I'm writing account research into HubSpot from a Claude workflow. The agent uses Coresignal's MCP to pull company and employee context, then produces a brief and a priority score.

The brief gets regenerated when new information comes in. Historical headcount trends are part of the analysis, so sometimes the score changes even when the latest headcount hasn't moved much. The longer-term picture has changed, or the model has enough new context to reassess the account.

Right now, the latest output replaces the previous one. I'd like to retain the old score, the new score, and a short explanation of what caused the change.

One option is to store a JSON snapshot for each run in a separate table and keep only the current version in HubSpot. Another is to write a small change log to a custom object.

If you've done this, which approach turned out to be less painful to maintain? I'm trying to avoid building a whole audit service for something fairly simple.