Badread | long read simulator that can imitate many types

Data in TTrees are compressed by baskets, so if a basket's compression is corrupted, no data can be read from that basket but all the other baskets are potentially fine.

Uproot's array-reading functions give up if any baskets raise an error, but you can use the more low-level TBranch.basket method to read baskets one by one, catching any exceptions along the way. Get a TBranch object from the TTree with dict-like access (e.g. mytree["branch_name"]) and call basket(i, ...) with the same arguments you'd pass to TTree.array but additionally with the basket number i. (They start at 0 and go up to but not including TBranch.numbaskets.)

There's also a TBranch.iterate_baskets, but it won't help you here because it would stop iterating when it hits an exception. You need to control the loop over baskets to wrap it in try-catch logic.

There's one more issue: you might need to correlate data from different branches, and their baskets might not begin and end at the same entry numbers. If you ask for TTree.clusters(branches_list) with the branches you're interested in, it will give you entry start and stop numbers at basket boundaries that are common to the set of branches you provide. Using these entry numbers as entrystart and entrystop in the normal TTree.arrays method would read only the requested baskets, and you can put try-catch logic around that.