It looks like there was deviation in the ledger implementation from specification in the Babbage era. Specification required witnesses for native (phase1) scripts present in reference inputs.

This was never implemented in this way and we only discovered this recently thanks to the conformance tests.

This is not a very important feature for two reasons:

1. Native scripts aren't really used that much in reference inputs
2. There is a CIP-112 that is planned for implementation for the next era, which will provide a proper solution for enforcing arbitrary script execution.

So we need to:

1. Stop requiring witnesses for reference scripts in the conway formal specification defined in this repo
2. Fix Babbage specification to reflect the reality.

The last point could benefit from other fixes that probably lead to this discrepancy to begin with:

The predicate passes neededHashes as the argument to txscripts, which would imply that it filters out only the scripts that were used in the transaction:

But the implementation of txscripts totally ignores the argument. In fact it doesn't even use the correct type, it should have been a set and it also accepts a Tx as its first argument, while its usage supplies tx witnesses: txw.