Ground truth
The screenshot is the exact analysis frame the ground truth is computed on. Labels are imported (AtariARI / OCAtari) then verified causally on the bit-exact machine.
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.carnival).
← back to the Ground Truth ROMs · Paper 2 audit · method catalogue
The screenshot is the exact analysis frame the ground truth is computed on. Labels are imported (AtariARI / OCAtari) then verified causally on the bit-exact machine.
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 |
|---|---|---|---|---|---|---|
| A2 · Single-unit lesions | causal / intervention | 1.000 | n/a | n/a | 0.464 | 1.000 |
| A8 · Whole-state recording | descriptive | 1.000 | n/a | n/a | 1.000 | 0.117 |
| Activation patching / causal mediation | causal / intervention | 1.000 | n/a | 1.000 | 1.000 | 0.975 |
| Causal scrubbing | causal / intervention | 1.000 | n/a | n/a | 1.000 | 0.667 |
| Interchange interventions / DAS | causal / intervention | 1.000 | n/a | n/a | 1.000 | 0.895 |
| Attribution / edge patching | gradient / correlational | 0.998 | n/a | n/a | 0.951 | 1.000 |
| Logit / tuned lens | causal / intervention | 0.975 | n/a | n/a | n/a | 0.168 |
| Path patching / IOI circuit | causal / intervention | 0.667 | n/a | n/a | n/a | 1.000 |
| Occlusion | causal / intervention | 0.499 | 0.999 | 0.000 | 0.526 | 1.000 |
| Extremal / meaningful perturbation | causal / intervention | 0.472 | 0.945 | 0.000 | 0.500 | 1.000 |
| Grad×Input / DeepLIFT | gradient / correlational | 0.469 | 0.938 | 0.000 | 0.526 | 1.000 |
| Linear probing + control tasks | probing | 0.419 | n/a | n/a | n/a | 1.000 |
| KernelSHAP / Shapley | gradient / correlational | 0.414 | 0.827 | 0.000 | 0.447 | 0.403 |
| On-distribution counterfactual | causal / intervention | 0.383 | 0.766 | 0.000 | 0.474 | 1.000 |
| ACDC — automatic circuit discovery | causal / intervention | 0.364 | n/a | n/a | 0.222 | 1.000 |
| LIME | gradient / correlational | 0.331 | 0.661 | 0.000 | 0.474 | 0.403 |
| RISE | gradient / correlational | 0.330 | 0.661 | 0.000 | 0.474 | 0.403 |
| A4 · Pairwise correlations | gradient / correlational | 0.254 | n/a | n/a | 0.244 | 0.178 |
| A1 · Connectomics / data-flow graph | causal / intervention | 0.200 | n/a | n/a | 0.997 | 1.000 |
| A7 · Dimensionality reduction (NMF/PCA) | dim_reduction | 0.200 | n/a | n/a | n/a | n/a |
| A6 · Granger causality | gradient / correlational | 0.118 | n/a | n/a | 0.974 | 1.000 |
| A3 · Tuning curves | gradient / correlational | 0.000 | n/a | n/a | -0.028 | 1.000 |
| NMF/PCA dictionaries | dim_reduction | 0.000 | n/a | n/a | 0.986 | 1.000 |
| Expected Gradients | gradient / correlational | 0.000 | 0.000 | 0.000 | 0.526 | 1.000 |
| Guided Backprop | gradient / correlational | 0.000 | 0.000 | 0.000 | 0.500 | 1.000 |
| Integrated Gradients | gradient / correlational | 0.000 | 0.000 | 0.000 | 0.526 | 1.000 |
| Sparse autoencoders | dim_reduction | 0.000 | n/a | n/a | -0.112 | 1.000 |
| Vanilla gradient (saliency) | gradient / correlational | 0.000 | 0.000 | 0.000 | 0.526 | 1.000 |
| SmoothGrad | gradient / correlational | 0.000 | 0.000 | 0.000 | 0.526 | 1.000 |
Source: site_data.json
· generated by gen_site_data.py from the
committed §R records + leaderboard.json.