Static implementation findings

• Quantile ties and endpoint counts. The rule

does not enforce an exact retention count when magnitudes tie. At $s=1$, it retains the largest-magnitude entries instead of deleting every entry. Exact top-count selection with a declared tie rule is required for the identities in Section sec:definitions.

• Possible mask reactivation. Recomputing a density mask over already masked weights can produce a zero threshold. The rule $\text{abs}W\ge 0$ then marks deleted zero positions active. The default early-stage call starts from the dense baseline, so this is not asserted to have corrupted the supplied tables. It is a defect of the general pruning interface. The repair rejects an attempted increase in active count.

• Fractional pruning does not use its computed count. The original function computes $\lfloor f{n}_{\mathrm{active}}\rfloor$ but subsequently uses a quantile rather than deleting exactly that many positions. Quantile rounding and ties can therefore disagree with the stated removal count.

• Ignored allocation flag. The fractional-pruning function accepts per_layer but always performs layer-wise pruning. A repaired implementation must implement the global branch or explicitly reject it.

• Two density conventions. The pruning manager reports weight-mask density. The comprehensive evaluation counts all numerically nonzero parameters, including unpruned biases. The two denominators must not be interchanged.

• Random-cue probe is a shared random vector. The original call torch.rand_like(correct_cue) creates one two-coordinate vector and repeats it for all test examples. Its coordinates need not sum to one. It is neither an independent per-example cue probe nor a normalized ambiguous-cue sweep.

• Zero-cue ablation is out of the training cue set. The zero vector differs from both one-hot cues and from their convex combinations. Zero-cue invariance alone does not prove endpoint cue neglect. Endpoint switching, wrong-cue behavior, and cue removal must be reported separately.

• The probe is not a trained decoder. The supplied cue-ablation function evaluates accuracy under changed cues. It does not train or evaluate a representation decoder. “Decoder probe” should be reserved for an actual decoding procedure.

• Evaluation consumes training randomness. Stress, blended-input sampling, random cue generation, and model construction use the global Torch random stream. Adding an evaluation or changing branch order can therefore change later minibatch orders. Dedicated generators and branch replay are needed.

• Reproducibility has an environment boundary. A fixed seed is not a guarantee of bitwise identity across PyTorch versions or platforms PyTorchReproducibility. The software versions, hardware, generator states, and determinism settings must be recorded.