Figure 16.4 Eval-driven development as plumbing. Every change runs against two suites: the
capability evals it is meant to climb, and the regression evals it must not break; tasks
that stabilize are promoted from the first into the second. The change reaches a gate (in
accent) where a threshold is enforced, not glanced at—the numbers, not moods, decide
whether it merges or is blocked.
