The exact ground truth for this ROM and how every interpretability method scores
against it — one of the 42 scored games. Every number is read from
site_data.json (games.jamesbond).
← back to the Ground Truth ROMs · Paper 2 audit · method catalogue
Each XAI / mechanistic-interpretability method's faithfulness on this ROM, sorted by F (click a header to re-sort). The left border marks the tradition: green = causal / intervention, red = gradient / correlational, amber = other — the split the paper's headline turns on. F is all-regime; content-F and position-F are the two output regimes (n/a where this game has no record for that regime).
| Method | Family | F (all) | content-F | position-F | S | M |
|---|---|---|---|---|---|---|
| A8 · Whole-state recording | descriptive | 1.000 | n/a | n/a | 1.000 | 0.055 |
| ACDC — automatic circuit discovery | causal / intervention | 1.000 | n/a | n/a | 0.385 | 1.000 |
| Activation patching / causal mediation | causal / intervention | 1.000 | n/a | 1.000 | 1.000 | 0.909 |
| Causal scrubbing | causal / intervention | 1.000 | n/a | n/a | 1.000 | 0.500 |
| Interchange interventions / DAS | causal / intervention | 1.000 | n/a | n/a | 1.000 | 0.769 |
| A2 · Single-unit lesions | causal / intervention | 0.985 | n/a | n/a | 0.600 | 1.000 |
| Logit / tuned lens | causal / intervention | 0.764 | n/a | n/a | n/a | 0.105 |
| A5 · Local field potentials | gradient / correlational | 0.743 | n/a | n/a | n/a | n/a |
| Occlusion | causal / intervention | 0.724 | 0.747 | 0.700 | 0.893 | 0.964 |
| KernelSHAP / Shapley | gradient / correlational | 0.719 | 0.747 | 0.691 | 0.893 | 0.288 |
| LIME | gradient / correlational | 0.684 | 0.727 | 0.640 | 0.893 | 0.288 |
| RISE | gradient / correlational | 0.644 | 0.685 | 0.603 | 0.893 | 0.288 |
| A7 · Dimensionality reduction (NMF/PCA) | dim_reduction | 0.600 | n/a | n/a | n/a | n/a |
| Sparse autoencoders | dim_reduction | 0.580 | n/a | n/a | -0.097 | 0.875 |
| On-distribution counterfactual | causal / intervention | 0.578 | 0.747 | 0.408 | 0.893 | 1.000 |
| Grad×Input / DeepLIFT | gradient / correlational | 0.447 | 0.865 | 0.028 | 0.893 | 1.000 |
| Extremal / meaningful perturbation | causal / intervention | 0.402 | 0.489 | 0.314 | 0.893 | 0.964 |
| Expected Gradients | gradient / correlational | 0.388 | 0.747 | 0.028 | 0.893 | 1.000 |
| Integrated Gradients | gradient / correlational | 0.388 | 0.747 | 0.028 | 0.893 | 1.000 |
| Vanilla gradient (saliency) | gradient / correlational | 0.388 | 0.747 | 0.028 | 0.893 | 1.000 |
| SmoothGrad | gradient / correlational | 0.388 | 0.747 | 0.028 | 0.893 | 1.000 |
| Guided Backprop | gradient / correlational | 0.374 | 0.747 | 0.000 | 0.893 | 1.000 |
| Linear probing + control tasks | probing | 0.362 | n/a | n/a | n/a | 0.800 |
| A4 · Pairwise correlations | gradient / correlational | 0.272 | n/a | n/a | 0.345 | 0.107 |
| A3 · Tuning curves | gradient / correlational | 0.188 | n/a | n/a | 0.130 | 0.875 |
| Attribution / edge patching | gradient / correlational | 0.174 | n/a | n/a | 0.917 | 1.000 |
| A1 · Connectomics / data-flow graph | causal / intervention | 0.000 | n/a | n/a | 0.981 | n/a |
| A6 · Granger causality | gradient / correlational | 0.000 | n/a | n/a | 0.853 | 0.150 |
| NMF/PCA dictionaries | dim_reduction | 0.000 | n/a | n/a | 0.915 | 1.000 |
| Path patching / IOI circuit | causal / intervention | 0.000 | n/a | n/a | n/a | n/a |
Source: site_data.json
· generated by gen_site_data.py from the
committed §R records + leaderboard.json.