Question 1

How are AI submissions evaluated on SN121?

Accepted Answer

Each Challenge publishes a suite.yaml manifest, dataset with multiple scenarios, and a rubric. Validators load a miner's submitted SKILL.md into a fixed agent harness and run it across every scenario. Four graders (skill_use, scenario_quality, holistic rubric, novelty_check) score each run with published weights, producing a per-scenario score (weighted average of graders) and a final score (mean across scenarios). All grader weights, logic, and per-test rationales are published so anyone can recompute scores.

Question 2

What do miners submit to compete on SN121?

Accepted Answer

Miners submit a SKILL.md file: structured instructions and procedures that teach an agent how to complete the Challenge. The skill file is what gets tested during evaluation and is the same format that sundae_bar hosts in its commercial library. Miners can iterate on submissions during the open Challenge window subject to daily and per-Challenge limits.

Question 3

How does a Challenge start and end on SN121?

Accepted Answer

sundae_bar identifies business demand and publishes a Challenge (brief plus evaluation suite). Miners submit Skills while the Challenge is open. Validators score submissions continuously and update the leaderboard. When any submission reaches the qualifying bar (currently 80%), sundae_bar's research agent reviews whether the capability meets commercial standards, and a closing window begins. When the Challenge closes, emissions stop and the next Challenge cycle begins.

Question 4

Who earns emissions on SN121?

Accepted Answer

The miner holding the #1 position on the leaderboard earns emissions. Rank is determined by highest score first, then earliest qualifying submission. The current share is 40% of validator weights assigned to #1, accruing for as long as that miner holds the lead across weight-setting cycles. When a Challenge ends and no new Challenge is active, all miner emissions are burned.

Question 5

Can I audit the evaluation suite and scoring logic?

Accepted Answer

Yes. Every Challenge publishes its suite.yaml, dataset.jsonl, rubric.txt, and grader weights on the Lab challenge page. The evaluation harness is fixed and the grader logic is deterministic. You can download the suite, inspect how scores were computed, and verify results against the published graders and weights.

Question 6

What happens to high-performing Skills after a Challenge ends?

Accepted Answer

High-performing Skills may be integrated into sundae_bar's commercial platform and listed in the Skills and Agents library. sundae_bar may modify a skill to fit product and customer use cases better. When that happens, the listing is labeled as built on the sundae_bar Lab. Skills that are integrated can be drawn on by Scout when composing solutions for customers.

sundae_bar

sundae_bar

Hardware Requirements

Hardware Requirements

Miner Setup

Validator Setup

Participate

Participate

Hardware Requirements

Hardware Requirements

Miner Setup

Validator Setup