This sounds like the perfect invitation for some old school over engineering.
I'm already having so much fun running through every possible input in my head, and I would inevitably write a serious mountain of steaming code to support it.
any time they enter and expression (auto complete), it they introduce a new one, they needed to add the range.
this did the job.
the time I spent the most was to sort the existing data and restore it in the new dictionary.
My main testing dataset is the 470,000 records from the Met, with 33k unique date values. Fortunately they include machine-readable dates I can validate against.
You have things like "in the Xth year of the reign of King Y", where we can easily relate multiple entries with different values for X, but don't actually know which CE years they correspond to. Even weirder is the Roman habit of recording "the year of the consulship of X and Y", which doesn't even allow you to relate any two different years at all without a reference table (which we don't have completely). And no, "years from the foundong of the city" wasn't a thing.
https://en.wikipedia.org/wiki/Varronian_chronology
> The Varronian chronology was adopted by the Roman state during the first century BC and gave rise to the traditional years ab urbe condita ("from the founding of the city"); most especially, those dates were used in monumental Augustan-era inscriptions, the fasti Capitolini and the fasti Triumphales.[40]
There are references to the Islamic and Japanese calendar systems, but always next to the CE equivalent.
Data entry is fortunately being done by modern people, so the translation to CE/BCE is usually baked in, and all you need to support is every possible way somebody could say "the early half of" and "5th millenium B.C. to mid 1914"