A destructive action sharing its shape with a routine one is a high-risk configuration

What it is

The previous two leaves both dealt with "which of two similar objects got picked by mistake." This one flags a special case: when one of the two easily-confused objects is an irreversible or high-cost destructive action (permanent delete, bulk clear, an overwrite send) and the other is a low-cost routine one (save, draft, cancel), the configuration itself is high-risk regardless of how low the misselection probability happens to be — because it ties the residual risk that a description-similarity error can never be driven to zero directly to a bet with wildly unequal stakes.

Why it happens

The probability that a description-similarity error occurs depends on how discriminable the candidates are, and that probability can only be lowered by any single measure, never eliminated. When that lowered-but-still-present probability sits between two options with roughly equal consequences (two sort orders, say), the actual cost of a misselection is low and the user usually notices and corrects it quickly. But when the same probability sits between options as lopsided as "delete" and "save," the expected loss attached to that identical probability number gets amplified sharply — the probability hasn't changed, the stakes have. This is also why "looking the same" isn't the whole story on its own: the real risk multiplier is the combination of "looks the same" and "unequal consequences." Either factor alone doesn't add up to the scenario this leaf is warning about.

Where it stops holding

If the destructive action already carries an independent, sufficiently strong secondary confirmation, and that confirmation step itself presents distinguishing features, the actual risk of this lookalike configuration gets diluted substantially. What matters here is the combination of "looks the same" plus "no extra protection" — once a destructive action has already gained a dedicated extra layer of protection, the source of risk shifts to whether that layer itself is reliable, and away from surface similarity.

Applying it

When reviewing an interface, specifically pull out every control tied to a destructive, irreversible action and check, one by one, whether it stands in the same kind of similarity relationship discussed under description-similarity error with any routine-action control nearby, in either appearance or position. Wherever that relationship holds, flag it as a high-priority item regardless of how low the currently measured misselection rate looks — the fix can be visual differentiation through position, shape, and wording, a much larger spatial gap, or giving the destructive action an entirely different interaction path from the routine one (a long-press instead of a tap, say). Verification: for every destructive-action control, track separately how many times it has historically been immediately undone or complained about due to a mis-tap or misselection. Even if that count is a small fraction of the control's total taps, as long as the absolute number is non-zero and the consequence is irreversible, it should keep getting reinforced rather than shelved on the grounds that the rate looks low.

Related

• Same group: A10.04.2 mis-tapping between adjacent controls that look identical · A10.04.5 use position, shape, and wording together to widen discriminability
• Nearby: A10.08 error tolerance and graceful degradation
• Search terms: irreversible action · destructive action design · delete confirmation