How it works
- 01
The numbers
Results come straight from the league's Yahoo data. Each team's daily lineup is added up player by player, counting only starters, and the build refuses to publish unless those daily lines match Yahoo's weekly totals exactly. Every statistic in the write-ups is filled in from that data by the build, never typed in.
- 02
The decider
A 5–4 week turns on one category. The decider is the closest category the winner took: flip it and the result flips. Close percentages are shown to two or three decimals when that is what separated them.
- 03
The full story
Every matchup has its own page: the week told night by night, how each player went, a box score for both rosters and the category score after every day. The box score's stat totals must equal the category table, and its games played plus starts sat out must equal Yahoo's games-played count, or the build stops. Yahoo counts a game whenever a starter's team plays, even if he sits out, so the page shows those nights separately as starts sat out. Yahoo records them as a blank line, which can't tell a night off from a scoreless minute; both count as sat out here.
- 04
Top performers
Players are ranked by Yahoo's standard fantasy points (points, rebounds, assists, steals, blocks, minus turnovers), counted only while they were in a starting slot.
- 05
Surprises and letdowns
Each player's week is measured against a projection: ESPN's pre-season numbers in week one, then the projection the model used for that week's calls (made before the first game from week three; week two's calls, and so its projection, came after day one). The biggest surprise gained the most fantasy value over the week against that expectation. The biggest underperformer lost the most, from among the highest-projected players who played. Both need at least two games in a starting slot.
- 06
The calls
A model projects each roster's week: ESPN's per-game projections, overtaken by each player's actual numbers as games accumulate, each team's real NBA schedule, daily lineup slots and injury status. It works out every category, names a winner, the chance of it and the most likely score. Week one used calls locked before the season started. Week two's were made on current rosters after day one, starting from the score so far. From week three the model makes its calls every Monday on current rosters, before the first game.
- 07
The record
Joey's spreadsheet made its calls in public and the league held it to account. Same deal here. Every call stays on the page. A call is right when it names the winner, and from week two an exact final scoreline is counted separately. Every team is modelled the same way and every call is scored the same way.