Thanks for your comments. I did look quite a bit at all those ideas I wrote about and couldn't find anything yet.
It's almost an "aesthetic" argument - I see the length of the ciphertext is 98 characters, which is 7 rows of a seriated digraphic cipher with width 7.
I still think the number of repeated digraphs at width 21 is a key observation, or indicated something which is not a coincidence...
For Trifid, I think I've gone over it fairly thoroughly, personally, but don't let that dissuade you. I could have missed something obvious. I think that's a lesson for me from the IRA cipher and Thouless cipher experience. The last time people tried to solve those, the tools were not as good as they are now ...
I used to wonder if "...ID BY ROWS" meant "TRIFID BY ROWS" (text missing from the start, like "T IS YOUR POSITION", confirmed by Sanborn to be "WHAT ...").
To answer the question "how much Trifid ciphertext do you need to find the plaintext without a crib", it's in the same ballpark as, say, Playfair. The unicity distance of Playfair is 22-27 letters - 25! different keys, assuming you have I/J combined. With Trifid you have 27! different keys, plus a period.
If you have 97/98 letters/characters, and 20 known plaintext letters, then it almost uniquely determines the key and the period. You could use "simulated annealing" to solve it at that length, without a crib - it's been described by Michael J. Cowan (ANCHISES) in The Cryptogram.
For more detailed description of unicity distance, there's an old article by Cipher Deavours in Cryptologia (1977) and a similar shorter version on James Lyons' (fellow Brisbanite) Practical Cryptography website.
It also might be fun to use code like Bion's trifid crib dragger to see what periods, if any, work with the known plaintext/ciphertext.