We are announcing the retirement and deprecation of blockprint. Despite our attempts with @miga_labs to revive the classifier and make it accurate post-Electra, today we are saying goodbye and moving on 🥲

@sproulM_ @ratedw3b @hanni_abu Clients retain a little room to express themselves through attestation packing — up to 8 attestations per block. In practice this isn't enough to differentiate them, as the attestations after the first are usually very similar: junk from out-of-sync validators.

@sproulM_ @ratedw3b @hanni_abu With Miga Labs we tried re-training Blockprint on post-Electra blocks, but couldn't crack the accuracy floor required to yield useful results. We had our machines grind through every combination of known features (grid search) and tried developing new features, to no avail.

@sproulM_ @ratedw3b @hanni_abu The blockprint GitHub repository has now been archived, and other repositories from the blockprint-collective will be migrated back under the sigp org in the coming days.

@sproulM_ @ratedw3b @hanni_abu To replace blockprint we are now relying on Miga's network crawler, which lacks a validator-centric view of client diversity but provides a good view of client diversity amongst running nodes. SigP may also have more to say about network crawlers soon, so watch this space.

@sproulM_ @ratedw3b @hanni_abu There are also other client diversity metrics in the works which are likely to be announced soon, and could help fill the hole left by Blockprint 🖤.