Comment by chaxor

You're right that Merkle trees are useful here, but adding proof of work to, e.g. run the code to produce some lfs data (database, NN weights, physical parameters, etc), which is checked by a checksum, could be a fantastic addition for ensuring reproducibility in science. IPFS could be great for reproducibility if the commit hash was linked directly to the data checksum. Blockchain may not be necessary, but it could be an easier implementation of this feature.