Personalized recommended items appearing and disappearing interrupt spatial memory

What it is

A “workspace recommended for you” appears today, vanishes tomorrow, and is replaced by something else the day after; the nav’s length and neighbourhoods move with it. Coming and going recommended items interrupt spatial memory. This is not reordering among a fixed membership. The array gains or loses a member, and every downstream slot shifts. “Settings is third” becomes fourth after one insertion, even if Settings was never named by the ranker.

Insert/remove hurts more than reorder because even the membership set is unstable. After a reorder the names can still be hunted; removing an item makes people think a feature shipped away.

Why it happens

Spatial memory encodes relative positions. Insert a slot, and everything to the right or below gets new coordinates; the whole memory voids together. Recommenders mint members per session, so the nav’s cardinality becomes a random variable. Visually it is still one nav bar; motor memory no longer maps.

A vanished recommendation that looked like an ordinary class updates the mental model as “the class was deleted,” and people may never look for the feature that still lives elsewhere. An inserted item that occupies a familiar slot takes the first point: either a misclick or a new “correct path,” both of which scramble learning of the original array.

Studying it

On a learned nav, insert or delete a recommended slot and measure pointing at the other items.

• Paradigms: five fixed items to asymptote; test phase inserts a recommendation at position 2, or removes a recommendation that had been present. Contrast with “recommendation in a module outside the array.”
• Independent variables: insert position (head / middle / tail), whether the recommendation looks like an ordinary class, whether removal leaves a gap.
• Dependent variables: pointing error on items the recommender did not name, reports that “Settings was deleted,” misclicks on the inserted item.
• Methodological note: measuring only the recommendation’s own click-through will conclude insertion “worked.” Measure harm to neighbours. Leaving a gap versus closing it are different translations; split them. Day-to-day insert/remove hurts more than within-session, because memory consolidated overnight is hit by a new cardinality the next day.

Where it stops holding

Legitimate add/remove from notices or permissions (a module turned on, a campaign retired) is not recommendation, but it still shifts slots and should use the same stability rule: do not insert in the middle of a learned array. A “today’s picks” block on a media home, if clearly a content stream and not a nav slot, was never a target of spatial memory. A user-pinned item that the system withdraws is a broken promise, more serious than recommendation jitter.

Applying it

• Do not insert recommendations into a learned primary array. Put them in a module labelled Recommended; jitter stays inside the module.
• If they must appear in the nav strip, fix one Recommended slot and change content inside it; do not add or remove the slot.
• Verify: after five items are learned, insert a recommendation. Points at unchanged items such as Settings should still land on the old coordinates. If the whole row translates, or people ask whether Settings is gone, insert/remove has already interrupted spatial memory.

Related

• Within the group: G2.10.1 Reordering nav items by frequency unfixes their positions and hurts predictability · G2.10.3 Stable core nav and personalized modules must be partitioned · G2.10.4 Personalized ranking belongs on secondary entrances, not primary-task nav · G2.10.5 People need an explicit way to turn personalization off and restore a fixed order
• Adjacent: G2.07 Navigation consistency · E5.16 Shortcuts and pinned items · G2.06 Global and local navigation
• Search terms: spatial memory · personalized nav insertion · recommended shortcuts