14 minutes, then 25
One backup center was projected for 14 minutes. He played 25.
His team's starting center missed the game, and the backup stepped into the starting role. Over the previous season and the early weeks of this one — 71 of his 79 games in the sample — he'd been a bench player getting 12 to 15 minutes against opposing backups. That's what a full-year average sees. It counted all those games honestly. It had no way of knowing he was now a starter.
At roughly one point of projected output per extra minute — the average rate for a healthy rotation player — an 11-minute gap is worth about 11 points of projected production. The numbers published for his individual stat lines were built on the wrong version of his job.
This is why the board posts projected scores for every upcoming game, but doesn't call a winner until the starting lineup is confirmed. You'll see both teams' numbers as soon as the game is listed. The recommendation — the edge, the lean — holds until the starters are known. Not because the earlier numbers are hidden or untrustworthy. Because a projection built on an expected lineup and one built on a confirmed lineup are two different things, and the board won't call them the same.
What a lineup actually changes
A lineup isn't just a roster. It's an allocation of minutes, and minutes are what everything else flows from.
When a starter misses a game, those minutes don't spread evenly across five players on the bench. They concentrate. The backup who's been playing 12 minutes might be getting 25 or 28 now. The player whose role was 'third string when the second guy rests' becomes the second guy for real. Someone else's ceiling for the night just moved significantly.
The cascade runs in both directions. Teammates who were getting extra run when the starter sat go back to their normal load. The player stepping up draws more defensive attention, which shifts the matchups around him. These aren't subtle adjustments — they're the difference between a guy hitting his prop line comfortably and a guy who can't get there even on a good night.
Before lineups confirm, the model prices on what it expects. That expectation is usually close: the most likely replacement is known, the minute projection reflects the anticipated situation. But expected and confirmed are different things. Unexpected scratches happen. Undisclosed rest decisions happen. Last-minute lineup changes happen with enough regularity that a recommendation built only on expectations is a recommendation built on incomplete information.
When one player is the whole story
In football, the same principle shows up in a single player more sharply than anywhere else in sports.
Switching a team's starting quarterback for a backup moves their projected team score by roughly four points. Not four points added to both sides — four points subtracted from one team. A team projected to score 24 with their starter is a 20-point team with a backup in. That's a meaningful number. It's the kind of difference that changes which side of a spread you'd prefer.
The measurement ran across three seasons, tracking teams against themselves each time the starting quarterback changed, adjusting for who they were playing. The gap held in every season, not just in aggregate across the full sample.
Before the official injury list posts 90 minutes before kickoff, that question is open. The projection is running with the most likely answer. A quarterback who's practiced all week but is listed as questionable might start and might not. Those two outcomes don't produce the same projected score. The board knows the difference between 'we think he plays' and 'he's confirmed to play' — and it doesn't recommend the same side until the second one is true.
The window the board waits for
The confirmation window is specific to each sport.
In basketball, official lineups are released before tip-off — typically in the 30-to-90-minute window before the game, through the league's own announcements and the confirmed sources that track them. The model updates its projections as that information arrives. Players get adjusted minute estimates based on recent games, not the full-season average that undervalued the backup center in the first section. The changes run through the whole rotation.
In football, the official inactives list posts 90 minutes before kickoff. That's when the puzzle locks. A player listed as questionable all week either plays or doesn't. The team either starts their quarterback or they don't. Everything the projection needs to know about who's on the field has an official answer.
The board shows projected numbers before that window closes — both teams' scores, the spread, the total, right there on the matchups page as soon as the game is listed. But a recommended side doesn't post until the starters are confirmed. Not because the earlier numbers aren't honest. Because a recommendation made on expected information and one made on confirmed information are different calls, and the board treats them differently.
How the model closes that gap
There's a specific reason the system projects minutes from recent games rather than from a season-long average.
A flat full-season average doesn't know that a player changed roles. It just counts. A reserve who played 12 minutes a game for the first 70 games of a season and then stepped into a starting role for the last eight has a season average that looks like 13 minutes. That number is accurate about the past. It tells you nothing about what his job currently is.
The model now weights recent games heavily when it estimates playing time — the last several games count far more than games from earlier in the season or last year. If the recent evidence says a player is getting 22 minutes, that sits much closer to tonight's projection than 70 games of backup usage from before the starting role was his.
That adjustment closes most of the gap before the lineup is even confirmed. The backup center in the first section, priced off recent games instead of the full year, would have landed much closer to 25 minutes than 14. But closer is still not the same as confirmed. The last step — knowing who's actually starting, in tonight's actual game — is what turns an estimate into a priced call. The board waits for that step.
What this doesn't show
It doesn't tell you how often expected lineups differ from confirmed ones. The cases above — a backup stepping into a starting role, a quarterback's availability in question — are the clearest examples of a wide gap between expectation and reality. How frequently that gap shows up across a full season, and by how much on average, isn't something we've published, because it depends on how you define 'expected' and the sample varies meaningfully by sport.
It doesn't say the confirmed lineup eliminates projection error. Even with confirmed starters, a minute estimate can be off. A foul situation can shorten someone's night. A blowout can change the rotation. A coaching decision made at halftime doesn't appear in any pre-game announcement. Knowing who's playing is the most important thing the model needs to know — it's not everything.
And it doesn't say the board always has a recommendation after lineups post. Some games are priced close enough to the market that there's nothing to recommend. The confirmation opens the window for a call; it doesn't guarantee one. If the evidence isn't there once starters are settled, the board says so rather than manufacturing a lean from a game it doesn't have an edge on.
The numbers visible before the lineup lands are real projections of real games. They're worth looking at. They're just not the same as a call — and the board holds that line.
Sources
Every number here comes from the graded ledger the boards publish. Research from a statistical model, not betting advice; no outcome is guaranteed. Method: how MatchWiz works.