As per conversation in our catch-up this week, we have discussed the usage of source ‘ground truth’.
We all agreed that ‘Ground Truth’ should be the most reliable label source, which means Ground Truth >> other sources.
In the meeting, we clarified that, Ground Truth labels:
Should be captured by the labeller(s) via conducting actual transactions on chain
Should be backed by evidence - transaction hash, screenshots, etc
We understand that labels in the scenario below are also being identified as ground truth labels at the moment, but they are not true ground truth labels:
Via identifying transaction patterns on chain - e.g. addresses that are interacted with DeFi protocol are labels as defi users
Via Third party purchase that does not include transaction evidence
Via research over publicly available - forum, twitters, etc
I think for the next step, we can clarify the definition of ‘Ground Truth’ in our Taxonomy doc (with examples).
Contributors may also consider reassigning source values to the ones that are not real ground truth.
discord: https://discord.com/channels/1070229930668982352/1174412237474111488
As per conversation in our catch-up this week, we have discussed the usage of source ‘ground truth’. We all agreed that ‘Ground Truth’ should be the most reliable label source, which means Ground Truth >> other sources.
In the meeting, we clarified that, Ground Truth labels:
We understand that labels in the scenario below are also being identified as ground truth labels at the moment, but they are not true ground truth labels:
I think for the next step, we can clarify the definition of ‘Ground Truth’ in our Taxonomy doc (with examples). Contributors may also consider reassigning source values to the ones that are not real ground truth.
Current ground truth labels from Messari: