Too-frequent intervention turns review into a rubber stamp

What it is

The gate is in the right place and the materials are there, but it arrives every five minutes, and people shrink review to “press approve so work can continue.” Content is no longer read; mean dwell drops below a title. Too-frequent gates become rubber stamps is quantity eating quality. The card on nominal loops is causality not wired; here causality is wired and habit shorts it.

Stamping is not an attitude problem. It is a rate problem.

Why it happens

Attention and suspicion have refractory periods. Each gate has to re-raise the stance “this might be wrong.” Intervals shorter than the refractory period never raise it, and the act is taken over by procedural memory: the lit button sits in the same place, the hand beats the eye. Alarm fatigue and confirm-dialog fatigue are the same curve. A generate agent that gates every tool call will finish that curve in an afternoon.

Accuracy helps in the wrong direction: almost every gate should pass, negatives are sparse, and people learn faster that pass is the default. The sparse negative then lands when the habit is deepest.

Studying it

Manipulate arrival rate (gates per hour) and negative density; watch dwell, pupil or regressions, and catch of harmful items decay over time. Independent variables: whether gates are merged by consequence, whether a forced slow-down is inserted. Dependent variables: slope of decay, stamp classification (dwell below a readable floor).

Decay must be drawn by shift, not as a session mean. A mean will wash the first hour’s care and the fourth hour’s stamp into “fine.”

Where it stops holding

Rare, high-stakes gates (two irreversible acts a day) almost never stamp; frequency is not the issue. Safety checks that intentionally stop every time are another training regime and do not move onto an office agent as-is. After automation lowers frequency, remaining gates are rarer and heavier, so materials must be more complete, or they will be rare and blind. This entry does not treat how accuracy itself corrodes supervision, only arrival rate.

Applying it

• Merge by consequence: no gate on reversible steps; several preparatory steps of one outward act become one gate. Do not gate every step.
• Give the shift a review budget: over budget, lower automation or add people, rather than let the same person stamp faster.
• Log dwell. Releases shorter than a readable floor go into the quality report, not as successful releases.
• Check: dwell distribution in the second half of a real shift. A pile at 400 ms means stamping has started. Then plant a harmful item in that half — if it is not caught, rate is eating quality. Merge gates until the second half can still finish reading the objects.

Related

• Same group: L1.09.1 Where to put the intervention is decided by reversibility · L1.09.2 Intervenors need the context that reconstructs current state · L1.09.4 The higher the system’s accuracy, the worse actual supervision · L1.09.5 Putting a person in the loop also puts liability on them; that transfer must be said, not assumed
• Nearby: L1.05 Human in the loop · L4.03 Automation complacency · L4.11 Pre-action confirmation and irreversible operations
• Search terms: rubber-stamp review · alarm fatigue · confirmation frequency