I'd keep the recorded diagnosis and the historian's interpretation visible separately. That would let readers see where a conclusion rests on the old description and where a later framework has been applied.

Display still matters for people who'll never request research access. I'd like museums to explain a few examples of work done on stored objects. That would make this argument easier to assess.

A long prompt can contain useful context too. I'd judge it by whether the agent can locate the behavior being changed. Cutting out the relevant file names to make a prompt shorter would lose something.

I'd also read an explanation from the person who asked it. They might recognize immediately which background detail an outsider needed. The finished answer probably hides some of that back-and-forth.

Preferences also expire. I'd want each note to show when it was added and which conversation it came from, so an old project requirement doesn't become a permanent assumption about me.

I'd be cautious about judging Charles solely by what later rulers did. A deal that reduced an immediate threat could still have been reasonable given his alternatives at the time. I'd like those alternatives spelled out.

I'd separate the work needed for a second platform from the benefit of keeping that option. It could be worth doing for a specific customer even if CUDA stays the first target.

The earlier spending could've been real and the emergency could still have been invented. Those are separate claims. That's where I'd focus the discussion of what evidence was available.