Track draft versions and compare

An edtech marketing team rewrites a course-description page for the third time this month, and the newest version is actually weaker than the one from two weeks ago — a well-meaning edit dropped a detail that mattered, and nobody remembers exactly what the earlier draft said or how to get back to it. Trying to reconstruct an old version from memory or a saved email attachment is a slow, unreliable way to undo a mistake.

Every save here creates a version, whether the change came from Lekha's generation or a person's manual edit, and any two versions can be compared side by side or an earlier one restored with one click. This is a workflow and record-keeping capability, not a copy-generation one — it exists so a business never loses a version of its own writing to an edit that made things worse.

This is deliberately the lowest-risk capability in the whole suite from a copy-generation standpoint, because it doesn't generate or alter any text on its own — it records and displays what already happened, and comparing or restoring a version is a mechanical operation rather than one requiring a check against a source brief or a regulatory rule the way generation does. There is no hallucination risk here, no drift, no claim to verify, because nothing new is being written; a version is simply a stored copy of text that existed at a specific point in time.

The practical consideration that does apply is storage growth, which becomes a real factor if a version is created on every keystroke rather than on every meaningful save — a document edited heavily over months could accumulate an unwieldy number of near-identical versions if the granularity is set too fine, most of which nobody would ever actually want to restore individually. A periodic cleanup pass that prunes very old, minor versions while keeping meaningful milestones intact is a sensible way to manage this as volume grows.

What this doesn't do is judge which version is actually better — it makes every version visible and restorable, but the decision about whether the current draft or an earlier one reads stronger, or whether a fact was more accurate in one version than another, is still a human judgment call the diff view supports rather than replaces. It's the record-keeping half of good editorial practice, not the editorial judgment itself.