Replication package: The Scaffold as Specification. A Longitudinal Case Study of Clone-and-Own Reuse in LLM-Assisted Development of Four Production Web Applications

Measurement code and derived datasets supporting the study The Scaffold as Specification: A Longitudinal Case Study of Clone-and-Own Reuse in LLM-Assisted Development of Four Production Web Applications. The study reconstructs, from repository artefacts, how four production web applications were built by a single practitioner with an LLM coding assistant between February and May 2026, and how each was derived from an earlier one by clone-and-own reuse. Contents. Five measurement scripts, an export script that writes the derived datasets, a verification script that recomputes every table in the paper from the deposited data alone, and a figure script. The derived datasets are: one record per non-merge commit (project, commit hash, date, assistant version recovered from the Co-Authored-By trailer, files changed, insertions, deletions, subject line); one record per tracked source file (path, SHA-1 of the newline-normalised content, line count); one record per same-path file pair across project pairs with its similarity ratio; the structural metrics per project; and the ORM model names. What is not included. The four systems themselves are not released. They are internal applications of a university research unit, they run in production, and they hold operational and personal data. No file contents appear anywhere in this package: the per-file table records a cryptographic hash of each file, which is what the reuse measurements are computed from, and the hash does not permit reconstruction of the source. Session logs from the development period were not retained, so no prompt-level or interaction data exists. Scope of verification. The package allows a reader to audit how every reported number was produced, to recompute it, and to apply the same procedure to other repository sets. It does not allow independent confirmation that the inputs were the repositories described in the paper. That limit is inherent to the setting and is stated in the paper. Licence. Measurement code is released under the MIT licence; the derived datasets under CC BY 4.0.