Road test

Freedom Roleplay runs over a hundred custom cars from bought packs, and every vendor ships its own handling. Some were seconds off their real 0-100, some slid around every corner, and police reported that starter cars outran every patrol car. Fixing that by hand means hours of test driving per car. I wanted a test that drives itself.

Every custom vehicle, more than 130 by now, is tied to a real make, generation and trim, with a source for its top speed and 0-100 time; unknown figures are marked as estimates. Road holding comes from its class: 11 classes, from super cars to trucks, with cornering grip bands taken from published skidpad tests. Off-road behaviour comes from a terrain class per model. Every one of the 1,014 models in the catalogue has an explicit decision, and the game's own surface material under the wheels decides how much grip a road car loses on dirt.

A planner on the server hands one step at a time to a staff client. That client spawns the car on a private test track and drives scripted runs while it streams telemetry back every 50 milliseconds:

• Speed: sprints to a stable top speed, judged against the real 0-100 and top speed within 0.2 seconds and 2 km/h.
• Handling: keyboard-style manoeuvres such as full lock under throttle, braking in a corner at 100 km/h and quick steering swaps, scored on cornering grip, understeer and peak slip. A spin always fails.
• Tuning: between runs the server adjusts power, gearing, drag, grip and chassis levers, each within fixed bounds, until the car passes or is flagged for review.

Before it judges anything, the test checks itself. A reference car is measured first, an old setup that is known to slide must fail and the current one must pass. If that calibration does not hold, no car is judged. Progress is stored on the server, so a test can be paused, resumed or picked up by another tester without losing a run.

The first version stepped power by fixed percentages and needed several sprints per car. A rework fitted a power-law model on 4,434 stored sprints from 125 models; after one fitted sprint, 76 percent of missed 0-100 targets landed within 0.2 seconds. Top speed is now corrected in one physical step, because it scales with the square root of power over drag.

On 3 October a whole morning of testing finished eight cars, only five of them passed, so I went through the logs. 38 percent of the time went to sprints for estimated in-between speeds that no real figure backs up, plus braking runs nothing used. Those were cut. The first run after the change did eleven cars in 64 minutes, about six minutes per car.

The measured figures are what the dealer, the garage and the tuning menu show, side by side for standard and fully tuned, with "not reachable" where a car cannot reach a speed band. Results reach the live server as data releases, so players drive what was measured.

Players can order an imported car through the in-game store. The bought vehicle packs sit on the server as dormant stock: about 825 models in 18 packs that stay out of the catalogue, the test queue and storage until an import is activated. Activation writes the profile, terrain class, storage and value, and refuses known bad vendor setups, such as a missing gravity flag that costs about a fifth of grip and braking, unless a written reason is given. Only an import with a passed road test can be set ready to claim. The buyer then gets a popup in the store and claims the car with a free plate of choice. The price class follows the real car's top speed, under 280, 280 to 330, or above 330 km/h, with the odd exception where a car is clearly better than its class.