The Philosophy of the Modern F1 Strategy Room / 现代F1策略室的哲学

The fundamental objective of modern Formula 1 strategy is not to win a single race, but to maximize the expected number of championship points accumulated over a 23-race season. This shift from a single-event mindset to a probabilistic, portfolio-theory approach defines the philosophy of the contemporary strategy room. Every decision—from tire compound selection to pit stop timing—is evaluated not by its potential for victory in the moment, but by its contribution to the long-term expected value of the championship tally. This chapter establishes that core principle, frames strategy as a dynamic optimization problem under uncertainty, and introduces the critical transition from driver-centric intuition to data-centric calculation that now governs the sport.

---

1. The Core Objective: Expected Championship Points Over a Single Victory

1.1 The Portfolio Theory of a Formula 1 Season

A Formula 1 season is a series of independent, yet interconnected, probabilistic events. A team’s final championship standing is the sum of its outcomes across all races, weighted by the points system. The current FIA points system awards 25 points for a win, 18 for second, 15 for third, and so on down to 1 point for 10th place, with an additional point for the fastest lap (if the driver finishes in the top ten). This structure creates a non-linear relationship between race result and championship impact.

Consider two hypothetical seasons for a driver:

• **Season A**: Wins 3 races (75 points), but finishes outside the points (0) in the other 20 races. Total: **75 points**.
• **Season B**: Never wins, but finishes 2nd in 3 races (54 points), 3rd in 4 races (60 points), 4th in 5 races (50 points), and scores points in all 23 races. Total: **~350 points**.

Season B’s driver would be a credible championship contender; Season A’s driver would be an also-ran. This illustrates that consistency and point accumulation often outweigh raw race-winning speed. The strategic imperative is therefore to maximize the *expected* points yield from each race, even if that means sacrificing a potential win for a more certain podium or points finish.

The concept of *expected value* ($E[X]$) is central. For any strategic decision (e.g., “Do we pit for soft tires on lap 20?”), the strategist must calculate:

$$E[\text{Points}] = \sum_{i} P(\text{Outcome}_i) \times \text{Points}(\text{Outcome}_i)$$

Where $P(\text{Outcome}_i)$ is the probability of a specific race result (e.g., finishing position) given the decision, and $\text{Points}(\text{Outcome}_i)$ is the points awarded for that result. The optimal decision is the one with the highest expected points value, *not* the highest probability of winning.

Case Study: The 2016 Spanish Grand Prix

At the start of the 2016 race, Nico Rosberg (Mercedes) and Lewis Hamilton (Mercedes) collided, eliminating both cars. For the remaining 18 races of the season, Mercedes’ objective shifted from “win every race” to “maximize the expected points contribution of their single remaining car (driven by either Rosberg or Hamilton, depending on who returned) to secure the constructors’ championship and support the surviving driver’s title bid.” Every strategy call for the rest of the season was filtered through this portfolio lens. For example, at the 2016 Monaco Grand Prix, Mercedes pitted Hamilton early for supersoft tires, not to chase a win (Daniel Ricciardo’s Red Bull was faster), but to secure a safe second place and 18 valuable points, minimizing the risk of a low-scoring finish. The expected value of a conservative second (high probability) outweighed the expected value of an aggressive, low-probability win attempt that could have resulted in a retirement.

1.2 The Multi-Season Horizon: Rookie vs. Veteran Driver Valuation

The expected points calculus extends beyond a single season. A team’s valuation of a driver’s performance incorporates a multi-year horizon. For a rookie driver, a single points finish may be valued more highly than for a veteran, because it contributes to the team’s *long-term* expected constructor points through driver development contracts, sponsor retention, and future performance projections. Conversely, for a championship-contending veteran, a risky strategy that yields a win but carries a 20% chance of a DNF (Did Not Finish) might have a *lower* expected points value than a conservative podium finish, because a DNF could mathematically eliminate the driver from the championship, drastically reducing the *season-long* expected points total.

This is a risk-adjusted expected value calculation:

$$E[\text{Season Points}] = E[\text{Race Points}] \times (1 - \text{Probability of Championship-eliminating Event})$$

The strategy room must therefore model not just the immediate race, but the entire remaining season’s probability tree after each decision.

---

2. Strategy as a Dynamic Optimization Problem with Incomplete Information

2.1 The State Space, Action Space, and Reward Function

Formally, an F1 race strategy can be modeled as a Partially Observable Markov Decision Process (POMDP). The strategist must choose actions (e.g., “box for soft tires on lap 22”) based on an incomplete and noisy observation of the true state of the world.

• **State Space ($S$)**: The complete set of variables defining the race at time $t$. This includes:
• *Track State*: Asphalt temperature, rubbering-in level, wind speed/direction.
• *Car State*: Tire wear and degradation rates (per lap), fuel load (for the race start), brake wear, engine mode.
• *Competitor State*: Positions, predicted tire ages/compounds, fuel loads (estimated), likely strategies.
• *Environmental State*: Weather intensity and trajectory, probability of safety car/virtual safety car (VSC).
• *Regulatory State*: Tire allocation rules, power unit component penalties.

• **Action Space ($A$)**: The set of all feasible decisions at time $t$.
• Primary: Pit stop now? If yes, which tire compound?
• Secondary: Adjust engine mode (e.g., “preserve” vs. “push”)? Adjust brake bias? Defend position aggressively?
• Tertiary: Communicate specific lap time targets to driver.

• **Reward Function ($R$)**: The immediate and future points awarded for transitioning from state $s_t$ to $s_{t+1}$ after action $a_t$. The reward is *delayed* and *probabilistic*. A pit stop on lap 20 yields no immediate points; its value is realized 20 laps later via track position and tire performance. The reward is also stochastic: the same pit stop under identical conditions might yield a podium in one scenario (if rivals pit later) and a 5th place in another (if a safety car disrupts the order).

The optimization goal is to find a policy $\pi$ (a mapping from observed states to actions) that maximizes the *cumulative discounted expected reward* over the remaining race distance:

$$V^\pi(s_t) = E\left[\sum_{k=0}^{T} \gamma^k R(s_{t+k}, a_{t+k}) \mid \pi, s_t\right]$$

Where $\gamma$ is a discount factor (typically close to 1.0 for a 2-hour race, as future laps are nearly as valuable as current laps). The challenge is that $S$ is partially observable. The strategist never knows the true state with certainty.

2.2 Information Asymmetry and Bayesian Updating

The strategy room operates with imperfect information. Key unknowns include: 1. Competitor Intentions: Will Ferrari pit on this lap? Are they on a one-stop or two-stop strategy? 2. Tire Degradation Non-Linearity: How will a tire’s performance drop-off accelerate after a certain wear threshold? 3. Weather Forecast Accuracy: What is the true probability of rain in 10 minutes? 4. Driver Feedback Ambiguity: “The tires are gone” is subjective; how does that map to quantifiable lap time delta?

The strategist uses Bayesian inference to maintain a probability distribution over these hidden states. For example, upon seeing a competitor enter the pit lane, the strategist updates the belief