HarnessGrad
A gradient is the signal you need before you can improve anything — measurable, attributable, comparable, and only worth having if it can be trusted.
control▁▁0.00
treat-A▁██0.00 → 1.00
The local demo, end to end. 4 tasks, no network, no cost.
01
Scoring is not delegable
A subject that scores itself makes every comparison meaningless.
02
Methods live outside
A method returns candidate harnesses. The platform measures those.
03
Isolation is recorded
Runs made under different rules never share an axis.