1. Saturn attempts retrievals back to SPs
   1. L1s can horse-race cache-misses to both the IPFS gateway and SPs
   2. Logs of attempts, speeds, results valuable to bedrock / bootstrapping reputation
   3. Timeframe: could implement requests in Jan. Work for getting logs out/centralized will take a bit more time.
   4. Transport: HTTP car matching IPFS gateway
   5. Incentives: centralized audit, already written
   6. Bedrock AI: propose /add incentive to SPs for initial HTTP testing.
   7. Retrieval Incentives Day Notes:
      1. @Marten Seemann (Libp2p Steward): Why no Libp2p in Saturn L1? (not for browsers but SP fallback - would enable existing protocols)
      2. Maybe Saturn is a Fil+ provider to participating SPs, evaluates via auditing with existing auditing mechanism, stop sending deals to SPs that don’t meet criteria
      3. Egress payments on fallback to SPs? Why doesn’t this work?
2. Saturn L2s attempt retrievals back
   1. Bedrock preference: retrieval lives in trusted compute base outside of Saturn
      1. pros: more powerful API. can coordinate across multiple container residents. can have longer lived connections pre-opened.
   2. Timeframe: beta in march. launch June
   3. Incentives: big question mark - reason for delay to June
   4. Current Status: dagstore + HTTP interface, written in go
   5. Future: Thomas, Alex want to rewrite in rust using Myel Pop code
   6. Transport: flexibility here
   7. Retrieval Incentives Day Notes
      1. Multiparty data transfer doesn’t help with small content (NFTs) - they’re just one block
      2. Realistically even with multi-party data transfer, L2 is not a good CDN for small content. But it’s good for large content like video maybe (ideally direct to browser)
      3. Maybe L2s initially are validators:
         1. Audit SP fallback to augment L1. Verify transfer parameters (bandwidth, latency) are sufficient
         2. Audit L1s? verify transfer parameters
3. End to end testing
   1. To build the case for selling Saturn as a CDN, we need to understand what the network can do
   2. We need to build an end to end to end testing — starting at a browser and going all the way to Boost. (Maybe we can use L2’s for this!)
   3. This will tell us where our bottlenecks are, help us fix issues.