The Argument and the Letterhead — empirical data

Empirical data supporting Michele Loi, The Argument and the Letterhead: Source-position coherence in AI evaluation (https://arxiv.org/abs/2609.35286).

The deposit contains three registered collections: the initial US study (1,024 usable ratings; 1,038 client attempts), the US/Germany/Switzerland extension (1,664 ratings; 1,669 attempts), and the European Jev supplement (288 ratings; 347 attempts). Total: 2,976 selected ratings, 3,054 attempts and 154 cells.

The complete ZIP includes 9,159 original saved request/response/result files, unchanged completed descriptive reports, five CSV tables, nine previously public frozen scientific packages, post hoc calculation code and outputs, a data dictionary, provenance records, SHA-256 hashes and an offline verifier. Scientific code and preregistrations: https://github.com/MicheleLoi/Petri_studies.

The tables include 41 full-sample source-pair comparisons and both planned chronological halves (123 rows), plus 52 post hoc comparisons and 104 nominal p-values. All selected ratings, saved-response parsing, cell summaries, contrasts and existing p-values were checked offline; no new experiment or inferential analysis was conducted for this deposit.

Three technical-failure attempts lack saved responses; this missingness is explicit. Original flags and retrospective eligibility reviews remain distinguishable. The post hoc tests were not preregistered. The earlier 26-log exploratory audit is separate and not included. Private human-assistant conversations, Giano records, credentials and account/billing records are excluded. Historical references to a private raw archive describe the September 2026 state; this deposit releases the scientific data and completed reports.

The ZIP contains the complete dataset and verification materials. Its tables/ directory provides CSV access to individual ratings, attempts, cell summaries, contrasts and post hoc tests.