Showing "them" would be pure social validation and, according to your parent, not recommended. Showing "that" would be sanity protection, just as you understand it.
Feynman is clearly saying disregard others while you are doing the work, not disregard others when you present it for feedback and criticism. That is the spirit with which I responded.
As a developer, there is a third option where your work is simply ignored. If it is functional for you and you are satisfied with your own work, then sure, feel free to disregard others lack of interest, like an artist who values their own work but meets a deafening silence. But please don't disregard serious criticism.
Could there be an algorithmic reason for this surprising behaviour, or is this a manual intervention?
Very little. The chosen primitives are naturally resistant to timing attacks, thanks to the utter absence of secret dependent branches and secret dependent indices. I did have doubts about constant time comparison (code generated under -O3 was crazy), which I solved by changing the API.
Locking the memory is not implemented, only discussed in the manual (Monocypher wants to stay portable).
Zeroing out memory is neither implemented nor discussed. I'm not sure what to make of it, actually. If it is useful, then wiping the contexts is not enough, you likely also have to wipe temporary buffers above the top of the stack (I don't think anybody does this). In any case, if the attacker can read your memory, you're probably screwed anyway.
Now if some attacker can read your memory only after you've processed your secret stuff, a `crypto_wipe()` function would have its uses. But if you've processed secret stuff once, you're likely to do it again, and then crypto_wipe won't save you.
I'm also afraid zeroing out memory gives a false sense of security. I'm more in favour of segregating the sensitive stuff in a separate process, like Qmail.
Shameless plug: I've written my own crypto library¹. Not just for the lulz, I fully intend to use it in production. To do that, I had to disregard the crypto community, that basically says anyone who does that deserves to burn in Crypto Hell the time it takes them to count to 2^256. (I might exaggerate a tad).
They say that for a reason however. I had to seek and listen to external advice eventually. Which, judging by the holes they poked through my library, was invaluable.
I think there are two phases. The "I'll show them" phase where you hide under your cave doing your thing, and the phase where you actually show some results and confirm whether this was a worthwhile endeavour.
[1]: http://loup-vaillant.fr/projects/monocypher