Skip to content

Latest commit

 

History

History
35 lines (24 loc) · 1.68 KB

File metadata and controls

35 lines (24 loc) · 1.68 KB

NeuroTrace — offline benchmark results

Seeds: 16 · samples/circuit: 512 · each cell is the mean AUROC of recovering the known ground-truth circuit (1.0 = perfect, 0.5 = chance). No API keys, no downloads.

AUROC by method and circuit

method family clean (control) confounded suppressor
random baseline 0.496 0.496 0.496
correlation proxy 1.000 0.750 1.000
gradient proxy 1.000 1.000 0.750
integrated_gradients causal 1.000 1.000 1.000
patching causal 1.000 1.000 1.000

Effect 1 — confounding breaks correlation, not causation

On confounded, a non-causal feature is correlated with the output. Correlation-based attribution falls from 1.000 (clean control) to 0.750, while activation patching stays at 1.000 — it reads the finite causal effect, which the decoy does not have.

Effect 2 — saturation breaks gradients, not causation

On suppressor, a non-causal feature has a large local gradient but ~zero net effect (two near-cancelling steep gates). Gradient saliency falls from 1.000 (clean control) to 0.750, while activation patching stays at 1.000. Integrated gradients — the path-integrated correction to plain gradients — recovers too.

Scrambled-label null (sanity)

Shuffle the ground-truth labels and every method collapses to chance, so the AUROC above is not an artefact of the metric.

method null AUROC
random 0.495
correlation 0.490
gradient 0.486
integrated_gradients 0.505
patching 0.505

Reproduce: python -m evals.harness (writes this file), python -m evals.gate (asserts the dissociation).