This is part of our series on a year of Bittensor experience, leading up to our anniversary at the 13th of July. After discussing miner archetypes and incentive landscapes, the validator design is now discussed. After discussing validator opacity we now shortly touch on sharing of validator logging and data. Please let us know what you think in our Discord channel!
When analyzing subnet mechanics, we often scrape validator logs and miner submissions. We combine that data with our own internal analysis and the output of a locally run validator. Often we improve the locally run validator code to output more detailed data on submissions and scoring calculations.
All this is required to do proper big data analysis, which reveals patterns and insights that can be used to optimize mining or uncover validator bugs (as explained before, bug hunting is mining as well). The big data analysis can also be used to optimize the incentive structure and the subnet in general.
Instead of requiring the miners to do the hard work of data collection, the validator could, or perhaps even should, publish structured data for offline analysis. Simple JSON or CSV dumps on a webserver will do wonders.
Not only will data sharing allow for improvements, it also serves as a way towards validator accountability. Incidentally, subnet owners and/or validators are accused of wrongdoing, with them having not much more than the logs to point to, to defend themselves. Having the full data on display will allow them to more easily prove nothing is out of the ordinary.
The next article will discuss the singular validator.
0 Comments