Why it exists
Vendors publish energy advantages measured on their own terms: different workloads, different system boundaries, cooling in or out. Those claims cannot be compared. This benchmark fixes the task, states the meter boundary and counts only correct answers, so wrong runs make the number worse, exactly as they should.
The pipeline
Every release is tagged, every run keeps its evidence in the repo under docs/evidence.
First results
The first task is the H2 ground state via VQE, solved to chemical accuracy. On this workload the MacBook beats the datacentre GPU by roughly forty times, because a two qubit simulation cannot fill a 450 watt board. The interesting question the benchmark now makes answerable is where the crossover sits as circuits grow.
Roadmap: a real QPU through a cloud provider, then an AMD Kria class edge board as a permanent lab fixture. One backend per month, each a tagged release.