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 44 scored games. Every number is read from
site_data.json (games.yars_revenge).
← 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 |
|---|---|---|---|---|---|---|
| A8 · Whole-state recording | descriptive | 1.000 | n/a | n/a | 1.000 | 0.031 |
| Activation patching / causal mediation | causal / intervention | 1.000 | n/a | 1.000 | 1.000 | 1.000 |
| Causal scrubbing | causal / intervention | 1.000 | n/a | n/a | 1.000 | 0.857 |
| Interchange interventions / DAS | causal / intervention | 1.000 | n/a | n/a | 1.000 | 0.737 |
| Path patching / IOI circuit | causal / intervention | 1.000 | n/a | n/a | n/a | 1.000 |
| Attribution / edge patching | gradient / correlational | 1.000 | n/a | n/a | 0.983 | 1.000 |
| A2 · Single-unit lesions | causal / intervention | 0.999 | n/a | n/a | 0.693 | 1.000 |
| ACDC — automatic circuit discovery | causal / intervention | 0.941 | n/a | n/a | 0.385 | 1.000 |
| Logit / tuned lens | causal / intervention | 0.824 | n/a | n/a | n/a | 0.147 |
| A5 · Local field potentials | gradient / correlational | 0.607 | n/a | n/a | n/a | n/a |
| A7 · Dimensionality reduction (NMF/PCA) | dim_reduction | 0.600 | n/a | n/a | n/a | n/a |
| A3 · Tuning curves | gradient / correlational | 0.553 | n/a | n/a | 0.358 | 0.500 |
| Sparse autoencoders | dim_reduction | 0.545 | n/a | n/a | 0.181 | 0.500 |
| NMF/PCA dictionaries | dim_reduction | 0.464 | n/a | n/a | -0.355 | 1.000 |
| Grad×Input / DeepLIFT | gradient / correlational | 0.422 | 0.845 | 0.000 | 0.464 | 1.000 |
| On-distribution counterfactual | causal / intervention | 0.407 | 0.776 | 0.039 | 0.464 | 1.000 |
| KernelSHAP / Shapley | gradient / correlational | 0.406 | 0.771 | 0.041 | 0.464 | 0.673 |
| LIME | gradient / correlational | 0.406 | 0.771 | 0.041 | 0.464 | 0.673 |
| Occlusion | causal / intervention | 0.405 | 0.771 | 0.040 | 0.464 | 1.000 |
| A4 · Pairwise correlations | gradient / correlational | 0.401 | n/a | n/a | 0.250 | 0.286 |
| Guided Backprop | gradient / correlational | 0.385 | 0.771 | 0.000 | 0.464 | 1.000 |
| Integrated Gradients | gradient / correlational | 0.385 | 0.771 | 0.000 | 0.464 | 1.000 |
| Vanilla gradient (saliency) | gradient / correlational | 0.385 | 0.771 | 0.000 | 0.464 | 1.000 |
| SmoothGrad | gradient / correlational | 0.385 | 0.771 | 0.000 | 0.464 | 1.000 |
| RISE | gradient / correlational | 0.378 | 0.756 | 0.000 | 0.464 | 0.211 |
| A6 · Granger causality | gradient / correlational | 0.345 | n/a | n/a | 0.878 | 0.450 |
| Extremal / meaningful perturbation | causal / intervention | 0.260 | 0.520 | 0.000 | 0.464 | 1.000 |
| A1 · Connectomics / data-flow graph | causal / intervention | 0.200 | n/a | n/a | 0.949 | 1.000 |
| Linear probing + control tasks | probing | 0.072 | n/a | n/a | n/a | 1.000 |
| Expected Gradients | gradient / correlational | 0.000 | 0.000 | 0.000 | 0.464 | 1.000 |
Source: site_data.json
· generated by gen_site_data.py from the
committed §R records + leaderboard.json.