What it does
majordomus economics run executes each task of .ai/repo/benchmarks/economics/ in Claude
Code, once in the fixture as a team keeps it and once with Majordomus installed, and records
what the provider reported and whether the hidden acceptance tests passed.
majordomus economics measure counts, with no model, what the context compiler selects for
every issue of the plan. economics summary, explain and runs — and the same capabilities
over HTTP, MCP and the Cockpit — show every metric with its class, sample size and interval,
every pair including the invalid ones, and the verdict. economics check refuses a savings
number that no evidence produced.
What it does not do
It does not claim a saving the evidence does not support: below the publication rule the verdict says there is no verified claim. It does not compare counts across providers or tokenizers, re-price old runs, or record anything a session said.