Conditioning Strategies¶

qgate supports three conditioning strategies, configured via the variant field in GateConfig.

Global Conditioning¶

All N subsystems must pass all W monitoring cycles.

\[P_{\text{accept}}^{\text{global}} = \prod_{w=1}^{W} \prod_{i=1}^{N} p_i^{(w)}\]

Limitation: Exponential decay with N — unusable at N ≥ 2 under noise.

config = GateConfig(variant="global")

Accept if at least ⌈k·N⌉ subsystems pass each cycle.

\[P_{\text{accept}}^{\text{hier}} = \prod_{w=1}^{W} P\!\left(\sum_{i=1}^{N} X_i^{(w)} \ge \lceil k \cdot N \rceil\right)\]

Advantage: O(1) scaling — maintains high acceptance from N = 1 to N = 64.

config = GateConfig(variant="hierarchical", k_fraction=0.9)

Continuous metric combining LF and HF scores:

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

Accept if \(S_{\text{combined}} \ge \theta\).

Advantage: Soft decision boundary absorbs noise spikes.

from qgate import GateConfig, FusionConfig

config = GateConfig(
    variant="score_fusion",
    fusion=FusionConfig(alpha=0.5, threshold=0.65),
)