Put the experiment in temporal order and the interesting part is where the measurement sits. The signal that predicts overnight strengthening is recorded while the person is awake and learning [3], so the variance worth explaining exists before the lights go out. That reframes the search: instead of hunting through sleep physiology for the sorting rule, you look at encoding for a marker and treat the night as the machinery that acts on it. Sleep's role in stabilising new learning was already well established [11]; a pre-sleep predictor of which items get the benefit is the new part.
What the press summary does not give you is arithmetic. No sample size, no effect size, and no description of a comparison group that stayed awake across the same interval [7]. That third absence is the one I would want closed first. A trial with strong theta at encoding is plausibly just a trial that was encoded well, and well-encoded items are recalled better whether or not the intervening hours contained sleep. Without the wake comparison, "preferentially strengthened during sleep" is hard to separate from "learned properly in the first place". The full PLOS Biology paper may carry that contrast [2]; the release does not say.
Then there is the verb. Lead author Dan Denis describes the result as a signature of neural activity that instructs the brain which experiences to process during sleep [5], and the team calls theta strength an instructional cue that organises sleeping activity [4]. A correlation between encoding theta and later consolidation supports the weaker reading, that theta indexes whatever does the prioritising. Instruction is a stronger claim, and the way to earn it is to move theta and watch the memory benefit move with it. Nothing in the material describes that manipulation.
The clinical extension deserves the same discipline. Denis notes that people with depression tend to over-allocate attention to negative information while downplaying positive experiences [8], and suggests the findings may open avenues for understanding and treating such conditions [9]; the researchers also say disrupted filtering could contribute to psychiatric illness [12]. What was actually tested was word-picture associations in a lab [1]. Between that and a mechanism of depressive memory bias sit several experiments involving emotionally valenced material and clinical samples, none of them reported here.
The capacity argument underneath all this is old and sound: remembering everything would overload the system, so something has to triage [10]. The contribution on offer is a candidate marker of that triage, timed at the moment of experience rather than inferred from what survived the night.
