Currently, we mint Discourse cred for activity; every topic and post creates new cred. I recently threw together a prototype which allows us to mint cred based on likes instead of posts. This post explores how this changes our cred.
In keeping with the practices we came up with during the CredSperiment weight changes thread, I’m not going to mention any individuals by name. However, the analysis happens at the level of individuals, so you can argue this is just a fig leaf. Let me know if you see a better way to handle this discussion, and, as always, please be respectful and considerate towards your fellow cred-tributors.
First, I’ll link the two prototypes of how our (Discourse-only) cred looks with this change. We have:
The first thing to note is that the default cred graph is a lot “spikier”:
whereas the cred-on-likes graph stays a lot more consistent over time:
Basically what we’re seeing here is that when there is a big spike in activity (more posts and topics), it doesn’t necessarily result in a spike of likes, so the second approach is a little more “conservative” in cred generation. (Although, counterfactually: someone could create new accounts and then like hundreds of posts, which would create a cred spike of its own.)
Diving deeper, we can look at how the cred distributions compare at the individual level, alongside statistics on number of topics, posts, and likes received. Here they are for the top 8 Discourse contributors by cred (names elided):
|User||Default Cred %||Likes Cred %||Cred Change||Topics %||Posts %||Likes %||Likes per Post|
|sum or average||100%||100%||100%||100%||100%||0.9|
Basically, the big shift is that a prolific contributor who made up a large share of the raw activity, but a smaller share of the likes (as seen by their having a likes-per-post that was about half of the average). Changing to cred-on-like brings their cred share closer to their share of likes received; this shifts cred towards the other contributors, as seen by everyone else having a positive cred change.
Overall, I think this is a constructive change to cred. It brings the principle of review culture to our Discourse cred; a like is an implicit, low-friction form of review. This is potentially vulnerable to gaming; for example, someone could register a dozen new accounts and use them to mint a lot of new cred. However, Discourse already has a built in concept of trust levels that we could apply here. We could adjust the heuristic so that, for example:
|Trust Level||Minted cred per like|
Note that likes by L0 or L1 users would still flow cred, they just wouldn’t mint any cred. If you read over the docs linked above, you’ll see that getting to L2 (the first level that can mint any cred) requires visiting on at least 15 separate days, spending an hour reading posts, receiving 1 like, etc. I think this is a sufficient barrier to deter gaming, at least for our current community trust level.