I've looked over the Bangladesh study (
Bangladesh Mask RCT), but it doesn't really seem like either a (1) very good study or (2) very good argument for masks.
(1) There's numerous methodological flaws with the study. The most important one is that they didn't use a baseline measurement of seroprevalence in the cohort populations. Without an established baseline you cannot measure change, so the findings of the study could easily be noise. We're supposed to assume that the criteria they used to randomize the populations had equal starting seroprevalence points, but we (and they) don't know that.
(2) The study doesn't demonstrate any effect for the under 50yo population. The study's results within the 95% CI can't rule out negative (for cloth) or no effect (for surgical). That's a pretty weak case.
We can go on and on and nitpick about it, but realistically I can't put any confidence in a study that's failing on both (1) and (2) here.
Nitpicking phase:
The mechanism by which seroprevalence is affected in an age-specific way is pretty unclear, and suggests to me that the more important thing here is not the masking element of the study, but the educational & enforcement aspect of the study leading to different behaviors. The study's seroprevalence measurement was only in symptomatic volunteers, and I don't think it's at all unlikely that people in the groups that were being instructed by local authorities (& paid for participation) knew the desirable outcome and thus this impacted reporting rates.
This isn't directly related to the study itself, but the results they are claiming are hard to square with real world data. The study discusses increasing mask usage from a baseline of 13% to 42%. In most developed countries mask usage for 2020-2021 was over 90% for months at a time. What we distinctly don't have is any demonstration in any locale that masks themselves were responsible for any kind of effect like the study claims. It's possible, since what the study is assessing is different from the normal measurement metrics (e.g. cases/hospitalization/deaths). We don't have any knowledge of how seroprevalence relates to metrics we actually care about (hospitalization/deaths).