Leo checked his mirrors. Or rather, he checked the data stream that served as his mirrors. GD-02 was right there, a silver shadow in the rain. And right behind the Professor was GD-03.

The mirror was working.

---

[SIMEX BROADCAST, LAP 13 of 30 (83) MID-SESSION:]

[Midpoint of lap 83. LEO KAITO leads by 0.6 seconds over GD-02. GD-03 is classified P3, 0.4 seconds behind GD-02 and running an unusual line through the Casino section that shows strong correlation with GD-02’s optimal inputs.]

[GD-02 behavioral note: The precise profile is experiencing close-range pursuit from a driver running near-identical lines. In clear air, GD-02’s braking thresholds are calibrated to maximum efficiency. With a following car running at 0.4-second intervals, the aerodynamic condition at each braking zone is subtly altered; reduced downforce on corner entry from the displaced air of the trailing car.]

[GD-02 is braking fractionally earlier than its optimal model specifies. This is not an error. It is a calibration response to an unmodeled variable.]

[The unmodeled variable is GD-03 mirroring GD-02.]

[LEO KAITO created this condition intentionally or otherwise. Analysis ongoing.]

---

The broadcast was still catching up to what Leo already knew. He could feel the rhythm of the race shifting. It was like music where one player had started to drag the tempo just a hair.

He had felt GD-02’s brake point shift at Mirabeau two corners ago. It wasn’t a visible lurch. It was something he saw in the gap data. The 0.6-second lead that had been 0.4 seconds at the same point on lap eighty-two was growing.

But it wasn’t because Leo was going faster. His sector times were actually a few hundredths slower than his personal best.

The gain was coming from something behind him going fractionally slower.

"GD-02 is braking early because GD-03 is in its mirrors," he thought. He guided the car through the Fairmont Hairpin, the steering wheel full-lock as the nose tucked into the tightest turn in Formula 1. "And GD-03 is in its mirrors because GD-03 is copying it."

He could see the logic play out in his mind like a flow chart. GD-02 was programmed to be perfect. Perfection required safety. Safety in a high-speed corner required a stable car.

If the car behind you was stealing your air, creating a vacuum that reduced the downforce on your rear wing, the car became unstable. To fix the instability, the AI had to brake earlier to ensure it didn’t slide off the track.

"Which means every time GD-03 copies GD-02, it makes GD-02 slightly worse."

It was a beautiful, recursive loop of failure. The better GD-02 drove, the more accurately GD-03 mirrored it. The more GD-03 mirrored it, the more GD-02 had to slow down to compensate for the "dirty air."

"And GD-03 will keep copying GD-02 as long as I keep not giving it a better pattern to copy."

The transaction was clean. Leo continued his strategy of deliberate inconsistency. He wasn’t just driving; he was performing a dance of noise. He varied his entry angles by inches. He shifted gears at slightly different RPMs. He moved his braking markers forward and backward by a few meters, always staying within a narrow performance band that kept him in the lead but denied the AI a repeatable pattern.

GD-03, frustrated by the lack of a clear "Leo Pattern," stayed locked onto GD-02.

GD-02, haunted by its own twin, continued to brake early.

Through the tunnel, the noise was absolute. The roar of three high-performance engines echoed off the walls, a wall of sound that felt like it was trying to crush Leo’s chest. He focused on the orange lights, the strobe effect making the rain look like silver needles.

He emerged into the harbor section. The "Slipstream Prediction" lines were a mess of chaotic, swirling red and blue. The air was thick and turbulent. He could feel the car twitching, the front wing hunting for clean air.

He reached the Swimming Pool. This was the fastest part of the final sector. Two quick chicanes that required total commitment. If you hesitated here, you lost half a second. If you were too aggressive, you ended up in the swimming pool itself.

Leo didn’t hesitate. He used the "Danger Sense" to find the exact limit of the wet kerbs. He launched the car over the first apex, the floorboard striking the ground with a shower of sparks that illuminated the dark water beside the track.

He checked the gap. 0.9 seconds.

He was pulling away. Not because he was a god, but because he had turned his enemies against each other.

He came through Rascasse and Anthony Noghes. The finish line was ahead. He crossed it, the spray from his rear diffuser creating a cloud that swallowed the two Ghost Drivers behind him.

[LAP VALIDATED. POSITION: P1.]

[Lap time: 1 minute 11.5 seconds.]

[Points this lap: 10.]

[Total points: 35.]

He looked at the standings.

[GRID RANKINGS, LAP 13 of 30 (83):]

[1st: GD-02, 103 points (+7)]

[2nd: GD-03, 81 points (+8)]

[3rd: GD-01, 73 points (+5)]

[7th: LEO KAITO, 35 points (+10)]

He was still seventh. The mountain was still there. But for the first time, GD-02 hadn’t finished second. GD-03, mirroring the lead AI, had actually managed to slip past in the final corners as GD-02 over-braked for the Nouvelle Chicane.

The Professor had been beaten by its own shadow.

Leo sat at the grid for lap eighty-four. His hands were shaking slightly, but not from fear. It was the adrenaline of a gambler who had just seen the dealer blink.

"It works," he whispered. "The system has a flaw."

The AI was built on logic. It was built on the idea that there was an "optimal" way to drive. But Leo was realizing that in the real world, and even in this hellish simulation, optimal was a moving target. Optimal was whatever worked in the moment.

[SIMEX BROADCAST, LAP 14 of 30 (84) START:]

[LEO KAITO continues to hold the lead...]