1.1 What is C-SWON?
C-SWON (Cross-Subnet Workflow Orchestration Network) is a Bittensor subnet where the mined commodity is optimal workflow policy — miners propose multi-subnet execution plans (DAGs), validators score them on task success, cost, and latency, and the network continuously learns the best orchestration strategies through competitive pressure.
"Zapier for Subnets" — The Intelligence Layer for Multi-Subnet Composition
The Problem
Bittensor hosts over 100 specialized subnets covering text generation, code review, inference, agents, data processing, and fact-checking — yet there is no native way to compose them into reliable, end-to-end workflows. Developers today:
- Manually wire calls to 5–10 subnets per application
- Guess at optimal routing with no objective benchmarks
- Rebuild orchestration logic from scratch every time
- Waste TAO on suboptimal routing through expensive or slow paths
Each new subnet that joins Bittensor increases the orchestration surface area — making the problem worse over time without a dedicated solution layer.
The Solution
C-SWON abstracts away manual orchestration for Bittensor's AI ecosystem. It turns any complex AI task into a single, optimized workflow through competitive pressure:
- Validators issue task packages with constraints (budget, latency, allowed subnets)
- Miners design DAG execution plans that chain multiple subnet calls
- Validators execute these plans in sandboxed Docker containers
- Scoring is deterministic: success rate, cost efficiency, latency, reliability
- Weights are set per tempo via Yuma Consensus — best orchestrators earn more
The Digital Commodity
The commodity mined by C-SWON is optimal orchestration policy: which subnets to call, in what order, with what parameters, to complete a given task at the lowest cost and highest quality.
The Kernel
Miners submit a WorkflowSynapse (an RPC endpoint receiving task packages and returning DAG plans). Validators pull responses and run a deterministic evaluation to produce weights. That is the entire kernel.
At its core, C-SWON is two Python files:
neurons/miner.py— receives task packages, designs workflow DAGsneurons/validator.py— executes plans, scores results, submits weights
Everything else in these docs explains those two things.
How It Prevents Gaming
C-SWON is designed to prevent miners from earning rewards without delivering genuine orchestration intelligence:
- VRF-keyed task selection — each validator derives its task from
hash(hotkey + block), making pre-caching impossible - Synthetic ground truth (15-20%) — hidden known-answer tasks verify miners are actually planning, not memorizing
- Benchmark rotation — tasks are deprecated when >70% of miners score >0.90 for 3 consecutive tempos
- Deterministic scoring — ROUGE-L, test pass rates, schema validation — no LLM judge, no subjective evaluation
Navigation
| ← Previous | Documentation Index |
| → Next | 1.2 Quickstart: Miner |
| Index | Documentation Index |
| Repository | github.com/adysingh5711/C-SWON |