Everything on NHLTrades is one of two things: a record of something that happened, or an estimate of what it was worth. This page explains where the records come from, how the estimates are made, and how we check them. The short version: the trades, picks, scores, stats and contracts are facts we collect and verify; the grades, values and ratings are our models' best read on those facts, and they are labelled as such wherever they appear.
Records: what we collect
- Trades. Every NHL trade from the league's early seasons to today, more than 6,000 in all. For each one we record the date, the teams, and every asset that changed hands: players, player rights, draft picks, cash, retained salary, loans and future considerations. A pick is tracked as its own asset from the team that originally owned it, through each trade, to the team that used it and the player it became. Conditions on a pick are recorded with whether they were met and how they resolved.
- Drafts. Every draft board, with each selection linked to the player chosen and to the ownership chain of the pick used.
- Games and statistics. Results, boxscores and player statistics from the league's public data feeds, back to the league's first season. Older seasons carry less detail (ice time, shot locations and many event-level statistics only exist for the modern era), and the models account for that rather than guessing at numbers that were never recorded.
- Contracts. Terms, cap hits, length, clauses and signing details for the salary-cap era (2005 onward), gathered from public contract reporting. Many contracts before 2005 have no reliable dollar figures in any source, so those stay blank rather than estimated.
- People. Players, general managers and coaches, with their tenures, so that each trade is credited to the front office that made it and each season to the coach behind the bench.
New trades are added as they happen. During the season, scores, stats and rosters are refreshed overnight, so the morning after a game everything is current. Contract data is refreshed through the day.
How records are verified
Each asset in a trade is matched to a single identity for the player or pick involved, so a player who was traded four times appears on one page, not four, and a pick's ownership chain has to connect from the team that first held it to the selection it became. Trades are reviewed before they are published. Readers report the rest: every correction sent through Contact or Feedback is checked against the record and fixed.
Estimates: how the models work
The analytics on the site, trade grades, Trade Value, WAR, archetypes and team and coach ratings, are model estimates. They are built in layers, each resting on the one below.
Era adjustment
Scoring in the NHL has swung widely: the 1980s produced far more goals per game than the dead-puck years around 2000. Before a player from one era is compared with one from another, their numbers are scaled to a common reference season using the league-wide scoring environment of the year they played. Schedule length and the number of teams are adjusted the same way.
Player value and WAR
A player's contribution is measured against a replacement-level player, the kind any team could call up or claim on waivers, and expressed in wins. That puts skaters and goalies on one scale. Where the data allows (the modern era), the measure uses shot quality, ice time and the strength of teammates and opponents; for older seasons it falls back on the statistics that exist, with wider uncertainty.
Picks, contracts and Trade Value
A draft pick is valued by what picks in that slot have historically produced, and a pick that has not been used yet is projected to a range of slots from where the owning team is expected to finish. A contract adds to or subtracts from what a player is worth as an asset, depending on how his production compares with what he is paid. Trade Value weighs how good a player is now, his age and his contract into one figure that can be set beside picks and other players.
Trade grades
Each side of a trade is graded on the value it gave up against the value it got back, measured as of the day of the trade and with nothing that happened afterward allowed to leak in. The difference is scaled to the size of the deal, so a small edge matters in a depth swap and not in a blockbuster, and then compared with the trades of its era to produce a letter and a percentile. Each grade carries a confidence figure, lower where the record behind it is thinner.
Realized outcomes
Separately from the day-of grade, we track how each trade actually aged: what every player did while with the acquiring team, what every pick became, and what any re-trades returned. The two are kept apart on purpose. The grade is a judgment of the decision; the realized outcome is the result.
Archetypes, team and coach ratings
A player's archetype comes from grouping seasons that look alike across several areas of play, so "two-way center" or "stay-at-home defenseman" is a description of how someone played, not a label assigned by hand. Team strength over time uses an Elo rating that rises with wins and falls with losses; it sets what a team was expected to do each season, and coaches are rated against that expectation.
How estimates are checked
- Against outcomes. Day-of grades are compared with how the trades and careers actually turned out. A model that grades a deal well in the moment and badly a few years later is a model we revise.
- With a fixed test set. A set of well-known trades, lopsided and even, is graded by each new version of the model and checked against what we expect of it before the version goes live.
Because of this, a number can change: as a season plays out, as picks resolve, and as the models improve. The "Model estimates" note that appears over grades and analytics links back to the FAQ for that reason.
What we don't do
- We don't invent numbers where the record is silent. Pre-cap salaries, ice time before it was tracked, and shot locations before they were recorded stay blank.
- We don't let hindsight into a day-of grade. The realized outcome is reported beside it, not folded into it.
- We don't present an estimate as a statistic. Scores, stats, contracts and the details of each trade are records; everything graded, valued or rated is a model estimate and labelled as one.
Questions and corrections
The glossary defines each term used above, and the FAQ covers the common questions. If something looks wrong, a record or an estimate, tell us through Contact. That is how both get better.