Eirik Botten Nicolaysen
Hypothesis. Among 20 confirmatory genealogical axis units (116 languages), pronunciation forms of three segments recur identically across at least three genealogically independent units less often than each unit's own phonotactics predicts: obs/E < 1.0 at form length 3. The direction is specified in advance; a ratio above 1.0 disconfirms the hypothesis rather than supporting it. Design. Confirmatory replication of a count. Forms are normalised to CLTS/BIPA, filtered by a grammatical-word exclusion, and grouped into clusters of identical segment sequences. A cluster counts when attested in at least 3 axis units and 3 languages. The observed number of length-3 clusters is compared with the expectation under a per-language positional bigram null refit on the same filtered corpus, reported with two uncertainty sources that are never pooled: Monte Carlo over 1000 null replicates, and a bootstrap over the 20 axis units. The confirmatory arm has not been analysed. The registered quantity has never been computed for any confirmatory unit. The blind is verified, not asserted: urortkontroll.py, included here, checks four independent traces and reports one stated limitation rather than claiming absolute untouchedness. The decision rule was fixed in advance and is cryptographically timestamped: if the 95% interval from either uncertainty source covers 1.0, the result does not stand. That record is anchored in Bitcoin block 960700. Identity relation, null model and adequacy bands were each fixed in a decision record committed before the measurement it governs. Resource type: Zenodo's vocabulary contains no 'preregistration' type. 'Preprint' is the nearest available and is used for that reason alone. Not included: the corpus, the population files, and the exploratory/confirmatory split assignment — publishing the assignment would reveal the confirmatory half.