Adaptive Deep Funding

TL;DR: What if we could build a mechanism where Gitcoin donors directly steer “deep funding” towards repositories/projects they care about?

Deep Funding

Deep Funding - a visual guide in 3 easy steps

The existing design for deep funding assumes that it’s possible to create a quality model (or ensemble of models) that sufficiently answers the question: “What is valuable?”

  • Even if multiple models are in place this feels like approach of central planner.
  • This approach doesn’t appeal to me as a donor. I want to direct my money to projects I care about, not rely on a third party’s determination. I am much more liberal when it comes to money that comes from Gitcoin round sponsors.
  • Deep funding also emphasizes good models as a key part of the mechanism. I’d rather have donors drive the mechanism, adjusting rules toward the desired outcome: funding dependencies.

I am going to take part in “model contest” and maybe it would produce good models that I like to use, but really interesting question for me is - can we build better deep funding using wisdom of crowds, instead of AI/ML?

Main Idea is to propose “Adaptive” flavour of deep funding:

  1. Initialize deep funding by starting with weights in a neutral position.
  2. Allow these weights to be adjusted through Gitcoin donations.
  3. Stream sponsor money to projects according to the current weights.

You can imagine this process as donors paying to teach a neural network to steer sponsor funds toward the outcomes they value.

If some donor prefers to use “AI agent” to decide where his funds should go - they could do that in this system.

This is still popularity contest, but thanks to “trickle-down” effect it does better job funding dependencies than current approach where each project in a round is independent.


Example Initial Position:

  • “Every dependency gets a fixed share of the matching pool per hour.”

Simple Round:
A round with four projects:

  • “Gitcoin,”
  • “Gitcoin-Citizens,”
  • “Chainlink,”
  • “Chainlink-Marines.”

Each starts earning $3/hour. Donating to the “Gitcoin” project would also benefit “Gitcoin-Citizens” (as part of its dependency tree) - but not as much as if we donated to “Gitcoin-Citizens” directly.

Total rate of funding coming from sponsors is some fixed number (e.g. 12$/h).

Key Differences from Gitcoin Today:

  1. Dependencies:
    Projects must declare dependencies (e.g., “Bankless” → “Bankless-London”) or infer them (e.g., software dependency trees).
  2. Trickle-Down:
    Matching needs to account for the “trickle-down” effect when calculating final distributions.
  3. Trickle-Up:
    If dependencies get donations “upstream” project should benefit as well.
  4. Streaming Matching:
    The matching pool could either stream in real-time or calculate at the end of a round.

Considerations:

  1. Exact algorithms for weight updates/trickle-up/down are required.
  2. “Initial distribution” could be more complex - something metrics based. It doesn’t need to be perfect, because donors should quickly correct it.
  3. Projects may excessively split into smaller units to capture more initial funding (happens already).

I am still trying to iron out how such “deep funding through crowdfunding” mechanism could work so would be happy for any feedback or alternative ideas!

2 Likes

Love this model, perfect example of multiple mechanism funding, leveraging various capital allocation mechanisms.