They had the original SAP to start with They failed the PFS primary endpoint so they tossed it.
They data dredged a new recurrent GBM arm once they were able to see if they had enough of a sample (only 64). and rewrote the new SAP after they saw the data.