Progressive road load — source-level physics integration

Real Unreal chassis source changed, but NOT engine/runtime-qualified. This iteration targets the rigid-body Unreal Engine 5.8 car driven by the player and the single AI rival, not the separate native 2D Win32 experimental preview.

Handling problem and change

Previously the prototype configured the simulated chassis with SetLinearDamping(0.5f). That is a speed-proportional constant resistance applied by Chaos regardless of current road speed. In a force-driven arcade prototype it can dominate propulsion and make vehicle speed/coasting feel unconvincing. It is not an aerodynamic model, tire friction or a complete realistic vehicle system.

This iteration lowers the constant damping to 0.08 and adds a readable, speed-dependent, planar force opposing actual chassis velocity during a qualified racing tick:

• A simple quadratic drag term grows rapidly near racing speeds, rather than a constant percentage drag rate throughout.
• A bounded smooth rolling-resistance ramp vanishes near rest; the car is not pulled backward by an invented constant friction force.
• Existing road-surface observation increases resistance 45% after confirmed off-pavement travel.
• Arcade and Simulation Inspired use different force coefficients. The real player and real rival share the same chassis implementation.
• The force is proportional to the actual measured mass, uses the actual world-space XY velocity (including reverse and side-slide) and never modifies Z velocity, gravity, wheel contact, recovery coordinates or race clock.
• Before any force application the policy validates frame duration, finite XY velocities, finite realistic mass and bounded motion. It limits its own deceleration to no more than 80% of the sampled planar speed in one frame. Failed samples neutralize the physics driver's current frame rather than requesting a nonfinite force.

This is gameplay-tuned simplified drag, not validated aerodynamics, vehicle dynamics, tire contact, real-world braking performance or an official speed benchmark. The existing reverse limit and safety top-speed cap are deliberately retained.

Portable tests and source binding

Tests/Standalone/ProgressiveRoadLoadTests.cpp exercises zero speed, coasting, three levels of forward drag, backwards movement, sideways drift, diagonal force direction, off-pavement resistance, the two handling profiles, near-zero continuity, long frames, nonfinite speed, invalid mass and one-frame reversal limits.

Tests/Editor/UnrealProgressiveRoadLoadBindingTests.py verifies the actual Unreal pawn consumes that policy after its real circuit / Chaos readiness checks and before accelerating, that load is a separate XY AddForce and that it cannot write results/rewards. The C++ test is compiled with strict warnings on three OSes in CI. Neither test runs Unreal Engine or measures real gameplay.

Real playable acceptance gate

1. Compile in genuine Unreal Engine 5.8, package cooked Win64 and confirm the physics component applies the force every active frame.
2. On an installed build drive the six selectable cars in both handling modes. Measure 0–60 km/h, coast-down from 100 km/h, 60→0 braking, and steady-state top speed in straight-line travel.
3. Compare against the previous driving build with identical hardware/physics timestep; tune drag/rolling constants to avoid stalling before intended car top speeds or making cars too fast.
4. Validate driving clockwise through all original four gates, with handbrake slides, reverse, nitro and rival AI. Check no Z interference on slopes, bumps or Chaos contacts.
5. Test off-pavement transitions and recovery; confirm no abnormal speed reversal, acceleration on coasting, crashes or NaN force.
6. Verify 30/60/120 FPS behavior and hitch/focus/pausing safety, log performance; test an installed Windows build, not source only.

The game still needs genuine driveable build evidence, polished art, real wheel/suspension simulation, more tracks and professional UX. See playable release gate #148.