A: 1) If you have a dataset that includes snapshots you can load it. It would not leverage Scoop's native snapshotting feature. That needs to run on its own to capture snapshots over time.
But you can definitely upload a csv with historical data and analyze it. It does need structured tables to work.
2) LLMs are used to create the high level query, so it is possible that the same question might be interpreted if it is vague as asking a slightly different question. However, the LLM does not do the actual aggregation/calculation - that is done by a deterministic BI back end (columnar database with powerful query generation engine). So the same query will generate the same results on the same data. Definitely. You can see the query that the AI thought you were asking in the details (columns used, filters added, etc.)
3) MCP is terrible for analytics. It is really a document oriented API and is like drinking through a very tiny straw. (Which is why the Hubspot MCP integration on ChatGPT is so awful). Best is to extract data in bulk and load into Scoop directly, or load into a database or even a large spreadsheet.
4) You can share an interactive Canvas using our UI. Once you export it to Slides/PPTX, then it is static. But the live, shared version is promptable and drillable. You can use any slide background you want with any branding or color scheme. Literally just upload your powerpoint deck.
5) This kinda depends on your model. We just increased limits and believe that this should be reasonable for most people. If it's not, email me, and we'll discuss. The goal of the T3 limits is not to be reasonable, we are still getting a feel for how AppSumo customers would look to use it.