Baseline — threshold

benchmark version 1.1.0 · held_out results v1.0.0 · splits v1.0/held_out, v1.1/multi_grid
updated 2026-09-29 13:18 UTC · git 7243f310ddbc
What it does. A conventional overcurrent relay rule: trip when the current magnitude exceeds a threshold fitted on the training split. It represents a century of protection practice in one parameter and is the domain-classical reference the ML baselines must beat — defined only for the binary local fault-detection task, exactly like the physical device it mimics.

Reference: conventional protection practice (ANSI 50/51 style) · Implementation: src/evemtbench/baselines/threshold.py

Scope: threshold is a single-terminal overcurrent relay rule — it applies only to the binary local fault-detection task (coverage.baseline_applies).

Results (held_out)

threshold — every task × grid; click headers to sort
taskgridbaselineheadlinetestbenchmarkparamsfit/seedtrace
fault_detection_localcigre_mvthresholdbalanced_accuracy ↑0.660 ± 0.000 (n=5)0.810 ± 0.000 (n=5)—2strace: output · eval-code · log · train-code · config · W&B
fault_detection_localdouble_linethresholdbalanced_accuracy ↑0.729 ± 0.000 (n=5)0.916 ± 0.000 (n=5)—2strace: output · eval-code · log · train-code · config · W&B
fault_detection_localtestgrid_110kvthresholdbalanced_accuracy ↑0.755 ± 0.000 (n=5)0.886 ± 0.000 (n=5)—2strace: output · eval-code · log · train-code · config · W&B
fault_detection_localieee39thresholdbalanced_accuracy ↑0.793 ± 0.000 (n=5)0.914 ± 0.000 (n=5)—2strace: output · eval-code · log · train-code · config · W&B

Protection-grade & per-class metrics

Dependability/security rates and per-class F1 for this baseline — the same held_out result.json cells the Supplementary (Package C) tables select, traced here so every reported SI number resolves on the dashboard.