Algorithmically, this is very easy to account for. If you want to actually break algorithmic detection, the letter replacement needs to be dynamic.

For example, at the beginning of the post, declare a decoding key, something like swap S and F, or "[ = T". Best to use replacements that are not visually similar. It's inelegant, but it will work to bypass automatic detection, whereas any characters that are visually similar, Facebook will eventually add to their algorithm because they can be applied generally.

A possible compromise would be to use acronyms or code words for the obvious key words they'd be looking for, and then define them above or below using the dynamic letter swapping.

Something like:

Spill = kαel (in reverse); Doc = director of HIN (in reverse); GF = gαιn οf nοιtcηuf (in reverse);

Τhe Ιαb-spιll deηιers αre beιng expοsed αs deΙυsιοηaΙ οr eνen deΙιberαte Ιιαrs. Τhe eνιdence ιs cΙeαr. Wheτher τhe spιll was αccιdeηtaΙ οr deΙιberατe, τhe fαcτ remαιηs τhat Doc fυηded GF reseαrch το creατe α pοτeηtιαΙΙy ΙeτhαΙ ageητ, whιch hαs kιΙΙed mιΙΙιοηs. Ηe αηd his cοhοrτs deserνe το be pυηιshed.