← Back to technical notes

// benchmarks

.knowledge benchmarks and measurement notes

Treat benchmark numbers as local evidence about a repository and workflow, not as universal claims about every codebase or every agent.

// direct answer

What benchmark data can prove

Benchmarks can show whether route-first context reduced repeated repository crawls in a measured setup. They cannot prove a universal token-saving percentage for all teams.

// method

Measurement rule

The useful measurement boundary is the repository under test: generated routing bundle, metrics outputs, local release logs, and the exact commands that produced them.

  • Keep raw runs and release logs inspectable.
  • Report estimated savings as estimates, not guarantees.
  • Prefer before/after workflow evidence over broad marketing claims.
  • Link benchmark claims back to GitHub proof or generated local artifacts.

// proof links

Implementation references