https://twitter.com/JenniferNuzzo/status/1621881097452666883
Jennifer Nuzzo has a solid thread providing context. I could go into it more in general about Cochrane, but their approach is very conservative and it tends to suffer as a result when the evidence is diverse and not high quality (i.e. when they say something works that's generally true, when they say something doesn't work that's not necessarily the case).
For masks, as Nuzzo notes, the bulk of the RCT data comes from the pre-COVID era. And most importantly has variable endpoints, low adherence, and high risk of bias (particularly around blinding unsurprisingly). This all makes a meta-analysis difficult as you are comparing apples to oranges.
Of the post-COVID era, what we generally are seeing is that masks work, but the effect may not be substantial. But again adherence is the key. Larger effects are seen with better adherence and better masks. And unfortunately most of the data wasn't captured in this Cochrane review as it was limited to only RCTs (we only have a couple completed mask RCTs in the post-COVID era). Non-RCT data isn't useless and in some cases may have advantages in the context of masking ( e.g. hard to blind a RCT with masking, whereas a mechanistic study can show a direct pathway for efficacy).
There's been controversy with Cochrane over the years on a few matters. I think as long as you understand the limitations to their approach and the context of their approach it is fine.
The pre/post COVID distinction is highly relevant in this case. SARS-COV-2 is believed to have a not insignificant airborne aerosol component to transmission. So mask adherence is especially important. In a healthcare setting (where a lot of the RCT data in the pre-COVID era came from) that could be the difference between not getting exposed and getting exposed for a healthcare worker if they aren't wearing masks consistently and uniformly- e.g. only around patients/or during high aerosol generating procedures.
This isn't a knock on Cochrane. The review as much says all these issues with the data. They are doing the best they can with the data in regards to their usual approach. It just makes interpretation of their review important.