It'll be a win/win situation if baseless allegations of fraud are punished, and every detail of how voting machines work, including source code, are made public.
It's a debate we've had with secure messengers the benefit of having that many eyes on a piece of software outweighs the minor risk of a secret zero day.
The keys should be secure, but the design of algorithms and implementation of the system should not be.
If you hide away the source code, then only criminal actors and security agencies can evaluate security flaws. Open sourcing, at least, places legitimate security researchers on equal footing with these groups, so flaws can be disclosed publicly and rectified.
That being said, I think there are sufficient safeguards in the process that effectively mitigate any potential compromising of Dominion software. The software prints out the voters' ballot and the voter is given an opportunity to review the ballot before scanning it with a separate machine and it entering a locked box. This allows for a manual recount to guard against any possible shenanigans.
One additional step they could take, would be for the voter to verify their vote after scanning. This would ensure that they actually read the ballot. Though I could see that causing problems if the voter, for whatever reason, decided to repudiate their vote while travelling between the two machines.