# EPF Cohort 5 - Week 13 and Week 14

Date: 16/09/2024

### Updates

1. I was mistaken about the dataset that the codex team had scraped. They had just estimated the number of validators per node and not logged peering information. So we will have to switch to existing graph models like `erdos_renyi`(random peering), `watts_strogatz`(social peering) and `barabasi_albert`(social and popularity peering)
2. We finished on the basic mechanics of the simulator and tested the existing functions of the rated list for basic functionality. A lot of fixes were made and we will have to update the spec.
3. The sample gossip part of the spec has to change since it was majorly done for a 1D sharding design but we want rated list to be abstract enough to support any kind of DAS that transpires in the future.
4. For the simulator we tried testing against random graphs but we couldn't generate a large enough graph. We oughta try other libraries for the same or parallelize the process.

#### Anecdote on Rated List

Once a bare minimum simulator was ready there was an urge to run tests and verify the working of rated list. Writing a test is difficult when you cannot concretely define decentralized behaviour (common to most p2p constructions). So what I did instead was to test against a centralized network and check the functionality of the system, as follows:
1. First I constructed an acyclic graph that represents a tree.
2. Then I marked an entire subtree offline to check if the rated list filters out nodes from that tree and that tree alone.

It took a lot of fixing but I finally saw the result. After the node hits an offline peer it starts aggresively filtering out nodes from that subtree.


To go one step further I had written a small function to determine membership in a subtree. And after more fixing, I could finally also see the intended result (see the last line of execution)


> You can find the specs and the simulator code of rated list [here](https://github.com/dankrad/rated-list-specs)

### Next Steps

1. Test against random graphs.
1. Improve the node profile architecture
2. Watch peerDAS breakout room recordings to know the latest update on that front

### Personal Notes

I want to leave the reader with a question on DAS. The sole purpose is to obtain different perspectives. What is the purpose of extending blob data using erasure codes?