-3.5 to -10: the Week 2 spread that got less settled the more sources checked it

Most line disagreements close as kickoff gets closer. This one didn't — it widened, and the source that widened it is the one a lot of GameScript builds start from.

Four days, four numbers

SEA@ARI kicks off Sunday at State Farm Stadium. A daily research scan checked this specific line against multiple sources, every day, since six days out. Here is what "Seattle favored by" read, source by source, as each day's check found it:

† The one source all week to disagree on direction rather than magnitude — Arizona, not Seattle, favored. It did not recur on any later check. ‡ Both numbers checked again on 9/17 came back from pages dated 2026-09-15 (a Covers.com/Kalshi snapshot and an SI.com/onsi report of DraftKings' line) — reconfirmed, not fresh that day.

Direction settled fast: every source has favored Seattle since 9/15. Magnitude never did. On 9/16 — four days before kickoff — the spread across sources actually checked that day ran from GameScript's own -3.5 up to DraftKings' -10, a 6.5-point range on a single NFL spread, with every number coming from a live, direct fetch rather than a stale cache. That range is unchanged as of today's check, three days out.

The total, over the same four days, moved in a much narrower band: 44.5 at the open, settling to 40.5–41.5 everywhere it's been checked since 9/15. Whatever is driving the spread disagreement isn't showing up as disagreement about how many points this game will produce — only about how those points split between the two teams.

Why the house number is worth naming specifically

GameScript's own game-environment tool, dfs_get_game_context, didn't enter this picture until 9/16 — before that, the environment reads in this repository's own research were built entirely from the same public web sources listed above. When the house number finally showed up, it didn't side with either cluster. It landed at -3.5, below even the "everyone but DraftKings" range of -4.5 to -5.5, making it, as of 9/16 and again on 9/17, the low outlier on the opposite end from DraftKings' -10.

That's not a knock on the house model specifically — it's a reason to know what it's reading. Per this repository's own sourcing notes (docs/data-sources.md), the house build's default odds source since September 1 is nflverse's games.csv, described there as carrying "a closing consensus rather than any single book's live number," refreshed roughly every twenty minutes. Days before kickoff, "closing consensus" is aspirational language for a number that hasn't closed yet — it's whatever that feed's rolling consensus reads today, and today it disagrees with every other source checked this week, including DraftKings' own board. A consensus feed and a single book's actively-traded line can diverge for ordinary reasons — different books in the sample, a book managing its own liability rather than publishing a probability estimate — but the size of this particular gap, and the fact that it's grown rather than closed, is the thing worth flagging rather than assuming away.

Why the 6.5 points actually matter

Run the same 41.5 total through the two ends of this week's range and the implied scores describe two different games, not two versions of the same one:

Both rows describe a below-median total on this slate — this was never going to be the shootout game. But a below-median total with a tight spread and the same total with a double-digit spread sit in genuinely different buckets for a build: the first is a game where a volume-independent piece can still be defensible; the second is closer to a game to leave alone unless you need a specific correlated garbage-time role. Which bucket SEA@ARI belongs in depends entirely on which of this week's numbers you trust, and as of today, nobody's numbers have converged enough to answer that.

On the player side, the one thing every source agrees on: Drew Lock starts at quarterback for Seattle with Sam Darnold out, and no update to that has surfaced on any check this week. That's settled. The shape of the game he's starting in is not.

The rule

A single sportsbook's line — even the one running the slate you're playing — is a risk-managed number, not a probability estimate, and a "consensus" feed is only as settled as the market it's sampling that day. When multiple independently-checked sources land within a point of each other, treat that as real agreement. When they don't — and a 6.5-point spread on one game, growing rather than shrinking across four days of direct checks, is not noise — don't average the numbers and call it a lean. Note which shape each candidate spread implies, note that the total is the more stable half of this particular line, and treat any spread-dependent read on this game as provisional until a source or two closes the gap. For SEA@ARI, that hasn't happened yet, and it's worth one more direct check — DraftKings, FanDuel, and the house build all named specifically — before lock.