Fusion Scoring

Score fusion combines low-frequency and high-frequency monitoring scores into a single continuous metric.

Formula

\[S_{\text{combined}} = \alpha \cdot \bar{S}_{\text{LF}} + (1 - \alpha) \cdot \bar{S}_{\text{HF}}\]

Where:

• \(\bar{S}_{\text{LF}}\) = mean subsystem pass-rate over LF cycles
• \(\bar{S}_{\text{HF}}\) = mean subsystem pass-rate over HF cycles
• \(\alpha\) = weight for the LF component (0 ≤ α ≤ 1)

Configuration

from qgate import FusionConfig

fusion = FusionConfig(
    alpha=0.5,         # Equal LF/HF weighting
    threshold=0.65,    # Accept if combined ≥ 0.65
    hf_cycles=None,    # Default: every cycle
    lf_cycles=None,    # Default: every 2nd cycle (0, 2, 4, ...)
)

Why Score Fusion?

On real IBM hardware, logical (hard) fusion with per-frequency thresholds is extremely sensitive to HF noise, causing 100% false-reject at moderate noise levels. Score fusion provides a soft decision boundary that absorbs these spikes.

Method	γ ≤ 5.0	γ = 10.0
Logical fusion	100% false reject	~50% accept
Score fusion	Robust	Robust