Loading...
Loading
sundae_bar
Ⲅ (Gamma)SN121An enterprise AI platform that turns real business needs into ready-to-use AI solutions.
Overview
Registration Activity
Neuron registration trends
This Interval
0
— burn · — PoW
Target / Interval
Loading
An enterprise AI platform that turns real business needs into ready-to-use AI solutions.
Neuron registration trends
Businesses need AI capabilities tailored to specific workflows, but there is no systematic way to identify which capabilities matter most or to benchmark them against real-world performance requirements. Demand signals remain fragmented across customer conversations and market research, and AI development happens in isolation without pressure to solve problems that actually generate revenue.
A decentralized, incentivized network of developers can address this by connecting real business demand directly to competitive AI development. When business requirements are translated into structured Challenges with published evaluation criteria, developers compete to build the highest-performing solutions. Multiple independent graders and transparent benchmarks ensure quality and prevent gaming. The best capabilities feed back into commercial platforms where they generate revenue.
SN121 implements this model on Bittensor by having sundae_bar publish Challenges informed by actual customer needs, miners submit SKILL.md files that teach agents to solve those Challenges, validators autonomously score submissions across multiple scenarios using published rubrics and grader logic, and the top-performing Skills become candidates for integration into sundae_bar's commercial platform. Emissions reward the miner holding the #1 position, creating direct incentive for continuous improvement.
Use cases were auto-generated from the GitHub repository and may not reflect the latest changes.
SN121 is an incentivized AI development network on Bittensor that creates and benchmarks commercial AI capabilities for real business use. Developers submit AI Skills designed to solve structured Challenges informed by business demand, which are autonomously evaluated by validators against published benchmarks. The highest-performing Skills can be integrated into sundae_bar's commercial platform, creating a direct loop between business demand, competitive AI development, and commercial deployment.
SN121 operates a rolling Challenge cycle connecting business demand to AI development. sundae_bar identifies business requirements and publishes a Challenge (brief + evaluation suite with suite.yaml, dataset.jsonl, rubric.txt). Miners submit SKILL.md files (instructions for agents to follow). Validators run a fixed harness that loads the skill and executes it across multiple dataset scenarios, applying multiple independent graders with published weights (skill_use, scenario_quality, rubric judge, novelty_check) plus hidden pre-gates (skill_alignment, dataset_derived).
Scores aggregate per-scenario (weighted average of graders) and across scenarios (simple mean). A deterministic leaderboard ranks by highest score, then earliest qualifying submission. Validators post on-chain weights following the #1 miner (currently 40% of emissions).
When a submission reaches the qualifying bar (80%), a research review occurs and a closing window begins. Challenge then closes, emissions cease, and the cycle repeats with the next Challenge. Some validators child-key to sundae_bar's published scoring rather than re-run the full stack.
An enterprise AI platform that turns real business needs into ready-to-use AI solutions.
This information was auto-extracted from the GitHub repository and may be incomplete or out of date. Always refer to the subnet's GitHub repository for the latest instructions.
High-performing Skills feed into sundae_bar's commercial platform. All evaluation logic is auditable: suite, weights, and per-test rationales are published.
Each Challenge runs a rolling cycle: Challenge released → miners submit Skills (subject to daily and per-Challenge limits) → validators evaluate against published suite (multiple scenarios, multiple graders) → leaderboard updates and on-chain weights follow #1 → when qualifying score is reached, a research review occurs and a closing window begins (typically hours) → Challenge closes and emissions for that Challenge end. No Challenge active = 100% emissions burned. Scoring is fully auditable: published suite.yaml, grader weights, per-test rationales, and results JSON allow anyone to recompute scores.
Economic model includes pre-revenue (controlled emissions) and future commercial revenue (intended to fund ALPHA token buybacks, creating direct link between sundae_bar usage and subnet demand).
The sections above were auto-generated from the GitHub repository and may not reflect the latest changes. View repository
FAQs were auto-generated from the GitHub repository and may not reflect the latest changes.
Holder & total-staked metrics as of 1h ago (hourly snapshot)
Miners submit SKILL.md files (instructions and procedures for AI agents); hardware requirements depend on the model chosen for the specific Challenge's harness, specified in suite.yaml
Validators run evaluation harness against submissions using models specified in each Challenge's suite.yaml. Some validators child-key to sundae_bar's scoring validator rather than re-running full evaluation stack
Review the active Challenge brief and download the public suite bundle (suite.yaml, dataset.jsonl, rubric.txt)
Access sundae_bar Lab challenge page
Develop skill instructions and procedures that teach an AI agent how to solve the Challenge. Submit through sundae_bar Lab while the Challenge is open
Create SKILL.md file
Submit SKILL.md for evaluation; subject to daily limits (rolling 24-hour window) and per-Challenge limits. Monitor remaining submissions shown on Lab
Submit skill and iterate
Download grader rationale and inspect scored results JSON to understand how score was computed across multiple scenarios and graders
Review evaluation results
Obtain suite.yaml, dataset.jsonl, and rubric.txt from the Lab challenge page
Download Challenge suite bundle
Configure the fixed evaluation harness with load_skill tool to inject miner submissions. Use models specified in suite.yaml for that Challenge
Set up evaluation harness
Set up all configured graders (skill_use, scenario_quality, rubric model-judge, novelty_check) with published weights from suite.yaml
Implement graders
Process queued submissions against all dataset scenarios, apply pre-gates (skill_alignment, dataset_derived), compute weighted scores per scenario, and average across scenarios
Run evaluation queue
Submit setWeights transaction to SN121 based on leaderboard ranking (highest score, then earliest qualifying submission). Current #1 miner receives 40% of emissions
Post weights on-chain
Use the same weights from sundae_bar's extensive scoring rather than re-running full evaluation. Does not change incentive mechanism; amplifies the published vote(optional)
Optional: Child-key to sundae_bar validator
