Skip to main content
EducationalNot betting advice · research use only.
GameTime Picks
Today· Sep 6Pregame slateSettled· Sep 5

NFL

Preseason archiveFri, Aug 14 – Sat, Aug 29times in ET

Games

22 scheduled · 22 with a report · 0 with a supported read · 22 started or final

Started or final

NFL · public betaUpcoming· 1 gameUpcoming · 2026-09-09
🏈

NFL

Experimental preseason simulations

Wednesday, September 9 · 1 game on the slate

Games on the slate12026-09-09
Simulated0none published
Sportsbook prices1captured 17:50Z
See the slateSportsbook prices

Experimental, educational, paper-only. This model has not been shown to beat the sportsbook market — nothing here is a pick or a recommendation to wager.

Experimental preseason simulations — not picks. Tested on a full held-out preseason, this model picked winners no better than a coin flip. Its win percentages are therefore deliberately pulled close to 50% — the widest it will ever go is 45.3%–50.6%.

Where the data comes from
Everything on this page derives from committed public captures, and the coverage table below states each layer's exact status and the reason, in words. These simulations are not a claim to beat the sportsbook market.
Slate · 2026-09-09

Wednesday, September 9 — 1 game

Every game on this slate, with the simulation we published for it before kickoff. 0 of 1 are simulated; a game without one says so rather than showing a blank. Tested on a full held-out preseason, this model picked winners no better than a coin flip. Its win percentages are therefore deliberately pulled close to 50% — the widest it will ever go is 45.3%–50.6%.

EXPERIMENTAL
regular season · week 1
New England Patriots logoNEatSeattle Seahawks logoSEA

Wed, Sep 9, 8:20 PM ET

Lumen Field

No simulation was published for this game.

Simulations generated 2026-08-29T18:16:23Z under model nfl-preseason-public-beta-v1 before every kickoff shown, and frozen at that moment — the forecast never changes after the fact, and each one is settled against the official result. Publishing it with its results tracked is how it earns — or fails to earn — a stronger status over time.

Per-game reports

Every forecast we published

Each one shows what the model said before kickoff, and is labelled with where that game now stands.

Schedule

Later this preseason

From the committed schedule capture (2026-09-05T15:41:47Z). Simulations publish inside each game's own event window, not weeks ahead.

regular season · week 1
San Francisco 49ers logoSFatLos Angeles Rams logoLAR

Thu, Sep 10, 8:35 PM ET

Melbourne Cricket Ground

regular season · week 1
Tampa Bay Buccaneers logoTBatCincinnati Bengals logoCIN

Sun, Sep 13, 1:00 PM ET

Paycor Stadium

regular season · week 1
New Orleans Saints logoNOatDetroit Lions logoDET

Sun, Sep 13, 1:00 PM ET

Ford Field

regular season · week 1
New York Jets logoNYJatTennessee Titans logoTEN

Sun, Sep 13, 1:00 PM ET

Nissan Stadium

regular season · week 1
Baltimore Ravens logoBALatIndianapolis Colts logoIND

Sun, Sep 13, 1:00 PM ET

Lucas Oil Stadium

regular season · week 1
Atlanta Falcons logoATLatPittsburgh Steelers logoPIT

Sun, Sep 13, 1:00 PM ET

Acrisure Stadium

regular season · week 1
Chicago Bears logoCHIatCarolina Panthers logoCAR

Sun, Sep 13, 1:00 PM ET

Bank of America Stadium

regular season · week 1
Cleveland Browns logoCLEatJacksonville Jaguars logoJAX

Sun, Sep 13, 1:00 PM ET

EverBank Stadium

regular season · week 1
Buffalo Bills logoBUFatHouston Texans logoHOU

Sun, Sep 13, 1:00 PM ET

Reliant Stadium

Prices · captured 2026-08-29T17:50:39Z

Sportsbook prices for this slate

These are the sportsbooks' own numbers, shown as facts with attribution — captured before every kickoff below, and GameTimePicks publishes no NFL prediction beside them. The win percentages are each book's price with its margin removed, then the median across books; they describe the market, not a forecast of ours.

GameKickoffBooksMarket win % (home / away)Spread (home)Total
Chicago Bears logoCHI atTennessee Titans logoTENSat, Aug 29, 6:00 PM ET952.7% / 47.6%-1.037.0

Source: The Odds API, median across the captured books per game. Prices move after capture; the stamp above is the moment these were read, and nothing here is a recommendation to wager. The same authorized capture probed player markets for this window and found none offered — so no player or touchdown prices appear anywhere on this page.

Results

Recent finals

Finals join by durable event identity; a result without pre-event schedule lineage is quarantined and reported, never settled.

  • WSH logoWSH 3 — 41 BALBAL logofinal · 2026-08-28
  • ATL logoATL 17 — 12 MIAMIA logofinal · 2026-08-28
  • HOU logoHOU 13 — 16 CARCAR logofinal · 2026-08-28
  • NYG logoNYG 23 — 6 NYJNYJ logofinal · 2026-08-28
  • TB logoTB 0 — 19 JAXJAX logofinal · 2026-08-28
  • NO logoNO 27 — 24 DALDAL logofinal · 2026-08-29
  • ARI logoARI 38 — 42 GBGB logofinal · 2026-08-29
  • SEA logoSEA 9 — 9 KCKC logofinal · 2026-08-29
  • CIN logoCIN 30 — 13 PHIPHI logofinal · 2026-08-29
  • MIN logoMIN 6 — 34 DENDEN logofinal · 2026-08-29
  • DET logoDET 25 — 16 INDIND logofinal · 2026-08-29
  • CHI logoCHI 24 — 15 TENTEN logofinal · 2026-08-29
Picks vs outcomes · 43 graded

How the model's picks actually turned out

Preseason game forecasts: which side the model favoured, graded against the official final.

Full record →
24 hit · 19 missed2 void — a condition that did not hold is never scored as a miss55.8% hit rate

43 graded picks is still a small sample. A hit rate over this many outcomes has a wide enough margin of error to be consistent with almost any true accuracy.

EXPERIMENTAL. The NFL team model has not cleared a preregistered bar and is not promoted into any paper product. A tie is recorded as void, not a miss.

DateEventModel saidWhat happenedResult
2026-08-29DET @ INDAWAYAWAYHIT
2026-08-29CHI @ TENAWAYAWAYHIT
2026-08-28LAR @ LACAWAYAWAYHIT
2026-08-29CIN @ PHIAWAYAWAYHIT
2026-08-28WSH @ BALAWAYHOMEMISS
2026-08-28NYG @ NYJAWAYAWAYHIT

Showing the 6 most recent of 43 graded.

Status

Coverage, market by market

Readiness is stated per layer — a page section is never filled merely because it exists.

LayerStatusWhat that means
Schedule & identitiesLIVEdaily capture · 16 events in window, captured 2026-09-05T15:41:47Z
ResultsLIVEofficial finals join by durable event id; results without pre-event schedule lineage are quarantined and say so
Team game simulationPUBLIC_EXPERIMENTALWe publish a simulated score range and win chance for every game, clearly marked experimental. Tested on a full held-out preseason this model picked winners no better than a coin flip (1 games forecast today under nfl-preseason-public-beta-v1), so its win percentages stay close to 50% on purpose and it makes no claim to beat the sportsbook market. Every forecast is frozen before kickoff and settled against the official result. Next: A validated pick additionally needs performance, calibration, a current price, and settlement evidence — bars this version has not met.
Moneyline / spread / total pricesLIVECaptured before every kickoff shown, across up to 9 sportsbooks. These are the books' own numbers with attribution — not GameTimePicks predictions.
Passing yardsNO_MARKETTwo things are missing, and either alone is enough to hold: nobody can say which players will take the field in a preseason game, and the sportsbooks are not offering this market for these games. The projection model is still research — its ranges are not yet reliable enough to publish. Next: Regular-season participation evidence and an offered market.
Rushing yardsNO_MARKETTwo things are missing, and either alone is enough to hold: nobody can say which players will take the field in a preseason game, and the sportsbooks are not offering this market for these games. The projection model itself meets its accuracy bar; only the game-day evidence is missing. Next: Regular-season participation evidence and an offered market.
Receiving yardsNO_MARKETTwo things are missing, and either alone is enough to hold: nobody can say which players will take the field in a preseason game, and the sportsbooks are not offering this market for these games. The projection model is close to its bar and runs privately. Next: Regular-season participation evidence and an offered market.
ReceptionsNO_MARKETTwo things are missing, and either alone is enough to hold: nobody can say which players will take the field in a preseason game, and the sportsbooks are not offering this market for these games. The projection model is close to its bar and runs privately. Next: Regular-season participation evidence and an offered market.
Anytime touchdown · Endzone VaultNO_MARKETThe scoring model is calibrated, but the sportsbooks are not offering touchdown markets for these preseason games, and preseason playing time is unknown. Both must change before anything publishes. Next: An offered anytime-touchdown market plus current role evidence.
SettlementDEPLOYEDteam and scorer results are graded automatically from official finals; each pre-kickoff forecast settles exactly once when its result exists
Paper products · Bank BuilderNO_EVENTSno pre-kickoff NFL event was available to evaluate — this says nothing about the model
Paper products · MoonshotNO_EVENTSno pre-kickoff NFL event was available to evaluate — this says nothing about the model
Paper products · Card builderNO_EVENTSno pre-kickoff NFL event was available to evaluate — this says nothing about the model
Paper products · Endzone VaultPRODUCT_GATEDthe Vault reports NO_VAULT on its own evaluation: no upcoming NFL event in this window to evaluate
Player projections

We built player projections for these games, tested all four families, and rejected every one.

Passing, rushing, receiving and touchdowns — each judged on its own, against bars committed before any of them was built. The thing each model had to beat was deliberately simple: 'what a player of this usage level normally does in a preseason game'. If a player's own history cannot beat that, his history is not adding information.

FamilyGames of evidenceResultWhy
Passing attempts497REJECTEDimproved on the simple baseline by 0.1121 of an attempt where a full attempt was required — and its ranges were far too narrow, containing the real result only 52% of the time instead of 80%.
Rushing attempts1,232REJECTEDimproved by 0.1557 of a carry where 0.75 was required. Its ranges were honest (77% coverage), but the projection itself was not better than the simple version.
Receiving targets2,490REJECTEDvery slightly WORSE than the simple baseline (1.2352 vs 1.2309 targets of average error).
Touchdown scorers1,553REJECTEDscored worse than simply using the league-wide rate (0.1082 vs 0.1055, lower is better).

In preseason, how much a player plays is decided by coaching rotation on the night — not by what he did in past games. Our player histories could not predict it, so we are not publishing per-player projections for this weekend.

Separately, our odds provider offers no NFL player markets for these games at all, so there would have been no sportsbook price to compare against even if a family had passed.

Playing time · 1 games

We do not know who will play in these preseason games — and we would rather size that uncertainty than hide it.

No source we are authorized to use publishes confirmed actives, inactives or preseason playing-time plans for this window. So no player on this slate is listed as expected to start or expected to rotate; every one is marked role-uncertain.

What we do instead

For each listed player we publish a RANGE of how much of his team's work he might take, not a single number. The range is deliberately wide because August starters often play one or two series, and the part of the game nobody on our list accounts for is published too, as unallocated share.

Why it matters

A player projection is only as good as the playing-time assumption behind it. Quoting a regular-season workload in August would make a projection look precise while being wrong for a reason that has nothing to do with the player.

Graded · 2026-08-13

We scored all 6 of our 2026-08-13 NFL forecasts against the official finals.

Every forecast on this slate was a baseline: the model had no measured read separating the two teams, so these numbers grade a league-average preseason prior, not a game-specific prediction. 6 games. A record this small says almost nothing about a model; it is published because grading every forecast is the only way the numbers stay honest, not because it is evidence of skill.

GameWe saidResultMargin off byTotal off byInside our range
DET @ CIN19-1914-1627yes
GB @ PIT19-189-28200no
IND @ NE · tie19-1813-13112yes
ARI @ LV19-1827-14134yes
LAC @ HOU19-1827-7193no
TEN @ SF19-1819-1345yes
Winners called
3 of 5 (1 tie excluded)
Score off by, on average
6.25 points
Result inside our range
margin 67% · total 100% (aiming for 80%)

Our winner probabilities scored WORSE than the sportsbook consensus (0.2391 vs 0.2237, lower is better) across 5 decisive games.

What this model knows

This model does not currently tell these teams apart — and we would rather say so than pretend otherwise.

Preseason games on this slate look similar to each other because, right now, this model genuinely cannot tell them apart. Scoring comes from one league-wide preseason average, and the team-strength input was measured and found to carry no usable signal, so it is switched off. Similar-looking numbers are the honest output of a model that knows very little — not a coincidence, and not a bug.

Who wins, and by how much

We tested whether this model can tell which of two preseason teams is better, and it cannot — the measurement is indistinguishable from no effect at all. So we switched that part off rather than publish a favourite we cannot justify. The small differences you see between games are the simulation's own randomness, not a view on the teams.

How many points are scored

This part does NOT look at the two teams. Every game draws its point total from the same preseason average, so if two games show a similar total that is not a claim about those teams — it is the same starting number in both.

What we found and fixed: An earlier version of this page applied the team-strength input anyway. Because the measured effect pointed slightly the wrong way, it was quietly favouring the WEAKER side in every game. We caught it, tested the input properly, and switched it off.

A per-team scoring adapter built from the preseason corpus, which would have to beat a preregistered bar on a season it had never seen before it is allowed to move a published total.

Products · run 728d1e09ea33 · next 2026-09-07T14:30:00Z

All four NFL lanes ran today

All four NFL lanes ran and none produced a card. That is the result of the checks working, not of the products being switched off: our team model is a labelled baseline, none of our player projections passed their own tests, and the sportsbooks are not offering NFL player markets for these games anyway.

ProductResultCandidates looked atWhy none qualified
Bank BuilderREFUSED0—
MoonshotREFUSED0—
Card builderREFUSED0—
End Zone VaultREFUSED0—

Three separate things would each have to change. the team model is BASELINE_ONLY (the P181 challenger was rejected on bars frozen before it was built); all four player families were rejected on their own frozen bars (P183 Release B); the provider offers no NFL player market in this window (offeredMarkets: []).

Evaluated 2026-09-06T22:43:01Z

Why NFL is not in the paper products

There was no pre-kickoff NFL game to evaluate.

    This evaluation runs on every NFL event window and is recorded, so the answer for any past day stays recoverable. The same gate is enforced in the money path itself — an ineligible sport is refused there too, not merely omitted from the pool.

    Where to go next
    MethodHow this worksWhat the model does — and cannot do.
    RecordTracked resultsEvery forecast graded against the official result.
    CoverageAll sportsWhat is live and what is deliberately not.
    MLBMLB hubThe settled, longer-running side of the site.

    The NFL simulations on this page are published as experimental and are never presented as picks. Player markets and the Endzone Vault stay gated on their own separate evidence — team readiness never auto-publishes a player market. See methodology and all sports coverage.