learning ModelsGianni Pischedda
Already hadSet aside
  • Already had: Already had / already tested (the bulk of the paper)
  • Set aside: / ✅ verdict - no NEW metric; one existing take ENRICHED
MACDONALD-11A Regression-Based Adjusted Plus-Minus Statistic for NHL PlayersBrian Macdonald
AdoptedTested, not adoptedAlready hadSet aside
  • Adopted: RAPM - the 2nd new player-valuation metric from a source paper (after Action Impact)
  • Tested, not adopted: Tested / logged (his future work, not yet built)
DOUGLAS-21Valuing Individual Contributing Events (V-ICE) in HockeyEthan Douglas, Sean Clement, Nick Wan & Ian Greengross · ~2021
Tested, not adoptedAlready hadSet aside
  • Already had: Already had - V-ICE's purpose IS our Action Impact (and richer where it counts)
  • Tested, not adopted: Tested, the one piece worth a later look (not a player metric)
  • Set aside: Not adopting (the player metric)
RYDER-04bPoisson Toolbox: a review of the application of the Poisson Probability Distribution in hockeyAlan Ryder · Hockey Analytics, 2004
Already hadSet aside
  • Already had: Already had - and the Toolbox is its derivation (each cross-checked, not asserted)
  • Set aside: Not adopting - the one distinctive capability is real but largely redundant
PITASSI-25Puck Possession and Net Traffic Metrics in Ice HockeyMiles Pitassi · University of Waterloo MMath thesis, 2025
Tested, not adoptedAlready had
  • Already had: Already had / ⚪ not adopting
  • Tested, not adopted: Logged - a genuine capability gap we cannot fill without PPT
JABLONSKY-21Individual and team efficiency: a case of the National Hockey LeagueJosef Jablonsky · Central European Journal of Operations Research, 2021
Already had
  • Already had: Already had / ⚪ not adopting (ship nothing)

Each source is tagged [AUTHOR-YY]; when an idea lands in code, the commit cites the tag, so any model choice traces back to its origin. “Set aside” means we tested it and it didn't beat what we already had - sometimes it was outright refuted on our data.