I've been thinking about proofs for Hivemind for a while, as I realized earlier on that there aren't any.

Short of a periodic hash fucntion tied to a data set that changes every block, but doesn't take too much to compute, I can't think of anything right now.

I also thought of hashing the various live objects that make up the Hivemind codebase, and have that exposed via API so people can poll randomly, or have the nodes broadcast it on chain, timestamped. I'm still reading on the `inspect.getsource()` function in Python, which gives results on actual live code. Still studying it so I can't say that it can work. Obviously taking a hash of code that's not live can be faked.

It's not feasible to hash the whole DB at intervals because read speeds and calc times are long.

I am a bit confused I always thought layer 2 was the things like hive-engine the sidechains and how scotbot basically runs a sidechain for things like leofinance? Although those posts go straight to the chain, the witnesses validate and the only place that hivemind comes in is like you said the DB hivemind is just storage but now I wonder because don't many of the old chain apis now exclusively write to hivemind? Then again all nodes are required to run hivemind I assume or how do they then provide data unless thre is an interface so you may choose to kinda "swap"  out the DB. Basically what I am saying is I feel hivemind or was under the impression hivemind is an upgrade to the current chain data far more integral than some layer 2 that you can do with or without... 

Hivemind just makes accessing the layer 1 data more efficient by storing the blockchain data in a sql database so it can be used in a Web3 application. All important validity checks are still performed in layer 1.