Synthetic Score Calibration

1. Calibration Methodology

To exercise the mathematical dynamics of the scoring functions, we construct synthetic input cases:

• Controlled Isolation: Altering a single variable at a time (property type, paint surface area, duration, concurrency, viewport size).
• Observation Focus: Recording the deterministic numeric score ($0 - 100$) and assigned tier ($S, A, B, C, D, F$).
• Boundary: These are model calibration examples, not browser measurements. They do not establish user-perceived frame rate, actual VRAM allocation, or causal performance impact.

---

2. Calibration Cases

Calibration 01: Property Cost Hierarchy

The implementation assigns an ordered cost vector (Compositor $=0$, Paint $=20$, Layout $=50$). The resulting score also depends on paint area, duration, concurrency, and the other viewport pillars, so property type alone does not determine a fixed score or tier.

---

Calibration 02: Paint Surface Scaling Law

Paint penalties use the documented ratio clamp $\operatorname{clamp}(0.25, 4.0, \rho / 0.10)$. The clamp is an operational calibration choice, not a measured GPU fill-rate law.

---

Calibration 03: Duration and Micro-Interaction Discounting

The duration factor is a model discount that reaches $1.0$ at $100\text{ms}$ and is capped below at $0.25$. It must not be interpreted as proof that every sub-$100\text{ms}$ animation is imperceptible.

---

Calibration 04: Consecutive Frame Thrashing Severity

The analyzer applies a consecutive-frame multiplier and then combines the thrashing pillar with the other viewport pillars. A long burst increases penalty, but does not by itself guarantee Tier F; the final tier depends on the computed score and bottleneck caps.