I did go back and look at 38'30 in his presentation, and you are right that he indicated that the SAP has to be etched in stone before you "look at the data".
However, at 36'20 MB indicates that in parallel there is a lot of activity going on. They have a CRO & "they hold all the data", "in a separate database that only they have access to", and they are continuously checking and validating the data".
So my current thought is that the data that the CRO is reviewing (blinded to NWBO) is the data that is produced as a result of the Soft Lock. Then the CRO and auditors go to work on cleaning up this data, and doing their queries and verifications etc .... in parallel (or simultaneous with) ... whilst NWBO is working on the draft SAP & working to get it finalized (without being influenced by the soft lock data that only the CRO has access to).
Bosch's meaning is that they will UNBLIND after data lock. They have LOOKED at the data, but not unblinded it. They are scrubbing the BLINDED data, and when the SAP is ETCHED IN STONE, they will unblind it.
Soft data lock occurs in January 2019. Then they gather it and scrub it and answer queries and examine their navels or whatever else this tedious business entails while they wait for the SAP to be FINALIZED. And when the SAP is finalized, it's "etched in stone" and they can THEN unblind the data. But the data that has been scrubbed and that will be what forms the content of the hazard ratios and p-values and mid-points and k-m curve possibilities of survival, etc., etc., will be based on what happened previous to the soft lock. What happens between the soft lock and the formal data lock will be additional data that won't be lost, but won't be part of what makes up the formal unblinded trial data.
This is how I understand it and what I think DI is saying on his phone.