San Francisco went 12-5 last season. The point differential — +66 on the year — translates to 10.3 structural wins through our fitted model, which means the actual record ran about 1.7 wins ahead of what the scoring margins would have predicted. That gap has a name: luck. Specifically, the 49ers went 5-1 in one-score games, and one-score records are the single most volatile number in football. Teams that go 5-1 in close games one year revert hard toward .500 the next. It doesn't mean they weren't good — they were — but the 12-win headline contains about 1.7 wins the schedule won't hand back.

The honest accounting still leaves you with a legitimately strong team. A 10.3-win structural baseline is real. The offense moved the ball, the defense bent occasionally but rarely broke, and the roster absorbed injuries that would have derailed thinner operations. The close-game luck is the caveat, not the eulogy. But if you're building a 2026 projection and you start from twelve wins, you're starting from the wrong place.

The offense was genuinely good through the air and genuinely mediocre on the ground. Brock Purdy posted a passing EPA of +0.19 per attempt — that's a real number, not a product of scheme flattery, though the scheme helps. The run game came in at -0.04 EPA per carry, which is slightly below replacement. Kyle Shanahan's offense is built on making those two numbers coexist profitably: threaten the run enough that the pass opens, even when the run itself isn't winning plays. It mostly worked. Third- and fourth-down conversion rate came in at 53.2 percent, and red-zone EPA sat at +0.20 per play — both signs of an offense that finished drives rather than just moved chains.

The defense was better against the run (-0.01 EPA per carry allowed) than against the pass (opponents produced +0.13 EPA per attempt against them). That's an inversion of what the front four's reputation would suggest, and it's worth watching. A pass defense that allows nearly a fifth of a win per ten attempts is vulnerable in the playoffs, when opposing coordinators have two weeks to find it.

Shanahan's decision index is where the conversation gets a little uncomfortable. His model-agreement rate of 78.3 percent sounds reasonable until you see the go-rate gap: his actual fourth-down go rate was 20.5 percent against a model-recommended rate of 38.6 percent. That conservatism cost the team an estimated 0.21 expected wins last season — not catastrophic, but the kind of leak that matters in a division where a game or two separates the wild card from the couch. No coaching change to discuss here; Shanahan is back. The chart is simply unfinished business.

The biggest arrival is Mike Evans, who comes over from Tampa Bay on a three-year deal at $14.2 million annually. The market prices him at 1.13 production wins worth of salary. Our model credits him with zero production wins last season — that's the 55th percentile of WR production paired with 95th-percentile pay, at age 34, with an aging multiplier of 0.81. The exit ramp is 2028 at $7.2 million dead. Verdict: overpay. The falsifier is specific — if Evans finishes top-5 in WR production in 2026, this price is fair and that verdict is a miss. He's done it before. He's also 34 with a curve pointing the wrong direction. Dre Greenlaw returns to the Bay after a stint in Denver on a $6 million APY deal. The model doesn't have a production reading for him given his injury history, so that one sits as an open question rather than a verdict.

On the way out: Jauan Jennings to Minnesota ($8 million APY, 0.04 production wins) and Kendrick Bourne to Arizona ($5 million APY, -0.01 production wins). Neither departure stings from a production standpoint — Jennings generated almost no wins above his salary, and Bourne was a slight negative. But the targets they vacated are real, and that's the DFS story rather than the cap story.

Two contracts earned verdicts. Trent Williams gets a WATCH at $25 million APY over two years — 96th-percentile pay, age 38, no aging multiplier available because he's genuinely unprecedented at this level. The exit is 2029 at $8.8 million dead. This one is graded on price discipline alone: the verdict is wrong if the left-tackle market's ceiling moves far enough past this deal by 2027 that it reads as a bargain. That's possible. It's also a 38-year-old tackle. Both things are true. No draft class data was present in the brief.

The production-basis net for the offseason is -0.03 wins. The market-basis net — what San Francisco paid versus what left — is +0.63 wins. That gap tells the story: the front office paid more than it recovered in production, but also replaced players whose production records were thin. Jennings and Bourne were cheap; their replacements cost more. Whether Evans can fill the production gap is the central question, and the aging curve says be skeptical.

The cap posture has genuine quirks worth noting. The biggest spending outlier by position group is tight end, which is overweighted relative to the league (z-score of +1.33). Interior defensive line is the biggest underweight at -1.38 standard deviations below market — notable for a team whose pass defense allowed 0.13 EPA per attempt. Safety is also light at -1.17. These aren't emergencies, but they're places where depth thins fast if a starter misses time. Brandon Aiyuk is the roster's clearest albatross: 4 percent of the cap, 3rd-percentile production, 86th-percentile pay. That combination is hard to build around, and it won't get cheaper until the exit window opens. Future cap proration of $201.8 million is the broader structural ceiling — that's a significant number sitting in the ledger for future seasons.

The bargains provide real cushion. Deebo Samuel is producing at the 68th percentile of wide receivers on 52nd-percentile pay — that's a genuine value on a contending roster. Mac Jones at 57th-percentile production and 24th-percentile pay is as good as backup quarterback value gets. Purdy's 7 percent cap hit is reasonable for a quarterback of his production level, and Nick Bosa at 6.8 percent is the kind of edge-rusher price that looks fair even when it stings.

The arithmetic is clean, if a little deflating. Structural base from 2025 point differential: 10.3 wins. Offseason production wins, net: -0.03. Coach decision delta: zero, because Shanahan is back and the model holds his baseline constant. Structural wins 2026: 10.3. This is an accounting projection from the ledger — not a pick, not a bet, not a promise. It's what the numbers say before a snap is played.

The honest caveat is the one-score record. The structural model already strips most of that luck out by working from point differential rather than wins. The 10.3 figure is the cleaned number. A 12-win repeat would require either another fortunate run in close games or genuine improvement — in the passing defense, in Shanahan's fourth-down aggression, or in Evans outrunning his aging curve. None of those are impossible. None are the base case. The read is wrong if San Francisco wins fewer than nine games; at nine or below, the point differential would need to collapse in a way that suggests real regression, not a one-score hangover. It's also wrong in the other direction if Evans genuinely returns to form and the passing defense tightens — eleven or twelve wins would tell you the underlying quality was higher than the model credits.

Jennings and Bourne combined for 143 targets and 265.4 PPR points last season. That's the opportunity pool now sitting unclaimed. Evans arrived to absorb some of it, but he caught 62 targets for 84.8 PPR points in Tampa last year — 1.37 PPR points per target, which is a useful efficiency baseline, but also a total that falls well short of what Jennings alone produced on 90 looks. The math says the targets are available, and Evans doesn't fill the whole hole.

The sleeper case points toward Deebo Samuel, not Evans. Samuel is producing at the 68th percentile of the position on 52nd-percentile pay, he runs routes from multiple alignments in Shanahan's system, and a suddenly open target share in one of the league's better passing offenses is exactly the kind of structural tailwind that reprices a player mid-season. The efficiency is already there. The opportunity just grew.

On Evans: the aging flag is real and the data supports caution. An 0.81 aging multiplier at age 34 means the model expects meaningful decline from his recent production baseline. He produced 84.8 PPR points last season — a modest total for a receiver priced like a top option. Roster him for the name if you like, but the curve and the number are pointing in the same direction, and it isn't up.

Everything above traces to a live instrument, and the receipts stay public: the GM boards carry Kyle Shanahan's decision card, the team's cap posture ($201.8M already charged to future caps), and every contract verdict with its falsifier and review date — including Trent Williams (WATCH), Mike Evans (OVERPAY). When the season grades these reads, the grades post whether they flatter us or not.