Confirmation fatigue strips high-frequency confirms of their protection; confirms must be graded by consequence

What it is

It is already known that a gate popped too often wears out. The countermeasure is not longer copy. It is splitting confirms by consequence: no gate, a light gate (a skippable hint), a heavy gate (release only after the instance is read). Grade confirms by consequence spends protection on the few real world-boundary crossings, and stops the rest from grinding the person down.

One dialog shared by every act is every act sharing the same fate of being clicked through.

Why it happens

Attention is finite; habituation tracks repetition. If high-consequence and low-consequence look the same and appear as often, attention calibrates to the low-consequence frequency and the high-consequence is clicked through with it. Grading uncouples look, resistance and count: heavy gates fewer, heavier, instance forced visible; light gates can be non-blocking; no-gate goes to undo. Fatigue’s diagnosis is frequency; grading changes the structure of frequency and resistance, rather than adding words at the same frequency.

Grading must follow consequence, not implementation convenience. A thousand internal API calls should not be promoted to heavy; one outward payment must not be demoted to light because “we are used to it.”

Studying it

Hold the same task flow, compare a single confirm style versus three grades. Plant a high-consequence harmful item late. Dependent variables: stop rate on high-consequence, dwell on low-consequence (should near zero), whether people can say “which kind will actually stop me.” Independent variables: visual difference between grades, frequency of high-consequence, whether low-consequence still blocks.

If people cannot tell grades apart, grading has failed. Being able to say “payment will halt, tagging will not” is the grade existing in cognition.

Where it stops holding

When consequence itself cannot be scored (a new tool, uncertain whether outward), grade up, not down for ease. The main prescription for reversible acts is undo, not a light gate to pad the grading. A batch list is how a heavy gate shows its inside, not the grade itself. Fatigue as a phenomenon already has its diagnosis; this entry only gives the move that splits count from resistance.

Applying it

• Add a consequence-grade column to the tool table: none / light / heavy. Cap how often heavy appears per day; if the cap is hit, first check whether low-consequence was mis-promoted, rather than asking people to adapt.
• Heavy and light must differ in look and in operation: heavy’s default is veto or neutral, instance column cannot be empty; light does not block the main path.
• Check: in the graded flow, plant one payment and one tag. Payment should release only after the instance is read; tagging should not pop heavy. If dwell on the two grades is close, look has not yet split them. After taking low-consequence off heavy, high-consequence stop rate should rise — that is evidence grading is doing work.

Related

• Same group: L4.11.1 Confirmation should show the concrete objects that will be affected, not just the action category · L4.11.2 Batch confirmation needs an inspectable list; a count is not enough to judge · L4.11.4 Reversible acts are cheaper overall if post-hoc undo replaces pre-action confirm · L4.11.5 Outward acts should be treated as irreversible even if they can be deleted later
• Nearby: L4.07 Pre-action Confirmation · L1.09 Intervention Points in the Human Loop · L4.02 Automation Bias
• Search terms: confirmation fatigue · consequence grading · habit interruption