Yes you are missing something.
The way Trezor works is that the untrusted host computer provides the Trezor wallet with every transaction that the to-be-signed transaction's inputs spend. All transactions refer to transaction inputs by a secure cryptographic hash, the transaction id. Thus it is impossible for the host computer to hide what transaction inputs are in fact being signed by the wallet - the worst the host computer could do is have the wallet harmlessly sign a completely invalid transaction.
Ahh, of course - I was overcomplicating things in my mind. Thanks for setting me straight. My test harness was making up these previous hashes for the purpose of tx signature testing, so I mentally wrote them off as arbitrary.
If my device is presented with the previous transactions for addresses A, B, and C (which include their total value) I can simply SHA256 these, make sure the hashes in the outpoint structures for the tx I'm going to sign match, and then I know the total alleged value for A, B, C and can present outputs and fees to the user. As you said before, the worst thing the SW feeding me stuff to sign can do is make me create a bogus transaction that will be rejected by the network. A bit of a hassle, but at least its just doing a streaming calculation of a SHA256 (not ECDSA thank goodness!) and no blockchain data is necessary. I feel alot better now, back to making progress

However, life as a HW wallet designer would still be much better with the proposed addition.