Diagnose broken loops by mapping onto…

When This Fires

• When a system or practice isn't improving despite apparent effort
• When reviews keep identifying the same issues — the loop is broken somewhere
• When building new feedback loops — verify all four components before launching
• During systems audits when evaluating which loops are functional vs. decorative

Common Failure Mode

Adding more observation to an observe-heavy but adjust-light system: "We need better metrics!" when the actual problem is that existing metrics are already showing the discrepancy but nobody is changing behavior in response. The metrics aren't the bottleneck — the adjustment mechanism is. Diagnose before intervening.

The Protocol

(1) When a feedback loop isn't producing improvement, map it onto the four-part structure: Act: is the behavior being performed? If not → the loop never starts. Fix: address the activation barrier (Diagnose failing behavioral agents by component — trigger…, Start every new agent at under two minutes with zero…). Observe: is the output being measured? If not → the loop runs blind. Fix: add a sensor (A complete feedback loop needs three elements — define measured output). Evaluate: is the measured output being compared against a standard? If not → the data exists but doesn't produce signal. Fix: define the standard and comparison protocol. Adjust: when discrepancies are found, does behavior change in the next cycle? If not → the loop is decorative. Fix: install specific adjustment rules (Translate every discrepancy into a specific behavioral…). (2) Fix the first missing component you find — don't improve components downstream of the break. (3) After fixing, verify: does the loop now complete a full cycle from Act through Adjust?