A shared prefix between a frequent and a rare path amplifies capture

What it is

The strength of a capture error isn't fixed — it's set jointly by how long a prefix two sequences share and by the frequency ratio between the high-frequency and low-frequency sequence. The more steps they share, and the more lopsided their relative frequency, the more likely the low-frequency sequence is to get hijacked at the branch point. This claim isn't about whether capture happens at all — it's about which architectural design choices make the risk of capture larger or smaller.

Why it happens

Every step in a shared prefix is effectively laying down reinforcement for whichever path runs more often, because most people passing through those steps are on the high-frequency task rather than the rare one — those steps get trained by the high-frequency path far more times than by the low-frequency one, even though the two sequences are, on paper, performing the identical operation at that point. The longer the shared prefix, the more this training imbalance accumulates; by the time the branch point arrives, the "trace" each path has left up to that step is badly lopsided, and because the physical actions up to that point were completely identical, the user has had no mid-sequence cue at all to remind them which path they're actually on.

Where it stops holding

This amplification effect only holds when the shared steps are literally identical operations — the same gesture, the same screen, the same button. If the two paths merely look similar but each step carries a slight difference in the actual operation, the "shared training" premise breaks down, because each step now carries its own motor or visual cue specific to its path, and the strength of capture drops accordingly.

Applying it

When designing information architecture or menus, first map the operation tree for core tasks and mark, at every branch point, the relative usage frequency on each side and the number of shared steps leading up to it. Branch points with a lopsided frequency ratio (an order of magnitude or more) and mismatched consequences should be flagged as high-risk, and the entry point for the rare, high-consequence action should be pulled out entirely rather than left to branch off the same starting point as the frequent, low-consequence one. If shared entry is unavoidable for other reasons, move the branch point as early as possible to shorten how many steps the two paths share, reducing how much the rare path gets assimilated into the frequent one's training. Verification: for the high-risk branch points flagged on the operation tree, sample the error logs for the corresponding rare action and check whether the erroneous outcomes cluster around what the shared-prefix high-frequency path would produce. If they do, the risk rating for that branch point is confirmed and should be prioritized.

Related

• Same group: A10.05.1 A familiar sequence hijacks the current intention · A10.05.3 Old habits carrying over after a redesign · A10.05.4 Post-change transition protection instead of a one-shot cutover
• Nearby: A10.04 Description-similarity errors
• Search terms: shared prefix · menu architecture risk · capture error