C1,C2 computationally reproduced. Executed unchanged code/run in isolated Python3.12 with exactly NumPy2.5.3, network disabled, no host mounts. All3 CLICS4 external files were reused from an earlier independent public download and reverified against this bundle's lengths and SHA256 digests(48,053,189bytes total). Complete generated R1 equals supplied R1, including all pair-level outputs, class tests and random-pool results. Runtime19.55seconds within5minutes. Registered mechanics.py matches the analysis script. Outputs, run environment and digest checks attached.
Read paper, claims, code, plan, deviations, materials, provenance, data specification and results. Hazard none: public CC-BY4.0 linguistic data and original code; no restricted/private content identified. Integrity categories empty.
Interpretation caveats, distinct from numerical reproduction: the stated range of random-set means in C1 excludes the path class, whose R1 mean is.353, so the all-four range is approximately.35-.72, not.48-.72. The cited result paths themselves match. Zero qualifying hits does not mean zero colexification, semantic absence, or universal separation of senses: the screen excludes pairs with fewer than30 jointly covered families, and imposes both a rank and3-family cutoff. Counts are selected concept labels, not37 exhaustive words or all mechanical domains. Upper-tail class p1.0 is failure to find enrichment, not evidence of equivalence or absence. The load class p.0633 similarly does not independently prove limited power; the plan explicitly treats this benchmark as a warning. General statements that weight is the only mechanical domain lexicalized psychologically exceed a finite, thresholded wordlist screen. Quantify coverage and report no qualifying pairs under this operational definition. These issues belong in methods/domain review and do not alter exact computational agreement.