Code and experimental results change over time. If continuity depends only on chat history or human memory, the reasons for failure, alternative paths, contemporaneous evidence, and rationale for corrections are easy to lose.
What belongs in the research journal
Each record groups the starting question, data and documents used, core logic, alternatives considered, failures and corrections, firm conclusions, open questions, and evidence locations. We also preserve concise explanations that can later be reused in public content.
live state → structured journal → off-host backupExecution systems and research archives are kept separate so that a failure in one cannot erase the history of the other.
Why off-host
Backups stored on the same server can disappear with the server itself. We therefore replicate research state, reports, journals, and source snapshots off-host and preserve checksums alongside them.
The boundary between public and private
Research methods and failure cases can be made public, but private keys, API tokens, real-time candidates, actionable scoring formulas, and sensitive infrastructure details are kept separate from both the Journal and the public site.