Subnet 43: Graphite
Graphite is Bittensor Subnet 43, a decentralized network for solving graph optimization problems. The Graphite README source describes its current focus as the Traveling Salesman Problem (TSP): finding the shortest route that visits a set of points and returns to the start. The subnet turns that computation into a rewarded service: validators pose TSP problems, miners return routes, and stronger solutions earn better rewards.
What the Subnet Produces
The subnet’s output is solved routes for graph problems. Because the optimal answer to a hard TSP instance is expensive to compute, there is room for miners to compete on both solution quality and speed. Problems can be expressed as coordinate inputs for undirected graphs or as edge inputs for directed graphs, so both documented input forms can feed the same TSP-style evaluation task.
The README says Graphite supports both coordinate inputs for undirected graphs and edge inputs for directed graphs. That matters because a TSP subnet needs a common problem format before miners can compete on solver quality. Different input styles can represent different route and ordering cases while still leaving validators with comparable submitted routes.
Solver Context
Graphite’s README lists several reference solver approaches: a nearest-neighbour heuristic, a dynamic-programming method, beam search, and a hybrid pointer-network approach. Those examples are not the only possible strategies; they show the shape of the competition. Miners can bring heuristic, exact, search-based, or learning-assisted approaches to the same TSP-style route task.
That solver diversity is the point of making the problem competitive. A miner can improve by finding better tours, finding them faster, or adapting to problem cases where simple heuristics fall short. The subnet’s useful output is therefore not a single model checkpoint or dataset, but a stream of candidate graph routes that validators can compare.
Reward Context
The README frames miner rewards around solution quality within a time limit. It says submitted solutions are evaluated relative to the best solution in the cohort and a benchmark solution, with validators penalizing solutions that fall short of the benchmark. This keeps the reward target tied to useful route improvement rather than to returning any valid path.
For readers, the important distinction is that Graphite scoring is comparative. A route can be valid but still weak if it is much worse than the cohort leader or fails to beat the benchmark heuristic. That makes the subnet closer to an optimization contest than a binary correctness check.
Miner and Validator Roles
Validators currently generate synthetic graph problems and send them to miners. Each miner solves the problem with whatever algorithm it runs and returns a route, within a time limit. Validators then evaluate the returned solutions. Miners compete by improving the route quality and speed of the solutions they return for those validator-generated problems.
Source and Live Data
Live SN43 data is available on TaoStats. The mechanism details in this article are tied to the Graphite README rather than to live identity fields.
Relationship to Yuma Consensus
Subnet 43 uses Yuma Consensus to convert the graph-optimization weight vectors that validators submit into the emission shares distributed to miners and validators within the subnet each tempo. The linked documentation describes how validator weight submissions are aggregated into consensus weights for each miner registered on the subnet.
In Graphite’s context, validators generate synthetic TSP instances and evaluate miner-returned routes by comparing them against the cohort leader and a benchmark heuristic within a time limit, then translate those comparative quality scores into weight vectors for the subnet. The Emission documentation describes how those consensus weights determine each participant’s share of the subnet’s accumulated emission each tempo.
Development Stage Context
The Introduction to Bittensor describes subnet development as moving from localnet to testnet and then mainnet. For Graphite (SN43), that sequence changes how readers should interpret graph optimization examples and solver evaluation outcomes.
In localnet, Graphite-compatible miners and validators can be developed and tested in an isolated environment. Localnet optimization scores and emission outcomes do not represent production subnet performance.
On testnet, Graphite-compatible graph optimization solvers can be exercised in a shared, non-production network. Testnet solver evaluations and validator scores are separate from mainnet subnet state.
On mainnet, Graphite (SN43) is the live production subnet where miners solve graph optimization problems and validators score those solutions to determine real Bittensor emissions. The Graphite repository describes the mechanism that applies on the production network.
The Bittensor Networks reference separates mainnet, testnet, and localnet. A graph optimization result or emission outcome from one environment should not be read as representing production subnet performance in another environment.
Reader Boundary
Subnet 43 Graphite should not be read as generic Bittensor subnet documentation, a model-training subnet, or proof that any valid route earns top weight. It names one subnet’s graph optimization competition on netuid 43 (Understanding Subnets, Glossary: Netuid).
Comparative Scoring Uses Cohort and Benchmark Routes
The Graphite README describes solutions being evaluated relative to the best solution in the cohort and a benchmark solution within a time limit (Graphite README).
A returned route can be valid but still score poorly if it falls short of those comparison points.
Validators Generate Synthetic Graph Problems
The README describes validators generating synthetic graph problems and sending them to miners for solution within a time limit (Graphite README).
Miner contributions are therefore judged on route quality for validator-issued problem instances.
Validator Weights Still Flow Through Yuma Consensus
Subnet 43 uses Yuma Consensus to convert validator weight submissions into emission shares each tempo (Yuma Consensus, Emission).