ProCreations/repro-causal-representation-learning-with-optimal-compression-and-complex-treatments

SkewD released-artifact reproduction

This semantic-v4 package executes all six registered claims against the exact official SkewD release at commit 5023b30…, including every released numerical cell required for the 13- and 19-benchmark aggregates, the ECM ablation, and the increasing-skewness comparisons. Three claims verify and three are literal falsifications; there are no toy or inconclusive claims. Two warning-strict replays are byte-identical.

The scheduled target was precreated as a private untagged quota-safety shell for an earlier candidate. Its immutable slug mismatch is explicit; the sole paper tag and every scientific byte refer to SkewD mXn4ueFKnQ.

PYTHONDONTWRITEBYTECODE=1 PYTHONHASHSEED=0 PYTHONWARNINGS=error python3 -W error native_skewd_audit.py --out outputs
PYTHONDONTWRITEBYTECODE=1 PYTHONHASHSEED=0 PYTHONWARNINGS=error python3 -W error native_skewd_audit.py --out packaged_replay
python3 validate_evidence.py