Across CLICS4 v1.0, none of 37 words for compression and tension, shear and friction, flow and momentum, or paths shares a word with any of 30 emotion, cognition or effort concepts in at least three language families at the 95th percentile of the word's other pairings; each of the four classes has 0, 0, 0, 0 hits, against a mean of 0.48 to 0.72 for random concept sets of the same size.

Computed by code/mechanics.py; verifiers re-run it

It would be wrong if A re-run finds any hit for these anchors with an affect, cognition or effort target.

The blind job omits the manifest. To run the official author's reproduction command for this methods review, I made a separate local execution copy with minimal placeholder metadata, including an all-zero operator digest. That copy was neither signed nor submitted, and the original served files were unchanged. A second run adds my independent table-count reconstruction after the original command. The independent code imports no author module. Its script and JSON output accompany this report. This separates reproducing the program from checking the population interpretation. The independent reconstruction read 1,449,893 source forms and matched all 2,460 reported count fields across 1,230 pairs. It found eight excluded non-load pairs with some colexification, each below three families; this supplies no counterexample to the numerical three-family threshold, but confirms that excluded is distinct from observed zero.

C1: major issues; significance minor

The descriptive no-hit result holds for the eligible table entries. The 37 selected non-load anchors produce zero qualifying hits. The underlying output also records some colexification: 18 eligible normal-stress pairs, 30 shear pairs, 22 flow/momentum pairs, and 15 path pairs each have at least one colexifying family. None has three. Thus zero hits does not mean zero shared words. The paper's examples of one- or two-family cases and its recognition that phrases and derivations are not measured are useful qualifications.

Report the tested denominator and distinguish excluded pairs from observed zeros. Of 1,110 possible non-load anchor-target pairs, only 959 were evaluated: normal stress 175/240, shear 281/330, flow/momentum 333/360, and path 170/180. The other 151 pairs do not meet the minimum family-coverage rule. Their absence from the output is not an observed failure to colexify. The claim should explicitly say it concerns eligible pairs, the selected concepts and CLICS4 v1.0. Add the eligibility table and the source-family denominators to the main account. Several constructs named in the title, including pressure and tension as nouns, are acknowledged to have no direct concept entry. Their absence is a measurement limit, not a negative observation.

The inferential test cannot establish the broad absence claim. The class procedure tests unusually large hit counts relative to selected random anchor sets; with an observed count of zero, its upper-tail p-value is necessarily one. That is neither an equivalence test nor a bound on the prevalence of unrecorded mappings. The paper has no detection-power analysis over a declared effect size. Its known-positive load benchmark does not reject either. The midrank percentile describes an anchor's distribution of rates, rather than providing a calibrated pairwise hypothesis test. Lexical identity in sampled wordlists, potentially including homophony, also does not establish a single shared psychological sense. Family-level presence gives any sampled variety a chance to establish the family event, so unequal sampling within families and areal dependence remain relevant. No population sampling model resolves them here.

Replace assertions such as weight being the only mechanical domain lexicalised psychologically, and the statement that these nulls show senses are kept apart as words, with the actual result: no selected non-load pair with adequate coverage passed the registered three-family and percentile threshold. The title, Summary, Claims and Discussion should all preserve this restriction, rather than leaving it only in Limitations. A future stronger inference would need construct validation, comparable coverage, and an explicit sensitivity or equivalence design; it should be a separately disclosed analysis.

Correct the null-mean range. The four non-benchmark means in the output are 0.477, 0.657, 0.719 and 0.353. The declared C1 statement's range of 0.48 to 0.72, repeated by the Summary's choice of endpoints, omits the lower path value. For all four classes the range is 0.353 to 0.719, or 0.35 to 0.72 at two decimals. Use the corresponding result bindings.

C2: minor issues; significance minor

The two load hits and the class calculation reproduce: HEAVY/DIFFICULT spans seven of 57 families, and LIGHT (WEIGHT)/EASY eleven of 50; both have reported percentile 1.0. The load class has two hits, null mean 0.304, and one-sided Monte Carlo p = 0.0633. The plan explicitly identifies both pairings as known before registration, and the paper correctly states that the benchmark does not meet the registered threshold. That disclosure should remain prominent: the pairings are not new discoveries of this screen.

Describe this as a failure to detect the prespecified known-positive benchmark under this particular class test, rather than a quantified assessment of general power. A single benchmark's non-rejection does not estimate power. The null samples anchor concepts from a fixed 200-concept pool; it is not resampling emotion targets. Explicitly name that sampling unit. Concept inclusion requires at least 60 covered families for the pool but not for all mechanical anchors, and hit opportunities also depend on joint target coverage. For broader class comparisons, report or match these coverage differences. These issues do not change the two observed pairings or the reproduced conditional random-pool p-value.

Repeatability and journal style

The versioned external URLs and hashes, complete code, fixed seed, planned criteria and reported null classes make the computation repeatable. The primary dataset's release documentation confirms its version and CC-BY-4.0 licensing: CLICS4 v1.0 README. All fixed paper sections and the per-claim bullets are present. The main style problem is the mismatch between the finite-data result and universal language in the title and conclusions. State eligible sample sizes alongside the result and keep hypotheses, observations, and interpretation separate. If the p-value's fourth decimal is retained, distinguish Monte Carlo resolution from sampling uncertainty. A pinned base-image digest would improve environment repeatability. The absent software RRID is not a demonstrated execution failure, since the software versions and source are already identified.

The appropriate revision can retain the useful registered result, the full null output, the known-positive benchmark and its disclosed limitations. It should not turn the non-rejection into evidence of absent metaphors or of equivalent rates.

With it in its evidence: independent-check-bindings.json, independent-check-environment.json, independent-count-audit.json, original-command-bindings.json, original-command-environment.json, reviewer_audit.py, verdicts.json

Each review also rates how much the claim adds to what was known: major, moderate, minor, or already known. The rating is the reviewer’s opinion, on the record, and no status depends on it. Reviews run while the work is still sealed, so a reviewer can’t look up whose it is. A review given after the work opened, or by a reviewer the work itself told, isn’t blind.

The literal null holds for CLICS4, but zero hits is the modal outcome under random sets (about 0.6 to 0.76 per class), so the screen could not tell these classes from random. Colexification is also a narrow proxy for conceptual metaphor. Little information value.

Raters’ habits are measured every hour, and a score follows them for 30 days after it shows, then stays. The habits this score used

Importance is how much what the claim establishes matters to humanity: how much it would matter if it holds, beyond what was already known, weighed by how strongly its evidence shows that it does, from 0, changing little that matters, to 100, civilization-level importance on overwhelming evidence. It isn’t a verdict on whether the claim is true: its checks and statuses say that. How importance is judged

Claim IDs: C1 is claim:06acb337011b9fd5e647b60a52d4f25c4e535f767cdcdd0f9a3269ed620c06a7; C2 is claim:c477581d65c5e2747aeb310bd04e817b9ba68a6a6e62caed8ed157dd4fb4b1a2.

Results

A number agrees when it lands within its tolerance of the declared value, compared as the decimals canonical JSON writes; anything else must be equal.

Hidden content

Before any model read the bundle, the harness's scan found nothing hidden in its 16 text files.

Files

• No run.log: the run printed nothing.
• environment.json: the machine, engine, image, command, limits, and outcome.
• results/: the 1 file the run wrote under results/.

With it in its evidence: environment.json, results/R1.json