The logic for placing particles on the right side of the screen while avoiding the form area is quite complex, especially the fallback mechanism after 30 attempts. This can be simplified for better readability and maintainability.

Instead of a limited number of attempts and a complex fallback, you can use a do-while loop that continues until a valid position is found. Given that the form doesn't cover the entire area, the risk of an infinite loop is negligible, and the code becomes much cleaner.

    let x: number;
    let y: number;

    do {
      x = canvasWidth / 2 + 20 + Math.random() * (canvasWidth / 2 - 20);
      y = Math.random() * canvasHeight;
    } while (isInFormArea(x, y, bounds));

To improve performance, you can avoid iterating over the particles array twice when rendering. You can do this in a single pass by using Path2D objects to build the paths for left and right particles separately, and then drawing them. This is more efficient, especially as the number of particles grows.

  const leftPath = new Path2D();
  const rightPath = new Path2D();
  let hasLeftParticles = false;
  let hasRightParticles = false;

  for (const dot of particles) {
    if (!isInFormArea(dot.x, dot.y, bounds)) {
      if (dot.x < canvasWidth / 2) {
        leftPath.rect(dot.x - 1.5, dot.y - 1.5, 3, 3);
        hasLeftParticles = true;
      } else {
        rightPath.rect(dot.x - 1.5, dot.y - 1.5, 3, 3);
        hasRightParticles = true;
      }
    }
  }

  if (hasLeftParticles) {
    ctx.fillStyle = LEFT_PARTICLE_FILL_STYLE;
    ctx.fill(leftPath);
  }

  if (hasRightParticles) {
    ctx.fillStyle = RIGHT_PARTICLE_FILL_STYLE;
    ctx.fill(rightPath);
  }

processAnimationFrame always calls renderFrame(), so the canvas still re-renders on every requestAnimationFrame tick (e.g., 240Hz displays will still render ~240 times/sec). This keeps GPU usage tied to refresh rate and likely undermines the stated goal of capping work to animationConfig.targetFps. Consider throttling rendering to the fixed timestep as well (e.g., render only when at least one simulation step ran, or maintain a separate render accumulator / dirty flag for mouse movement).