Today· Sep 17Pregame slateSettled· Sep 16

Regular season · week 2 · Sunday, Sep 20, 4:25 PM ET

Miami DolphinsatSan Francisco 49ersEXPERIMENTAL

Levi's Stadium

Miami DolphinsSan Francisco 49ersSaved on this device

72°F · 5 mph wind from the WNW · 0% chance of rain · mostly sunny — not used by the model; no weather term enters the numbers below.

Simulation story

5 chapters · read from the published artifact

Skip to the report ↓

› Loading the matchup model› Reading the outcome distribution from 10,000 simulated games› Preparing the score scenariosPress Play to step through what the model expects, chapter by chapter, or read the report below.

MIA @ SF

Miami Dolphins at San Francisco 49ers at Levi's Stadium.

Simulation

What our model expects

10,000 simulated games · model nfl-regular-season-public-v1

Projected score

MIA 19 — 27 SF

the middle of our simulated outcomes, not a call on the exact final

Win chance

MIA 23.1% · SF 74.4%

how often each side won in 10,000 simulations · ties 2.5%

Total points (both teams)

8 in 10 simulations landed between 29 and 63

SF winning margin

a minus means MIA wins by that much · 8 in 10 between -10 and 23

Model details

From the margin-of-victory Elo win head, published exactly as evaluated on held-out 2006–2021 — no shrink toward 50% is applied, and no claim to out-predict the sportsbook market is made. Generated 2026-09-17T14:17:00Z under model nfl-regular-season-public-v1 and frozen pre-kickoff; every forecast is settled against the official result.

Projected scorecard

expected statistical summaries · not one simulated game

MIA 19 — 27 SFwin chance MIA 23.1% · SF 74.4%total 46 (29–63)margin +7

MIA

Likely TD scorers

De'Von Achane 50.9% · available role uncertain

Malik Willis 31.7% · available role uncertain

Caleb Douglas 28.6% · available role uncertain

Passing

Malik Willis 239 pass yds (129–399) · available role uncertain

Rushing leaders

De'Von Achane 61 rush yds (17–154) · available role uncertain

Malik Willis 30 rush yds (5–97) · available role uncertain

New arrivals · not in these numbers

Per game at their previous club — history, not a projection.

Brady Cook 121.8 pass yds/g at NYJ

Receiving leaders

Caleb Douglas 5 rec · 61 yds · available role uncertain

De'Von Achane 4 rec · 33 yds · available role uncertain

Malik Washington 4 rec · 38 yds · available role uncertain

Likely TD scorers

Christian McCaffrey 52.5% · available role uncertain

Deebo Samuel Sr. 35.0% · available role uncertain

Mike Evans 34.5% · available role uncertain

Passing

Brock Purdy 217 pass yds (121–354) · available role uncertain

Rushing leaders

Christian McCaffrey 47 rush yds (12–123) · available role uncertain

Kaelon Black 45 rush yds (11–119) · available role uncertain

New arrivals · not in these numbers

Per game at their previous club — history, not a projection.

Mike Evans 3.8 rec · 46 yds/g at TB

Deebo Samuel Sr. 4.2 rec · 42.8 yds/g at WSH

Christian Kirk 2.9 rec · 29.7 yds/g at HOU

Receiving leaders

Christian McCaffrey 5 rec · 44 yds · available role uncertain

Deebo Samuel Sr. 4 rec · 44 yds · available role uncertain

Mike Evans 4 rec · 42 yds · available role uncertain

pass interceptions · First/last/2+ touchdown — withheld, with the exact bar each failed

pass interceptions: RESEARCH_ONLY
First/last/2+ touchdown: DISABLED — no ordering model and no calibration receipt of their own; never derived from anytime probabilities

Range

How wide the outcomes are

the 10th to 90th percentile of each team's simulated score

Team	Low (10th)	Projected	High (90th)
Miami Dolphins	8	19	32
San Francisco 49ers	14	27	38

Exact score

The likeliest final scores

from an event-based simulation of touchdowns and field goals, solved onto the median margin and total above

Score (MIA – SF)	Chance
17–20	0.80%
24–27	0.77%
17–27	0.74%
20–17	0.64%
20–23	0.60%
27–24	0.56%

No single scoreline is likely — these are the most common of hundreds. The projected score above is derived from the medians and answers a different question: the middle of the distribution, not its most common point.

Key numbers

How often it lands on 3, 7, 10 or 14

football's margins pile up on those four — this is the one thing a normal-draw model cannot show you

Margin	This game	Last 3 regular seasons
Decided by exactly 3	9.7%	14.2%
Decided by exactly 7	6.0%	8.3%
Decided by exactly 10	5.6%	4.2%
Decided by exactly 14	4.6%	4.4%
Any of the four	25.9%	31.1%

The right-hand column is what actually happened across 816 regular-season games (2023, 2024, 2025) — the same finals this engine was fitted to. It gets the shape right, with 3 the most common margin by a distance, and it is not calibrated to the number: it puts less weight on 3 and more on 10 than those seasons did.

MIA touchdowns

2.3

SF touchdowns

2.9

MIA field goals

1.5

SF field goals

2.0

Tied after regulation

3.8%

Comparison

Us versus the sportsbooks

two independent reads, shown side by side

No current sportsbook capture covers this game.

Player projections · 20 modelled

The player board

Every modelled player, with his availability. Players listed out carry no volume projection.

Every receiving family for one player on one row, ranked by expected receptions.

Player	Team	Availability	Receptions (10th–90th)	Rec yards (10th–90th)	TD chance
Christian McCaffrey	SF	role uncertain	5 (2–10)	44 (17–91)	52.5%
Caleb Douglas	MIA	role uncertain	5 (2–10)	61 (24–124)	28.6%
De'Von Achane	MIA	role uncertain	4 (1–8)	33 (10–75)	50.9%
Deebo Samuel Sr.	SF	role uncertain	4 (1–9)	44 (16–94)	35.0%
Mike Evans	SF	role uncertain	4 (1–8)	42 (14–91)	34.5%
Brandon Aiyuk	SF	role uncertain	4 (1–9)	48 (17–101)	31.8%
George Kittle	SF	questionable	4 (1–8)	38 (12–84)	28.5%
Malik Washington	MIA	role uncertain	4 (1–8)	38 (12–86)	25.3%
Ricky Pearsall	SF	role uncertain	3 (1–7)	35 (9–82)	23.5%
Demarcus Robinson	SF	role uncertain	2 (0–4)	19 (0–57)	17.8%
Greg Dulcich	MIA	role uncertain	2 (0–5)	23 (2–61)	14.7%
Chris Bell	MIA	role uncertain	1 (0–3)	13 (0–44)	9.8%
Kevin Coleman Jr.	MIA	role uncertain	1 (0–3)	11 (0–40)	9.8%

New arrivals · not in the simulated numbers above

Per game at their previous team — history, not a projection.

Player	Team	Availability	Last club	Per game (prior club)
Mike Evans WR	SF	active uncertain	TB · 8g	3.8 rec · 46 yds
Deebo Samuel Sr. WR	SF	active uncertain	WSH · 17g	4.2 rec · 42.8 yds · 5.5 rush yds
Christian Kirk WR	SF	active uncertain	HOU · 14g	2.9 rec · 29.7 yds
Brady Cook QB	MIA	active uncertain	NYJ · 8g	7.1 rush yds · 121.8 pass yds

Families not shown, and the exact bar each failed (2)

pass interceptions: RESEARCH_ONLY
First/last/2+ touchdown: DISABLED — no ordering model and no calibration receipt of their own; never derived from anytime probabilities

Scoring outlooktop 6 of 20 by TD chance · full list in the board’s TD tab

52.5% Christian McCaffrey · SF · available role uncertain
50.9% De'Von Achane · MIA · available role uncertain
35.0% Deebo Samuel Sr. · SF · available role uncertain
34.5% Mike Evans · SF · available role uncertain
31.8% Brandon Aiyuk · SF · available role uncertain
31.7% Malik Willis · MIA · available role uncertain

Anytime-scorer probability means scoring a touchdown — never throwing one. Void if the player does not play; questionable players carry their state above.

Reading key

What these numbers mean

Projected score: The middle outcome across every simulated game — not a prediction of the exact final.
Win chance: How often each side won across the simulations, after the calibration described above.
80% range: Eight in ten simulated games landed inside this band. Real games land outside it too.
pp (percentage points): The plain difference between two percentages. A gap is a difference, not an advantage.
Experimental: Published while its out-of-sample record is still accumulating. Every forecast is frozen before kickoff and graded against the official result.

Provenance

Where this came from

model: nfl-regular-season-public-v1 v2
simulations: 10,000
input hash: 6df6395781e7bedf
generated: 2026-09-17T14:17:00Z
kickoff: 2026-09-20T20:25Z
state: UPCOMING

Player-family model provenance (4 published)

Rushing yards: Share-level model: cleared every bar when re-tested on 2014–2021 (14,502 player-games) — a second look at seasons examined once before, not a blind test. Each week's forecast is frozen before the first kickoff and graded; its blind 2026 record is still accumulating (0 graded player-games so far).
Receiving yards: Share-level model: cleared every bar when re-tested on 2014–2021 (27,134 player-games) — a second look at seasons examined once before, not a blind test. Each week's forecast is frozen before the first kickoff and graded; its blind 2026 record is still accumulating (0 graded player-games so far).
Receptions: Share-level model: cleared every bar when re-tested on 2014–2021 (27,134 player-games) — a second look at seasons examined once before, not a blind test. Each week's forecast is frozen before the first kickoff and graded; its blind 2026 record is still accumulating (0 graded player-games so far).
Anytime touchdown: Opportunity touchdown model: cleared every bar on a blind test of 2014–2021 (35,128 player-games never used to build it). Probabilities condition on playing — a player who does not play settles void. Each week's forecast is frozen before the first kickoff and graded; its blind 2026 record is still accumulating (0 graded player-games so far).

Our win chances come from a team rating that weights how decisively teams win. Tested on 4,281 past games it had never seen (2006–2021), it scored 0.629 on log loss (lower is better) — better than our previous rating's 0.642 and a coin flip's 0.693, but short of the sportsbooks' own odds at 0.610. It has NOT been shown to beat the sportsbook market. Where our win and margin ratings disagree on the favourite, the previous rating is used for that game. The 2026 season evaluation still runs under bars frozen in August, so every forecast stays experimental.

Educational and paper-only — not betting advice.