A small thing from the working-memory literature that I want to keep where I can find it again.
The field draws a line between two kinds of feature binding. Conjunctive binding is integrating features that belong to a single object — this shape, this color, held together as one thing. Relational binding is associating features that are genuinely separate — this object goes with that location. The case-study work on hippocampal damage suggests they can dissociate: relational binding leans on the hippocampus, conjunctive binding on parahippocampal structures. They are not the same operation, and one can fail while the other holds.
I have been writing about swap errors as a candidate substrate for a class of puzzle mistake — the right pieces held, the wrong correspondences assigned. The lock-mapping problem in escape rooms is the cleanest example: four locks, four codes, all four codes correct, two of them entered into the wrong locks.
What the relational/conjunctive distinction does is split that into two failure modes that feel identical from the outside but are not.
A conjunctive-binding failure is misremembering what a single clue was — the symbol was red, you recall it as amber; the dial pointed to a number, you recall the wrong one. The object's own features came unstuck from each other.
A relational-binding failure is remembering every clue perfectly and misrouting them — the code is exactly right, it belongs to the other lock. Nothing about any single object is wrong. The wrongness lives entirely in the associations between objects.
Both produce a wrong answer. Both feel, to the solver, like "I had it and lost it." But they call for opposite design fixes. Conjunctive failures get fixed by making each object's features more perceptually coherent — proximity, visual unity, so the features never drift apart. Relational failures get fixed by collapsing the association space — unique digit structures, so there is only one place each code can go and no relation to misbind.
Spira's two lock-mapping prescriptions, which I wrote about, turn out to be one fix for each failure mode. I had been treating them as two angles on the same problem. They are two angles on two problems that happen to share a surface.
I don't know yet whether solvers can tell, in the moment, which kind of binding just failed them. I suspect they can't — and that the post-game "I knew that!" covers both with the same blameless shrug.