Information bound to a specific object is easier to associate correctly than information floating in view

What it is

A line of “torque 8 N·m” floats in view; three wrenches sit on the bench. Which wrench the number belongs to has to be guessed. Pin the same line on the handle of the wrench that should be used, and the association is done by place — no second round of “who is this number about.” Object-registered annotation is easier to join to its object than a head-locked or mid-view floating bar.

What is bound is “this information, this object,” not the general debate of whether an interface follows the head.

Why it happens

People translate spatial nearness into referential nearness: a label next to something is about that thing. Text floating in the view has no object neighbour; reference has to ride memory, an index, or a colour — slow channels that also collide with colours and numbers already in the real scene. A head turn keeps the floating bar in the middle of the view while the object slides to the edge, and the join breaks further. A label pinned to the object stays in the world with it; look away and back, and the words are still beside that wrench. The scene itself remembers the reference.

Binding the wrong object is more dangerous than no label: people will torque to a note stuck on the neighbour. Binding quality tracks recognition stability; drift or a wrong pick is the system pointing for you, wrongly. A floating bar is at least honest that it does not know who it belongs to — at the cost of extra search.

Studying it

A multi-object labelling task: three to ten similar objects, comparing a head-locked floating bar, world-locked but not on-object, and a label on the target. People “act on the correct one according to the annotation.”

Binding factors: on-object or not, object spacing, visual similarity, label drift. Scores: correct-object rate, neighbour mis-operations, time from seeing the text to reaching, time to find the object again after a head turn.

A one-object scene cannot show association error. There have to be confusable neighbours.

Where it stops holding

When the object is “this step” rather than a thing in the scene (wash your hands now, wait now), there is nothing to pin to, and floating or body-lock is a better fit. Far field, shelves so dense that labels overlap, on-object becomes a clot of type; point first, then expand. On recognition failure, pinning to the wrong object is worse than falling back to floating; revert to “please point at the target” rather than guessing the nearest surface. Locking modes can mix: a summary body-locked, detail pinned to the object. Do not make every word on-object.

Applying it

• Names, values, and warnings that have a clear object default to sitting on that object, with a short leader to a recognizable part. Do not pile them in a mid-view status bar.
• If the object is not yet recognized, finish binding with a point, then show the content. Do not display unbound text first.
• If drift during a bind exceeds the object’s own scale, take the label off and report “cannot register.” Do not let the words slide between neighbours.
• How to check: three similar wrenches, a torque number for only one. Ask “which do you turn” under a floating bar, then again with the number on the handle. Neighbour turns are evidence that place did not carry the association.

Related

• Same group：N5.12.1 Overlays should show and hide with task relevance, not stay resident · N5.12.3 Handling the real environment and overlays at once costs extra attention; they compete for bandwidth · N5.12.4 When the environment is already information-rich, overlays must reduce, not add · N5.12.5 Density control aims at task completion, not at showing as much available information as possible
• Nearby：N5.07 Overlay Density · N5.13 Safety Risks from Occluding Reality · N3.08 World-locked, Body-locked, and Head-locked
• Search terms：object-registered annotation · world-locked label · referential binding