One of the interesting solutions we discuss is that the unrestricted vs. forced-choice distinction is key.

We argue that incorporating aspects of natural behavior in VWM paradigms, can reveal a lot about how humans actually use their VWM.

Now for the less obvious beans: this pattern (longer view -> slower decay) does not show up in the typical (forced-choice) VWM paradigms: decay rates are independent from viewing time.

What might be up?

To spill the obvious beans first: memory performance got better with longer views, and it got worse with longer delays after viewing.
What's more: the LONGER a view was, the SLOWER performance got worse.

We had a bunch of people recreate example arrangements of funky shapes, however they wanted, while we tracked where they looked and what they did.