The manuscript also includes practical analytical tips for controlling swap errors, as well as a new non-parametric method for estimating biases.

Let us know what you think of this work! 6/6

We think these findings - on the interactive effect of load and retention on variability and the dissociation between variability and bias - have important implications for understanding working memory limits. 5/6

Very surprisingly, stimulus-specific biases remained stable, unaffected by either load or retention. This contradicts recent prominent attractor-based models, which predict that biases should grow as noise accumulates. 4/6

Using the same analytical tools across all datasets, we consistently found an interactive effect of load and retention on variability: the impact of delay was amplified at higher memory loads - in other words, people forget faster when storing more information in their working memory. 3/6

To address this, we examined the effects of load and retention on two components of recall error: unsystematic variability and systematic bias. Our dataset included 6 new experiments and reanalysis of 7 published ones, spanning orientation, colour, and location tasks conducted across 5 labs. 2/6

Ah, great point - thanks! I hadn’t really thought of serial dependence or those contextual biases in this context. Will check those out and see how they map onto what we’re finding.

... delay-related changes in bias, I’m not convinced they actually increase (or decrease) - even though my initial intuition was that they should. We're finalizing that work now and can share it with you in the next couple of weeks - would be great to hear if you think we're missing something.