Pittsburgh Steelers scored 22.4 points a game in 2025, 16th in the league. The model forecasts 22.0 for 2026, 22nd. Every page in this series then asks the same two questions of that forecast — how many fantasy starters does an offense like this actually support, and is the market charging the right price for them — and answers both from the data rather than from the shape of the template. So the question for Pittsburgh is not about the offense at all — the market has the team priced about right. It is about DK Metcalf, where the model and the price come apart.
Who this offense actually is
Over 2021–2025 this offense scored 20.2/18.1/17.8/21.9/22.4 points a game and threw on 62.2/53.3/51.7/49.3/57.9 percent of its snaps. Two separate readings come out of that. Its level: 22.4 in 2025, 2.3 points above its own five-year mean of 20.1. Its trend: a five-year slope of +0.8 points a season, which is genuinely rising. Both point the same way.
Across the same window it has leaned further into the run (-1.3 points of pass rate a season). In 2025 it ran 56.0 snaps from scrimmage a game, 26th in the league. That is play volume, not pace: it counts snaps, and snaps are driven as much by first downs, turnovers and game script as by how fast a team chooses to play. Genuine situation-neutral pace is not in this dataset, so this series does not claim it.
What changed, and why
Running back moves from Kenneth Gainwell to Jaylen Warren and quarterback, receiver and tight end stay where they were. The model's scoring forecast moves -0.4 points a game against 2025, with pass rate +3.4 points and snaps +5.2 a game. The forecast delta is the result of those changes; the job changes above are the cause.
The roster moves behind it: Rico Dowdle comes in from CAR, 216 PPR points last year, Michael Pittman comes in from IND, 202 PPR points last year and Kenneth Gainwell leaves to TB (221 PPR points).
Coordinator, scheme and offensive-line movement are not in this dataset and are not recorded for this team. They are left blank rather than guessed at.
Where the fantasy points go
For 2026 the model grades this offense QB 16th, RB 26th, WR 14th, TE 18th. Its relative strength is receiver — 14th, an above-average environment. Its relative weakness is running back — 26th, a bottom-10 environment. The gap between them — 14th against 26th — is how much the environment should move a close call between two players you rate similarly.
What the market believes, and whether it is wrong
Every roster in the league is priced twice: once at the market's ranks and once at the model's, both converted to expected fantasy points a game through the empirical price curve — what a player at that rank has actually returned, 2016–2025. Ranks are not linear in value, so summing them would flatter deep rosters and undersell top-heavy ones; points are.
No claim is available here. The market has this roster 29th, the model has it 24th — 5 places apart, inside the 8-place bar. Whatever else is true about this offense, “the market hasn't priced it in” is not a sentence this page has earned, so it is not written.
The biggest individual disagreements on this roster: DK Metcalf (+23), Michael Pittman (+9) and Jaylen Warren (+6).
What offenses like this one have produced
Take the 2026 forecast — 22.0 points, 61.3% pass rate, 61.2 snaps, 3.0 red-zone trips — and find the 30 most similar offenses among all 320 team-seasons from 2016 to 2025. Closest matches: Minnesota Vikings 2016 (20.4 PPG, 2 starters); Minnesota Vikings 2018 (22.5 PPG, 3 starters); Minnesota Vikings 2023 (20.2 PPG, 3 starters); Denver Broncos 2018 (20.6 PPG, 2 starters).
Those offenses produced a top-12 quarterback 10% of the time, two top-24 backs 57%, three top-36 receivers 97%, and a top-12 tight end 47%. That is where the starter count below comes from — it is measured off comparable offenses, not asserted.
Read off the 30 most similar offenses in 2016–2025, not asserted. A starter is a QB1, an RB2, a WR3 or a TE1 — top 12, 24, 36 and 12 at the position. Across all 320 team-seasons the average offense produced 2.6.
The player board
DK Metcalf clears the bar as a buy; nothing on this roster clears it as a fade.
| Player | ADP | Market | Model | Edge | PPG | Role | Risk | Call |
|---|---|---|---|---|---|---|---|---|
| Jaylen WarrenRB | 72 | RB29 | RB26 | +6 | 11 | 69 | Medium | FAIR PRICE |
| DK MetcalfWR | 77 | WR36 | WR15 | +23 | 13 | 71 | Low | BUY |
| Rico DowdleRB | 89 | RB33 | RB37 | 0 | 11 | 68 | Low | FAIR PRICE |
| Michael PittmanWR | 95 | WR44 | WR37 | +9 | 11 | 70 | Low | FAIR PRICE |
| Aaron RodgersQB | 176 | QB28 | QB40 | — | 12 | 61 | Medium | THIN MARKET |
| Pat FreiermuthTE | 199 | TE29 | TE69 | — | 5 | 67 | Medium | THIN MARKET |
Market and Model are ranks within the position. Edge is the model's position rank against the market's, after removing the drift that affects every player at that price — so it reads as “the model likes him this many places more than it likes the typical player who costs this much.” A call needs the edge to clear the position's bar (QB 7, RB 13, WR 11, TE 5 places), set at two-thirds of that position's own spread.
The three decisions
What would make this page wrong
- The pass-run balance doesn't move. The forecast has pass rate going +3.4 points to 61.3%. That shift is doing real work in the receiver grade above.
The one thing to watch in-season
Watch snaps from scrimmage per game. The forecast has it at 61.2 against 56.0 last season, and it is the input moving furthest from what this offense actually did. If it holds through the first month, the starter count on this page holds with it. If it snaps back, everything downstream of the forecast — the environment grades, the capacity number, the calls — moves with it.
Sources: Auto Draft Years FF Model workbook — Team Data By Year, Team Stat Ranking, 2026 Team Context; Auto Draft Years 2026 consensus board (16 authoritative market sources); Player outcomes 2016–2025, Pro Football Reference via the project dataset. Team environment grades, forecasts and player model ranks are proprietary. Every figure on this page is generated from the dataset at build time; none is typed in by hand.