Then as each head of lettuce is picked, it goes into a crate that's securely sealed and then signed with the farmer's private key in a way that originates that crate of produce on the ledger. When the farmer sells their produce to the middleman that transaction is also recorded, and so on, all the way to the end consumer who buys the produce in a grocery store or in a restaurant. (And consumers would just use their phones or credit cards for this, rather than using any sort of external fob.)
Then when the first person gets sick they report their illness as per usual. Nothing happens at this stage, because there's no way to narrow down what made the person sick. By by the time the second person gets sick (with E. coli of the same genetic signature), now you can find the furthest place back in the supply chain where both people's purchases intercept. So you can now see if e.g. the contamination came from a single farm, and if so only recall lettuce from that farm rather than all romaine lettuce produced worldwide.
Because the database is open anyone can download a copy, and there is no risk of a single entity imposing a 30% Apple tax on each head of lettuce or whatever. And each person benefits from participating, because it's a pareto improvement in terms of their profitability. (Now their products only get recalled when they are at fault, rather than their products getting recalled when anyone is at fault.)
Trust scales with something like Metcalfe's Law. E.g. a consortium of ten independent banks is probably 99% less likely to steal my money than just Wells Fargo. The idea that we need millions of independent entities to get substantially better security than the status quo is just propaganda that gets spread by Bitcoin maximalists.
There is a cost of duplication, but at the level of duplication you actually need the cost isn't that much compared to the benefit.
I've seen lots of vague "blockchain for supply chains is a great idea" pronouncements but I don't get how it would actually work.