The point is that we aren’t comparing the generations to one another; we’re comparing them to modern day chickens. And we aren’t comparing by checking to see if a given creature “looks completely unlike a chicken.” That’s not how we differentiate species today, so why would we use it here?

As one example, one common and fairly simple definition of a species is a collection of individuals that can breed with one another and produce healthy offspring (“healthy,” in this instance also meaning that they must be able to produce offspring of their own). Obviously this doesn’t apply to bacteria or other things that reproduce asexually, but for our purposes, it could be sufficient. So you take this and turn it into a test: “Can this creature breed with modern-day chickens and produce healthy offspring?”

Now, even that simple question may involve qualifications in order to allow a binary answer. For example, maybe modern day chickens can breed with only 30% of other modern day chickens (of the opposite gender) and that number steadily decreases as we move back in time. The threshold for species differentiation here is going to be arbitrary.

That specific question is a bad choice in this instance, since chickens are descended from red junglefowl and can breed with them. In fact, they’re sometimes considered to be of the same species - for our purposes, we want to know when we first had a chicken - red junglefowl don’t qualify. As such, with chickens specifically it likely makes more sense to make the distinguishing criteria something that would differentiate a chicken from a red junglefowl, like “Is it domesticated?” That even gives us a good place to start looking - current understanding is that all modern chicken owe their origins to a single domestication event in Southeast Asia, roughly 8,000 years ago. Another option would be basing it off the DNA similarity to modern-day chickens (red junglefowl have 71-79% of the same DNA as modern chickens), e.g., once the DNA is no longer at least 80.000% the same, it’s no longer a chicken.

And you’re not limited to a single question, so long as the outcome of the test is binary.

Regardless of the specific test, at some point, the answer will change from “Yes, it is technically a chicken” to “No, it is technically not a chicken.”

For what it's worth it's possible to test the contents of an egg, but it's moot because it doesn't actually matter when we know. It exists independent of observation.

so true, it could be bees for all we know.