I can see that this would be very useful.

This unfortunately requires some architectural changes to how sheriff-o-matic works. Right now, we just store a json file which is a list of alerts. We could provide some alert history, so you could go back and see different snapshots in time, if that would help. But we basically snapshot the state of the world every once and a while, and show you all the alerts based on the snapshot.

Does that make sense?

Oh, interesting. I don't think seeing old snapshots would be as intuitive without "diffing". Is there a way to track a particular alert across snapshots? 

I could imagine it working like this:
Take all snapshots for the last hour, put all alerts in a set (dedup), remove all the ones still active and show the rest in the "Done" area.

But I guess I don't know how difficult/expensive that would be for you to do?

We could change the server to use the same messages.Alerts struct that dispatcher uses (yay code sharing!) and actually parse the json into that struct on the server. Then store it in mdb as that struct instead of opaque json text. Then we can run queries against the datastore for historical analysis.

I think we could even run BigQuery over the datastore pretty easily.

Sean is correct in #3. That'd be the ideal case in the future. The original intent of my comment was to say that this would be a fair amount of work. Doing that refactor would allow us to do a lot of nice stuff, as well.