Skip to main content

Routing Rewards

Hokusai rewards contributors for measurable improvements to shared AI decision layers. For the Technical Task Router, the relevant improvement is better routing behavior for coding tasks.

What Creates Value​

Routing value comes from decisions that improve execution outcomes:

  • Higher task acceptance
  • Lower cost for comparable quality
  • Faster accepted results
  • Fewer retries
  • Better reviewer coverage
  • Lower regression rate
  • Better model selection under budget constraints

The router improves when it receives high-quality outcome data and when contributors improve the routing model, evaluation methods, or task representations.

Where Rewards Come From​

Routing calls can pay a per-decision fee. Those fees support the router's token economics. Contributors who create verified routing improvement can receive token rewards tied to measured performance lift.

The important distinction is that rewards are tied to observed improvement. The protocol should be able to compare routing behavior before and after an improvement and measure whether future outcomes got better.

What Counts as Contribution Data​

For the router, useful contribution data can include:

  • Task packets from real coding work
  • Route decisions and fallback behavior
  • Execution outcomes
  • Evaluation results
  • Human acceptance signals
  • Cost and latency records
  • Regression reports

Raw repository contents, secrets, customer data, and unnecessary artifacts should not be included in reward-oriented outcome reports.

Relationship to DeltaOne​

DeltaOne is Hokusai's unit for verified performance improvement. In the router context, a DeltaOne-style improvement can represent measurable lift in the router's objective, such as acceptance rate, cost-adjusted success, or reliability under an evaluation envelope.

Exact scoring depends on the benchmark or evaluation configuration. See Evaluations and Feedback.

Integrator Choices​

An integrator can decide how to handle rewards generated by its outcome stream:

  • Keep token rewards as infrastructure revenue.
  • Pass rewards through to users or task authors.
  • Split rewards between the harness operator and contributors.
  • Use rewards to fund further evaluation and routing work.

Those choices are integration policy decisions. The docs describe the protocol mechanics; integrators define the product experience around them.

Last updated on