Author(s): @David Dryjanski @Deleted User

Problem Statement

Many SPs do not accept any retrieval requests and there is no on-chain incentive for them to do so, nor there is punishment for not providing retrievals

• More detail in this presentation

Summary

Golden Retriever is a retrieval rewards program that evaluates retrieval performance on the Filecoin network and pays out rewards to the top performing Storage Providers.

• The MVP will leverage data captured through Autoretrieve bridge run by Bedrock to minimize the amount of engineering effort required to launch and go to market.
• Once retrieval data is more widely captured (e.g., via Validation bot), we can extend the program to data programs (i.e., Slingshot, Discover)

Links

Autoretrieve database

Autoretrieve dashboard

Goals & KPIs

• Main goals

  • Test the hypothesis that SP retrieval behavior can be altered by providing transparency into their retrieval quality or promise of a reward
    • KPI: Gather the data that allows to confirm the hypothesis and learn about SPs behavior
  • Incentivize SPs to serve retrievals reliably
    • KPI 1: Overall retrieval DSR in Autoretrieve
    • KPI 2: Number or SPs that qualify as “good retrieval providers”

• Additional Goals

  • Build a reliable DSR monitoring tool to objectively evaluate the state of the network.
    • KPIs: SP coverage (% of SPs), dashboard uptime (reliability), deals per week (scale)
  • Understand what amount of reward is needed to change SP behavior.
    • KPIs: Cost Per Retrieval Statistics

MVP Program (using Autoretrieve)

High level flow

• The rewards program will evaluate the retrieval performance of a set of SPs
• For every evaluation period (1 week):
  • Calculate metrics for all SPs
  • Identify a pool of “qualifying” SPs that pass all Retrieval Success Criteria, rank SPs and
    • Identify top 3 by the number of successful retrievals served
    • Run a random lottery to determine additional 5 winners