Link to study: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7587101/pdf/JCM.02107-20.pdf
This is a study that I wrote about when it was available as a preprint. I think it's worth revisiting now that the full paper has been published, as it provides a little more data about protective antibodies.
Back when the pandemic was young, there was a fishing boat departing from Seattle. In an attempt to prevent an outbreak on the boat, everyone was tested before boarding. Unfortunately, this was before we had fully realized that someone recently infected is not going to show up positive on a PCR test, and so one person who got on the boat was in the early stages of infection.
Unfortunately, the large majority of people on the boat caught covid. But fortunately for science, the pre-departure tests included taking blood samples to look for antibodies. Six of the 122 people on the boat had antibodies to the nucleocapsid protein before departure. Out of those six, three were infected. Because there was still a little bit of their earlier blood samples left, the lab was able to run more tests and determine that the three that were infected in spite of having prior covid did not have antibodies to the receptor-binding domain of spike. They had very low levels of antibodies to the whole spike, and no ability to neutralize virus through antibodies. The three with prior antibodies who did not get a second cases did have antibodies against the receptor-binding domain and also neutralizing titers.
Unfortunately there was not enough blood samples left to run the Roche / Labcorp test, but because it also looks for antibodies to the receptor-binding domain, presumably the three who tested negative for these antibodies in an in-house ELISA would also have tested negative via the Roche test. I would love to know what levels the three that were protected had, but we don't have that data. The lowest neutralizing titer of the three was 1:161, and I may be able to find a study somewhere correlating neutralization to the Roche test.