Research records

Reports, plans, and saved results behind Making ‘why’ testable.

These records support Daniel Ivanov's article.

This is a public export of reports, plans, settings, question data, figures, and saved textual/numerical results from the completed local studies. Absolute home-directory paths have been shortened. Reports describe the experiments as completed on 7 September 2026; their verification statements refer to those original runs.

Model weights, full activation/vocabulary arrays, execution scripts, and the local environment are not included. References to those files inside saved records describe the original study. This export supports inspection of the reported cases and statistics; it is not the full replication environment. Model and lens provenance are in the reports and sources. The downloadable archive contains this public export.

The publication manifest gives hashes of the exported files. Original experiment hashes retained inside the records refer to the original files, before publication path cleanup.