sundae_bar

![](./assets/logo_banner.jpg)

### Incentivizing The Creation and Benchmarking of Commercial AI Capabilities for Business

---

## Overview

SN121 is an incentivised AI development network designed to create and benchmark commercial AI capabilities for real business use.

The system operates through an open, competitive pipeline where real demand helps determine what gets built, and developer submissions are autonomously evaluated against structured benchmarks. The process works as follows:

1. **Real business demand informs development.** sundae_bar identifies demand signals, including customer requirements and broader market demand, that can be translated into structured SN121 Challenges.
2. **Developers compete to build the solution.** Developers (Bittensor miners) submit AI Skills designed to solve the Challenge.
3. **Submissions are autonomously benchmarked.** SN121 validators evaluate submissions against structured tests and scoring criteria, with emissions rewarding the highest-performing capabilities.
4. **The best capabilities feed into sundae_bar.** High-performing Skills can be integrated into sundae_bar’s commercial platform, expanding the AI available for customers to use.

This creates a direct development loop between commercial demand and decentralised AI development:

**Business Demand → SN121 Challenge → Developers Compete → Best AI Wins → Integrated into sundae_bar**

As the platform identifies new requirements, SN121 can create new Challenges around them. As developers produce better AI capabilities, sundae_bar’s Skills and Agents library expands, giving Scout an ever-growing set of capabilities to draw from when finding or composing AI solutions for customers.

SN121 therefore creates competitive pressure around AI that businesses actually need, using open development and objective benchmarking to continuously expand and improve the capabilities available through sundae_bar.

## About sundae_bar × SN121

SN121 is owned and operated by Sundae Bar AI PLC (AIM: SBAR), a London-listed company. Launched in June 2025, sundae_bar operates an enterprise AI platform where businesses can discover, deploy and manage AI Skills, AI Agents and bespoke AI solutions through one simple interface.

SN121 powers this ecosystem as sundae_bar’s decentralised AI development network on Bittensor. Real business demand informs structured development challenges, developers compete to build the required AI capabilities, and submissions are autonomously benchmarked to identify the highest-performing solutions. The best capabilities can then be integrated into the sundae_bar platform for commercial use, creating a direct loop between customer demand, AI development and commercial deployment.

## Platform Roles

**sundae_bar Platform:** Identifying Business Demand, Orchestrating and Commercialising AI Capabilities

**SN121:** Developing, Testing and Benchmarking AI Capabilities

## Economic Mechanism

SN121 is designed to develop toward a revenue-backed economic model that connects commercial activity on the sundae_bar platform with the subnet economy.

Businesses will access AI through a flat-rate Agent subscription together with usage-based fees. As commercial revenue grows, a portion of this revenue is intended to support buybacks of ALPHA (121), creating a direct link between commercial usage of sundae_bar and demand for the subnet token.

Separately, SN121 uses emissions to reward developers for building the highest-performing AI capabilities against structured Challenges and benchmarks.

Over time, the intention is to create an economic loop in which real business demand drives what gets built, developer competition improves the AI available through sundae_bar, and commercial usage creates demand for ALPHA (121).

### Operating Modes

1. **Pre-Revenue:** Controlled emissions designed to support developer participation and continued development on SN121.
2. **Commercial Revenue:** A portion of revenue generated through sundae_bar is intended to fund buybacks of ALPHA (121), creating a direct economic link between commercial usage and demand for the subnet token.

## Launch and Automated Evaluation

SN121 evaluates commercial AI **skills** against structured, challenge-specific benchmarks. Each Challenge is a bounded capability: a real business workflow.

**Challenges as benchmarks.** sundae_bar turns demand signals into Challenges: a public brief plus an evaluation suite. The suite is **open to audit** on the Lab challenge page: scenarios, dataset, rubric, graders, and `suite.yaml`, which tells the evaluation system how to run the suite. New Challenges are released as a rolling series so the capability library grows over time.

**Skill development.** Developers (Bittensor miners) submit a `SKILL.md`, a skill file that teaches an agent how to perform the Challenge. Submissions are made through the sundae_bar Lab while the Challenge is open.

**Automated evaluation.** Validators run every submission through the same harness and the published `suite.yaml`. A base agent loads the skill, works a **set of scenarios** (not a single prompt), and is scored by the suite’s graders. Multiple scenarios produce a more stable, integral score than one lucky run. Each scored result includes grader rationale that miners can download and inspect.

**Validator execution.** sundae_bar runs the scoring validator that evaluates submissions and posts on-chain weights. Ranking is deterministic: highest best score, then earliest qualifying submission. Incentive weights are decided from that leaderboard and submitted on-chain (`setWeights` on SN121).

The eval itself is built to be fair, non-biased, and fully auditable. Each skill is run across **multiple scenarios** and scored by **several independent graders** (procedure-following, output quality, holistic rubric, and deterministic checks such as novelty). No single prompt or single judge decides the result. Grader weights, the suite, and per-test rationales are pub