How do you handle pivots in Wayfinder without poisoning future Claude sessions?

[totoDoP]

So I am using Wayfinder. Previously I did grill-me's and made monster plans. Wayfinder is already better but how do I handle BIG pivots and second pivots?
So I have made fundamental decisions like how a drive should be mounted or picking database sources. I have done multiple rounds of research and decision tickets in linear trying to smooth out that design and then quite late something surfaces like my synology NAS hardrives are going to get murdered by this design and I need to pivot and this might be my second pivot. My map if FULL of research and decisions on the old design. How do people handle that kind of pivot in practice? Especially when part of it is implemented. I am worried future Ai sessions are all poisoned by the old paths.
I realize partly implementing things is a bad idea but a lot of discovery for me is implementing and trying it out to discover. These are particularly homelab related projects so until I offer it up to my infastructure I just don't know.

[hamedrabah]

I would preserve the old research, but stop presenting it as the current route.

Create a new decision ticket such as “Re-evaluate X after constraint Y changed.” Link it from the old closed ticket and mark the old conclusion superseded. Then revise or remove dependent open tickets, reconnect their blockers, and move invalidated work out of the frontier. Closed tickets remain the audit trail; open tickets should describe only the route you still intend to take.

That matches Wayfinder's model: the map is an index, while the decisions live in tickets. Tell Wayfinder plainly what changed and set the destination to the new decision. On a second pivot, repeat the same operation. Do not delete evidence that may still be useful; label the conclusion it supported as superseded so later sessions cannot mistake it for current guidance.

[707io]

I think the best approach is to treat a big pivot as a new branch rather than rewriting the old path. Keep the old research and decisions for historical context, clearly mark them as superseded, and carry over only what is still relevant. For partially implemented work, I’d document what needs to be removed or reworked so future AI sessions don’t treat the old design as still valid.