The Harness Layer
Loreto field paper No 01, September 2026. 29 pages for CIOs and CTOs, free to read. Three published results that do not agree with each other, the one comparison that holds the model fixed while changing the harness, and what it costs to build that layer yourself.
What it argues
Harness choice alone moved the pass rate twenty points
Eight harnesses, one model through one provider, 25 shared tasks and 200 scored results. Pass rates ran from 88 per cent down to 68.
The good news and the counter-evidence, together
Ten weeks of migration cleared in four days at one company; the same class of tooling measured 19 per cent slower in a controlled trial. Both are in…
Buying the harness and building it land in the same place
A loop that reads whatever your organisation has written down, and nothing else.
The better the model, the thinner the harness
Half of a harness compensates for the model and shrinks as models improve. The other half is your conventions, and nothing but writing them down shri…
How to get it
Free. Signed-in Loreto accounts download it from the page; everyone else leaves an email address and it arrives by mail.