The Adaptation Curve: How Long It Actually Takes a Driver to Settle In

The question gets asked every weekend. “Is he settled in yet?” It is asked of Hamilton at Ferrari. It was asked of Russell at Mercedes in 2022, of Sainz at Williams in 2025, of Tsunoda at Red Bull in 2025, of Ricciardo at McLaren in 2021. It will be asked, repeatedly, of whichever rookie or mid-career-switch driver is the new variable at the start of every future Formula 1 season.

The question is rarely answered with a number. The answer is some version of “still learning the car,” “starting to look more comfortable,” “another six races and we’ll know.” The honest version is “I don’t know,” because nobody seems to have measured it. We did.

The premise

What does “settled in” actually mean? You’d want it to be measurable as the gap between where a driver is right now and where they’re going to end up. A driver who is settled is at their steady-state pace. A driver still adapting is some distance behind where they will eventually be, and that distance is closing race by race.

This is operationalisable. For every driver-season-team combination since 2019, we have per-race median race-pace from FastF1’s lap data. We have their teammate’s per-race median pace, in the same car, on the same day. We can compute the pace gap to teammate for every race in their first season at a new team. We can normalise the trajectory to their final-three-races steady state. And we can stack the trajectories of thirty driver-team adaptations on top of each other to see what the typical curve looks like.

The result is messier than the rhetoric suggests, and more useful.

Thirty switches, seven seasons

Since 2019, thirty drivers have spent a first season at a new team that had FastF1 coverage. Some are headline switches: Hamilton to Ferrari, Sainz to Williams, Russell to Mercedes, Ricciardo to McLaren. Some are obscure: Albon to Williams (loan), Hülkenberg’s multiple reactivations, Zhou’s Sauber to Sauber-rebranded-as-Kick switch.

For each case, we computed the season-end teammate pace delta: where the driver finished the year relative to their teammate in the same car, measured as median race-lap-time deltas across the final three races of the season.

This is the headline distribution. Three things stand out:

Steady-state outcomes vary enormously. Some drivers, given a full season at a new team, end up significantly faster than their teammate by race-pace. Sainz at Williams ended 2025 nearly a second per lap quicker than Albon in race trim, a remarkable result given Albon had been Williams’s clear pace leader through 2022-2024. Ricciardo at Renault in 2019 ended -0.66s vs Hülkenberg. Hülkenberg at Haas in 2023 ended -0.38s vs Magnussen.

At the other end: Tsunoda’s 2025 Red Bull was the worst-trending adaptation in our entire dataset. He ended the season 1.68 seconds per lap off Verstappen in race trim. Not slower in any normal sense, slower in a way that no other driver in a year-old top car has been. The seat was lost mid-season. The data agrees with the decision.

Hamilton at Ferrari sits second-worst. His 2025 steady-state Δ vs Leclerc was +0.70 seconds per lap. Given Hamilton’s career, given his pre-2025 dominance over comparable teammates, and given the standard “give him a year” expectation, that gap is alarming.

Russell’s 2022 Mercedes move is conspicuously absent from the worst-case list. He ended 2022 at +0.14s vs Hamilton, which counts as settled-in, not behind. The narrative of the time treated him as the heir-apparent; the data treated him as comparable from the start.

Some adaptations go the wrong way. Ricciardo at McLaren 2021 ended the season +0.44 seconds behind Norris despite winning a race. The pace gap was not the win gap; race wins are dominated by tyre management and strategy, but the lap-time data is unforgiving. Ricciardo’s struggle at McLaren wasn’t a story the broadcast invented; it was the data confirming what felt like an off year.

What does the typical curve look like?

If we normalise each adaptation case to its season-end steady state, so race 1’s value is “how far from final pace was this driver at the start of the season”, and aggregate across all thirty cases, we get the canonical adaptation curve.

The result, frankly, is flatter and noisier than commentary implies. The median first-three-races extra-gap is +0.09, +0.07, +0.12 seconds, a tenth of a second worse than steady state. By race 5 to 10 the median oscillates between -0.15 and +0.20s. Race 20 onwards the median converges on the steady state by construction.

What this says: the typical driver is within a tenth of their final pace from race 1. They are not “spending the first eight races learning the car”; they are mostly at speed almost immediately. The cases where the curve is steep, drivers who are dramatically off in races 1-3 and converge by race 10, exist, but they are the exception. The typical adaptation is fast.

What is slow is the outlier band. The 75th-percentile trajectory shows extra gaps of +0.5 to +0.8 seconds in the first ten races, often slowly closing. The drivers who struggle at a new team show up here. The drivers who don’t, don’t.

This is, in some ways, the most important finding from the dataset: adaptation is bimodal. Drivers either click with the car quickly (most of them, within a couple of races) or they don’t (the slow-adapters, who may stay a quarter-second or more off teammate pace for half a season or longer). There is no smooth “race by race they get a bit closer” curve. There is a population that is close from the start and a population that isn’t.

The 2026 trajectories so far

Through Round 4, every driver on the 2026 grid is in some sense “adapting”; the regulation reset means everyone is in a new car, even drivers who stayed at the same team. The interesting question is how each driver’s per-race delta to teammate is moving.

A few patterns:

Antonelli at Mercedes: pure dominance trajectory. Δ vs Russell: R1 −0.07, R2 −0.04, R3 −0.61, R4 −0.88. Not just settled, actively pulling away. Antonelli is in his second season at Mercedes; we’re past adaptation here, into “outpacing the established teammate” territory. By the canonical curve’s standards he is running well ahead of any historical analog.

Hamilton at Ferrari: slow improvement, still adapting. Δ vs Leclerc: R1 +0.15, R2 +0.04, R3 +0.84, R4 +0.66. Average through four rounds: +0.42 seconds per lap behind Leclerc. That is slower than his 2025 season-end Δ of +0.70, so there is improvement, but it is gentle improvement, and the absolute gap is still in territory most drivers have closed by their second season. Hamilton is the canonical “still adapting” case in the 2026 dataset, and the data agrees with the broadcast framing for once.

Bearman at Haas: fast then volatile. Δ vs Ocon (until R2) then vs Hülkenberg (R3+): R1 −0.62, R2 −0.86 (faster than teammate in his rookie season’s start), R3 +1.76 (huge gap, likely a single bad weekend), R4 −0.38. The pattern is rookie volatility, not adaptation deficit. Bearman is capable of out-pacing his teammate, but he isn’t doing it consistently. Watch for that to settle by Round 8 or 9 if his career follows the canonical path.

Sainz at Williams (still): running ahead of teammate. Δ vs Albon: R1 +0.32, R3 −0.64, R4 −0.13. Two out of three races faster, one slower. This is consistent with the 2025 steady-state pattern where Sainz emerged dramatically faster than Albon over the year. He looks settled.

Hadjar at Red Bull: small sample, struggling. Δ vs Verstappen, R3 onwards: R3 +1.09, R4 (TBD/no data). The early signal is a one-data-point reading of struggle. Tsunoda’s 2025 ended at +1.68s vs Verstappen. Hadjar is on a similar trajectory through one race. The story to watch is whether Hadjar’s gap closes through rounds 5-10 or whether the Red Bull becomes the second-driver-killer it became in 2025.

Five drivers running ahead of their canonical curve

Drawing the 2026 per-race trajectories against the canonical curve gives us a way to flag who is adapting faster than the historical norm:

1. Antonelli. At -0.4 to -0.9 vs teammate, deep into “outpacing” territory. The canonical curve’s 25th percentile at this stage is around -0.3. He’s beyond it.

2. Sainz. Averaging -0.15 vs Albon. Within the canonical band but trending below the median.

3. Russell. Averaging +0.4 vs Antonelli (the other side of the same matchup). Not behind the curve, but the curve isn’t designed for “you’re at your old team but your new teammate is faster.”

4. Lawson at Racing Bulls. Δ vs Lindblad: R1 +0.31, R2 −0.55, R3 −0.47. Two out of three races faster than the rookie. Comfortable, ahead of canonical.

5. Lindblad. The inverse of Lawson, but for a rookie, a -0.5s gap to a sophomore in his first three races is good adaptation. The canonical curve for rookies (a sub-cohort with bigger gaps in the early races and slower convergence) would put a rookie at +0.5 in R3 typically. Lindblad is at +0.5 average across three. Roughly on-curve.

Three drivers running behind it

1. Hamilton. Δ +0.42 average vs Leclerc through four rounds. The historical adaptation curve median converges on steady state by race 6-8. Hamilton’s steady state from 2025 was already +0.70, and 2026 is showing +0.42. So he is improving on his 2025 baseline but not yet at the level the canonical curve predicts for a season-two adaptation. By the typical curve, a second-year-at-new-team driver should be at their teammate’s pace or within ±0.1s. Hamilton is three to four times that gap.

2. Hadjar. One race in, +1.09s vs Verstappen. Canonical curve median in R3 is +0.12. Hadjar is approximately ten times the typical gap. Sample is single-race so caveats apply, but the early reading is concerning.

3. Colapinto and Gasly at Alpine. Running noisily but with Colapinto consistently behind the senior driver. Colapinto’s R1 to R3 vs Gasly: +0.43, +0.81, +1.09. That’s not converging, it’s diverging. Adaptation curves do not look like this in healthy cases.

What this changes for the championship

The Hamilton-at-Ferrari trajectory is the most consequential piece of data in the table. Ferrari is currently P2 in the WCC, 70 points behind Mercedes. They need both drivers performing for any constructors’ title scenario. If Hamilton converges on Leclerc’s pace by the summer break, Ferrari’s championship odds shift materially. If he doesn’t, if the 2025 trajectory continues through 2026, Ferrari is essentially racing with one car on full pace and the other on adaptation tax.

The historical comparison cases for a star driver taking longer than typical to settle:

• Alonso at McLaren-Honda 2015. Took two seasons to be consistently faster than Button.
• Schumacher at Mercedes 2010. Never reached Rosberg’s pace; retired after three years.
• Räikkönen at Lotus 2012. Adapted within four races, won race five.

The Schumacher analog is the one Ferrari fans don’t want to hear. Schumacher came out of a partial retirement at age 41 to a regulation-reset car and never produced the steady-state pace of his early-2000s ceiling. Hamilton in 2025-2026 is 40-41 years old, in a regulation-reset car, and the data shows a slow trajectory of incremental improvement. The trajectory might break upward, Räikkönen-style, or it might not. The next eight rounds will tell us which.

For the WDC: Antonelli’s +20 point lead is built on the most-dominant teammate trajectory in our 2026 dataset. He is faster than Russell in three of the last four races by a margin of more than a quarter-second per lap. Unless something fundamental changes, his lead is going to grow at every race where the cars run clean. The chaos cluster from Canada onwards is the only thing that’s likely to compress it.

Caveats

The canonical curve is noisy. The median trajectory is much closer to flat than commentary would predict, but the variance is high; the 25th-75th percentile band spans about a second per lap. So “canonical curve” is more honest as “median trajectory, take it with the variance bars on.”

Adaptation versus regulation reset are different cases. Our 2018-2025 cases are drivers switching teams in a stable rule set. 2026 is the whole grid in a new rule set. We’re extrapolating somewhat. The closest analog is Russell 2022 (W13 was substantially different from W12), whose trajectory was fast convergence within 4-5 races. If the regulation-reset adaptation is similar, most of the grid will be at steady state by Round 8. The drivers who aren’t will be the structurally slow adapters.

Sample size per case. Some 2026 cases (Bortoleto, Hülkenberg) have only one or two rounds of data. Their trajectories are placeholder readings, not reliable signals yet. We’ve called them out where they appear in the chart but the round-by-round numbers should be treated as starting positions, not verdicts.

Within-team teammate strength. A driver whose teammate is exceptional (Antonelli currently, Hamilton circa 2018, Verstappen always) has a tougher adaptation curve than a driver whose teammate is average. The canonical curve doesn’t adjust for teammate strength. So Hamilton’s +0.70s 2025 steady state at Ferrari is at least partly a function of Leclerc being a top-five driver, not just a function of Hamilton being slow.

A future version of this analysis would adjust for teammate strength using the engine’s existing Elo ratings. That would shrink some of the alarming gaps (Hamilton vs Leclerc looks less concerning when you adjust for Leclerc being a 1900-rated driver) and inflate others (Tsunoda’s gap to Verstappen looks even worse adjusted, because Verstappen’s rating is near the all-time peak). The broad rankings would not change.

What this piece is for

Adaptation is the most-discussed and least-measured question in F1. Every season opener generates a wave of “is X adapting?” content. Almost none of it is anchored in data. This piece is the data, updated after each race weekend, available to point at when someone asks how Hamilton’s Ferrari move is really going or whether Antonelli’s lead is real.

The next time a broadcaster says “another five races and we’ll know”, you can look at the chart. Sometimes they’ll be right; the trajectory will close. Sometimes the curve has already flatlined and another five races won’t change it. The dataset is the only honest answer to the question, and now there is one.