98 Ethereum repos. $350k in funding. Build a model that predicts human jury scores before they go public — and earn from every correct call.
Human experts compare repos pairwise and score them. Your job: predict those accepted scores before they go public.
A meta jury of three experts — Andrew Coathup, Conor Svenson, and Devansh Mehta — reviews every evaluator submission and accepts only those with strong, well-reasoned rationale. Your model needs to mirror the thinking of accepted evaluators, not just average opinion.
Markets resolve when human evaluators randomly sample specific pairwise comparisons from the full possible set. If your predictions moved market prices closer to human judgment, you profit. If they moved prices away, you lose. Random sampling keeps the system honest: you cannot game specific matchups because you never know which comparisons will be selected.
Three interconnected markets determine the final allocation of funds across all 98 repos.
Divides the full $350k between all 98 repos. Weights must sum to 1.
Determines how much a repo keeps vs. passes to its dependencies. Trade UP or DOWN based on your assessment.
Divides funds flowing from a seed node among its dependencies. Weights must sum to 1.
Your trading credits are structured so your downside is zero from your own pocket.
Keep regardless of outcome. Win or lose, this portion is yours.
If you lose, return only what remains. If you profit, keep all profits and return the original 80%.
1 sUSDS ≈ $1.07 · Additional trading subsidies provided by Seer for all participants
A parallel ML competition running alongside the Seer markets — one submission, two revenue streams.
Unlike quadratic funding, manipulation here creates profit opportunities for you.
If a repo attempts to inflate its own market price by buying UP tokens beyond what evaluators would reasonably assess, that overpricing becomes your opportunity.
Suppose a seed node buys UP to 100% originality. Human evaluators later judge it at 80%. DOWN tokens redeem at 20¢. Anyone who bought DOWN between 0–20¢ profits. The repo attempting to game the system loses money on UP tokens that only redeem at 80¢.
In quadratic funding, sybil attacks carry only reputational cost. In deep funding, they become your profit margin.
The round closes after 2 months. Evaluations go public, profits and losses are settled, and allocations finalize.
Study the 98 Ethereum repositories being evaluated at deep.seer.pm. Understand the ecosystem, dependencies, and relative contributions.
Predict human jury scores across all three market types: seed node weights, originality scores, and child node weights.
Place your predictions as trades on Seer markets. You receive trading credits structured as a 20% grant + 80% forgivable loan.
Enter your model in the Pond ML contest (timeline TBA) for an additional $20,000 prize pool on top of your Seer market earnings.
When the round closes, evaluations go public. Accurate predictions yield profit. All participants receive leaderboard prizes and trading subsidies from Seer.
The meta jury reviews every evaluator submission and accepts only those with strong, well-reasoned rationale.
Anyone with an ENS name can become an evaluator at deepfundingjury.com. Each submission requires 10 pairwise comparisons between seed nodes and 3 originality judgments — taking 2–20 minutes per question. Quality matters more than quantity.