No, you're not better off. If you want to get a medical product like this licensed, you have to prove that you performed rigorous, multi-staged testing and document all your development including all emerging risks. I have participated in such licensing efforts and I doubt that an open source project has the means of providing such rigorous testing.

The result of this diligent process, of course, is how a broken CRC16 routine got shipped in this medical product.

It's the most trivial thing. Copy a public domain CRC16 routine, add a unit test with a test vector.

Can you be sure it wasn’t a badly implemented form of obsfucatiom?

It certainly slowed down the reverse engineers. If they didn’t get to the object code what would the next step have been? Cryptographic analysis?

5 of the bits were never set in their "obfuscated" variant? If you want to obfuscate CRC16 you would just choose a randomized starting value.

I didn't say it was implemented well. Perhaps they should changing shift operators would quietly change the values without any disturbance to checksum integrity.

Tidepool is indeed working on FDA approval for Loop.

https://www.tidepool.org/blog/tidepool-delivering-loop