The Ghost Drivers lined up differently on lap eighty-two.

The grid positions were assigned by the previous lap’s finishing order, the same as any race session, and Leo had earned pole. What changed was the behavior in the opening meters. GD-01, the aggressive profile, had taken the outside line at Sainte Dévote in every ranked lap since the grid race began. Lap eighty-two, it took the inside.

GD-03, which had been running the second row and attacking through the Casino section, sat three car lengths off Leo’s rear bumper at the standing start and didn’t close the gap at all through the first straight.

’They adjusted overnight,’ Leo thought, then caught himself; there was no overnight in here. There was no time outside the laps. ’They adjusted between laps. They know what I did on eighty-one.’

The inside line through Sainte Dévote was GD-01 saying: "I will not let you use the outside again."

The three-car-length gap was GD-03 saying: "I am not going to give you data about my counter until I have data about your next input."

They were treating him differently. Finally, they were seeing him as a threat to be managed.

Something settled in Leo’s chest that wasn’t quite satisfaction. Closer to recognition. The particular feeling of a problem being correctly classified.

For eighty-one laps, he had been a bug in the code, a minor variable to be smoothed over. Now, he was the primary target of the simulation’s defensive logic.

He gripped the wheel, the haptic sensors humming against his palms. The rain was a thick curtain, but he didn’t need to see the tarmac to know where the grip lived. He could feel the moisture level through the vibration of the steering column.

The lights on the gantry blinked. One. Two. Three. Four. Five.

The gap between the final light and the start was a microsecond of silence. Leo didn’t wait for the visual cue. He felt the tension in the simulated air break. He dropped the clutch.

The Arcadia car lunged forward. Because he was on pole, he had the advantage of the clean air for exactly forty meters. But GD-01 was charging up his right side, the aggressive AI using a high-torque launch profile that risked wheelspin for pure forward momentum.

It was coming for the inside of the first corner, a door-slamming move designed to force Leo to back off or crash.

Leo drove into Sainte Dévote and took neither the inside nor the outside.

He took the middle.

It was a line that made no sense. It left him vulnerable to an undercut from the inside and a sweep from the outside. But by positioning himself in the dead center of the track, he forced the AI to choose.

GD-01 had to sharpen its angle to stay inside the white line, which killed its exit speed. GD-03, waiting behind, had its vision obscured by the spray of both cars.

Leo braked later than the math allowed. He used the "Perfect Braking" to balance the car on the nose, letting the rear rotate just enough to point toward the exit without losing traction.

He emerged from the corner in P1.

---

[SIMEX BROADCAST, LAP 12 of 30 (82) SECTOR 1:]

[Race start: LEO KAITO from pole position. Notable; the human driver has selected a non-standard entry line at Sainte Dévote, splitting the gap between GD-01’s defensive inside position and the circuit’s conventional racing line.]

[The entry angle is suboptimal by conventional metrics. Exit speed, however, is 4 kilometers per hour above the session average for this corner.]

[GD-03 adaptive profile: no counter-measure deployed. Insufficient data on new input pattern.]

[GD-01: inside line held. Position: P3 at Massenet.]

[LEO KAITO: P1 at Massenet. Gap to GD-02: 0.8 seconds.]

---

0.8 seconds at Massenet.

In the previous ten ranked laps, the closest Leo had been to GD-02 at this point in any lap was 2.1 seconds. He read the gap notification through the corner without reacting to it, the number filed as information rather than celebration, and moved to the Casino section.

The air was heavy and wet. The "Slipstream Prediction" lines were glowing a faint, electric blue in the gray mist. He could see the wake left by the cars ahead, or in this case, the void he was creating for the cars behind. He was the one leading the pack through the uphill climb.

The Casino section was where GD-02 was fastest. The three medium-speed corners through the uphill section; Massenet, Casino Square, Mirabeau, were the precise profile’s strongest sequence, the lines optimized to a level that Leo hadn’t been able to match in any of the laps where he had been close enough to observe them.

The broadcast had noted it across multiple lap reviews. "GD-02’s Casino sector represents the benchmark for this session."

He wasn’t going to match GD-02 through Casino by running a better version of GD-02’s line. That was a fool’s errand. You didn’t beat a machine by trying to be a more precise machine. You beat it by changing the nature of the game.

Through Massenet he let the car run slightly wider on exit than the optimal, a fraction, the kind of margin that cost three hundredths of a second but loaded the left-rear tire differently for the uphill compression into Casino Square. He felt the weight shift, the suspension compressing into the tarmac.

Through Casino Square he used that extra load, the rear planted in a way that didn’t have a name in his technical vocabulary but expressed itself as a feeling of authority through the steering column, and carried speed onto the exit that the optimal line couldn’t access because the optimal line hadn’t built the load it required.

He wasn’t running GD-02’s line. He was running a line that GD-02’s model had no dataset for. The AI was programmed to find the shortest path with the highest grip. Leo was finding a longer path that generated higher momentum.

[SIMEX BROADCAST, LAP 12 of 30 (82) SECTOR 2:]

[SECTOR 2: LEO KAITO, 22.4 seconds. Session fastest by 0.5 seconds.]

[Notable: The human driver’s Casino section has recorded the fastest sector two time of the entire rankings simulation. The line employed does not correspond to any established optimal in the session dataset. Tire load telemetry indicates an unconventional load-transfer sequence through Massenet exit that appears to have enabled earlier throttle application through Casino Square.]

[GD-02 sector two: 22.9 seconds. GD-02 is no longer the session benchmark in this sector.]

[Gap at tunnel entry: LEO KAITO leads by 0.3 seconds.]