Coverage matrix

A cell is one (task × protocol × grid × baseline) runner invocation; it is done only when every expected test-set result.json is valid. Three non-green states are kept distinct and none is hidden: pending (queued sweep work), failed (partial artifacts), and N/A (excluded by a stated applicability rule, not missing).

• pretrain_grid_only — multi_grid_pretrain runs only on the template_20kv/template_110kv corpora; held_out and the transfer protocols run only on the four reference grids (coverage.protocol_grid_applies)
• threshold_scope — threshold is a single-terminal overcurrent relay rule — it applies only to the binary local fault-detection task (coverage.baseline_applies)
• transfer_deep_only — transfer protocols need transferable weights, so they apply only to the deep baselines mlp/gru/cnn/resnet (coverage.baseline_applies)

legend: done · pending · failed · N/A by rule (hover any square for the cell and its reproduce command)

held_out · 580 done / 0 pending / 0 failed of 580

columns per task strip: grid (voltage order: cigre_mv · double_line · testgrid_110kv · ieee39) × baseline (majority · threshold · random_forest · mlp · gru · cnn · resnet)

multi_grid_pretrain · 144 done / 0 pending / 0 failed of 144

columns per task strip: grid (voltage order: template_20kv · template_110kv · template_345kv) × baseline (majority · threshold · random_forest · mlp · gru · cnn · resnet)

transfer_zeroshot · 576 done / 0 pending / 0 failed of 576

columns per task strip: grid (voltage order: cigre_mv · double_line · testgrid_110kv · ieee39) × baseline (majority · threshold · random_forest · mlp · gru · cnn · resnet)

transfer_finetune · 576 done / 0 pending / 0 failed of 576

columns per task strip: grid (voltage order: cigre_mv · double_line · testgrid_110kv · ieee39) × baseline (majority · threshold · random_forest · mlp · gru · cnn · resnet)

The same matrix is committed as coverage_snapshot.json with the exact runner argument line per cell, and summarized by python -m evemtbench.coverage --summary.