There are a few misconceptions here:
- The ‘total amount’ of cred is actually independent of the edges in the graph. Right now every Discourse post is worth (say) 2 cred, and thus 2 cred is minted regardless of what that post is connected to. (In the future, I want most cred minting to happen based on community-evaluated value, e.g. initiatives and boosting, and not just based on raw activity.)
- Every node has an implicit connection that flows cred back to the “seed vector”, where the seed vector corresponds to which nodes minted cred. If there’s a node that doesn’t have any edges out, then all of its cred would just reset to the seed vector
- When a node does have edges out, the
alpha
parameter controls how much cred flows back to the seed vector, vs. how much flows to the edges out. It is possible to get a “cred black hole” when you have the following structure:
- A gives cred to X
- X gives cred to Y
- Y gives cred to X
Then A’s cred will tend to seep away to X and then “get stuck” bouncing between X and Y. If alpha is high, then this won’t be a big issue as it will pretty quickly all flow away to the seed vector. If alpha is low, however, then X and Y can crazily high cred scores.
Also, this construction can be made even simpler if X simply has a loop edge from X to X. Currently there’s no way for users to generate loop edges to themselves, but the two-step cycle happens all the time (I AUTHOR a post, and the post IS_AUTHORED_BY me).
Early in the CredSperiment, I increased alpha 4x precisely to mitigate this issue, you can read about it here.
I would really like us to build some tools (observable notebooks maybe?) that make it super easy to create little cred graphs and experiment with the cred flows. Then we can all get a good intuition for how cred really works, and make changes as necessary. See this initiative: Cred Analysis Notebooks.
Yeah, maybe we should switch the weights to put a little more love on GitHub and less on Discourse.
@Beanow would you be willing to come up with a proposal around weight changes, and then we can do some cred-weighted voting? Nice opportunity to keep testing our governance approach.
100% agree.