Existing analyses:limits and conditions.

What can be established about past work, what cannot be recovered, and how limited results should be read.

What cannot be recovered

Pre-registration: Registering a plan now does not show that it was specified before the data. Missing history stays missing.
The original program: Agreement with today’s recomputation does not prove that the original software produced the reported values in the same way.
Unknown preprocessing: Undocumented preprocessing is not reconstructed.

No code execution: Existing scripts are read to propose settings; they are not executed, and their correctness is not certified.
No extraction from PDFs or images: Data are not extracted from arbitrary PDFs or images.
No method changes to make it pass: Methods are not changed automatically until a guarantee becomes available.

Supported workflows only: Past analyses are recomputed only within the supported workflows. Without the original data, only a limited comparison of reported results is possible.
Limited is not complete: A limited result states what holds and what does not. It is not a confirmatory claim or a full-chain guarantee.
Assumptions stay assumptions: Calculations under assumed conditions are labeled as such and are not presented as confirmed for the real study.