The idea is to create a publicly accessible collection of an image of every single ballot, organized by the station from which they were collected. This can then be manually cross referenced, by anybody, against the reported vote count and result.
And yes, digital only ballots would pose a major problem here and in general.
The other practical issue with this concern is that buying or coercion would need to be of substantial scale, centralized, advertised, involved traceable transactions, and so on. And this is a criminal felony so the consequences for a leak are substantial. Ultimately, it just does not seem like a realistic attack vector.
So in many ways, I would consider this a feature more than a bug. You can personally verify that your ballot did indeed get counted (at least if you're willing to dig through thousands of ballots looking for your 'secret code') without allowing anybody else to see how you voted.
The problem is if the people casting the ballots can be identified by third parties. And in this regard, I don't understand what you mean or if that is what you are even talking about.
In particular, purely-digital systems are widely regarded to be too vulnerable to hacking to be safe. What you're describing sounds like DRE with VVPAT, which isn't considered to be a secure option, though it's better than DRE without VPAT since as you note you can in principle audit it. (However, note that most jurisdictions don't yet do RLAs to randomly audit, so right now digitizing, even with VPAT, could weaken the system.)
I think the basic idea here is -- if paper is secure and will be your fallback, and digital is insecure, you should just build your process to be optimized for paper-first, rather than digital-first. Digitizing as you suggest doesn't really gain anything over paper (except perhaps reporting provisional results faster, but you'd still need to do a risk-limiting audit to verify that your digital votes didn't get hacked so this might be a wash), but it does add more attack surfaces along the chain of custody.
Ultimately, paper is a very robust solution to the problem of making the system hard to subvert at scale; you can think of it as a sort of "proof of work", where it would be extremely difficult for, say, Russia, or the DNC / RNC to tamper with large quantities of ballots across the nation. Compare that robustness with a digital system, where IF it works you have the same properties... but around here we all know that almost all digital systems can be owned by a persistent enough adversary.
If you're willing to relax some of the requirements around refutability, there are some interesting e-voting schemes, for example you can do some cool stuff with homomorphic encryption like https://github.com/microsoft/electionguard/. But there is something to be said for having a tallying algorithm like "count the pieces of paper" that doesn't require a PhD to understand.