The widest published Week 1 gaps are NYJ at TEN on the total, BAL at IND on the spread, and ARI at LAC on the spread asOf 2026-09-01T18:55:36.615445Z. Each read is falsifiable now: refresh the scoring inputs, or move the market line through the model projection, and the conclusion weakens or reverses.
This is the board through a deliberately plain house model. It reads season-to-date points scored and points allowed for both teams from played NFL schedule rows. For a total, it averages each offense with the opposing defense and adds both expected scores. For a spread, it subtracts the expected away score from the expected home score and applies the model’s fixed home-field adjustment. The market line is the comparison threshold, not an input to those team-feature projections. A Sharkai context is attached upstream, but these scoring formulas do not read it asOf 2026-09-01T18:55:36.615445Z.
The total the board disagrees with most
NYJ at TEN: over. The published market total is 38.5 and the house projection is 46.01, a 7.51-point gap asOf 2026-09-01T18:55:36.615445Z. Those are the paired numbers that matter here. The projection is not a win probability, and the gap is not a promise that the game clears the total.
The drivers are Tennessee’s and New York’s season-to-date offensive points per game and defensive points allowed per game from played schedule rows. The model blends each offense with the defense it faces, then sums the two expected scores. It does not know whether a recent score came from a short field, a return touchdown, or garbage time. It sees points, opponent points, and the current market threshold.
What changes the read: refreshed scoring rates that pull the expected total down, or a market total that rises to or beyond 46.01 asOf 2026-09-01T18:55:36.615445Z. Either would erase the positive gap. Injury news, weather, and matchup detail matter only after they enter a measured feature or move the line; this baseline does not silently pretend to understand them.
The home underdog the model makes a favorite
BAL at IND: home. The stored spread is -3.5 under the feed’s positive-when-home-favored convention, so Baltimore is favored on the road. The model projects a 2.55-point Indianapolis home margin. That puts the model and market 6.05 points apart asOf 2026-09-01T18:55:36.615445Z.
The scoring-rate quartet drives the raw expected margin: each team’s offensive points per game and defensive points allowed per game. The spread model then applies its fixed home-field adjustment. It does not read quarterback health, protection matchups, coaching tendency, or the attached Sharkai fields. Calling Indianapolis the home side is therefore a narrow statement about this formula, not a complete football argument.
What changes the read: Indianapolis’s expected scoring margin falling, Baltimore’s rising, or the market threshold moving up to the 2.55-point model margin asOf 2026-09-01T18:55:36.615445Z. If projection and line meet, the coded spread gap disappears; if they cross, the side reverses.
The favorite the model will not lay
ARI at LAC: away. The market makes the Chargers a 10.5-point home favorite, while the model projects a 4.89-point Chargers margin. Arizona sits on the model side because the projection is 5.61 points below the market threshold asOf 2026-09-01T18:55:36.615445Z.
Again, the evidence is the team scoring-rate quartet plus the fixed home-field adjustment. That repetition is a feature of the baseline and a warning about it. These large gaps do not mean independent insights when the same compact input family generates all of them. A noisy scoring profile can move several reads together.
What changes the read: scoring updates that lift the projected Chargers margin toward 10.5, or a market line that falls toward the 4.89-point projection asOf 2026-09-01T18:55:36.615445Z. Once the line meets or crosses the projection, Arizona is no longer the model side.
No record yet means no record
The public NFL spread ledger has no eligible live graded outcomes asOf 2026-09-01T18:55:35.83182Z. In user language: this house model does not have a live graded record yet. The first published house pick is scheduled for September 9 asOf 2026-09-01T18:55:37.225438Z, so the record stays empty until the September 9 games finish and are graded.
That absence is not evidence for or against the three reads asOf 2026-09-01T18:55:35.83182Z. It is the starting line. Watch the published projections and market thresholds refresh before kickoff, then watch whether the eventual eligible results support the gaps. The read changes when the inputs, the price, or the evidence changes.
VERIFY
Run these read-only production queries. Q1 returns the three published line gaps above; Q2 proves the live graded-record absence; Q3 finds the first future published NFL house-pick kickoff. Query timestamps are the displayed asOf values.
-- Q1: public.picks published NFL house spread/total divergences
with eligible as (
select id, created_at, market, game_id, subject, line, side, predicted_value, edge, kickoff_at,
row_number() over (order by abs(edge) desc, created_at desc, id asc) as divergence_rank
from public.picks
where sport='nfl'
and source_kind='house'
and published=true
and kickoff_at >= '2026-09-09T00:00:00Z'
and kickoff_at < '2026-09-15T00:00:00Z'
and market in ('spread','total')
and line is not null
and predicted_value is not null
and edge is not null
)
select now() as as_of_utc, *
from eligible
where divergence_rank <= 3
order by divergence_rank;
-- Q2: live-pregame NFL spread outcomes
select now() as as_of_utc,
count(*) filter (where pr.bet_outcome in ('win','loss','push')) as graded_ats_rows
from public.prediction_results pr
join public.predictions p on p.id=pr.prediction_id
where pr.sport_key='nfl'
and pr.prediction_type='game_spread'
and pr.provenance_tier='live_pregame'
and p.provenance_tier='live_pregame';
-- Q3: first future published NFL house-pick kickoff
select now() as as_of_utc, min(kickoff_at) as first_published_house_pick_kickoff_at
from public.picks
where sport='nfl'
and source_kind='house'
and published=true
and kickoff_at >= '2026-09-01T00:00:00Z';This is analytics, not advice. Bet responsibly, set your own limits, and never chase losses.
NFL ATS cover-margin distribution
Bars count completed NFL schedule rows by closing-spread cover margin using the repository canonical home-margin grading convention.
Average NFL total points by recorded weather bucket
Average combined score is grouped only from completed NFL schedule rows with a recorded indoor roof state or numeric wind value.
