Thanks Philip and Doug !
See attached my simplified survey TestUnselectIf.xlsx (it is in French but I think you can understand the scope). I think my workaround looks like what you are suggesting. I added a field (inf_Confirmer) asking the user to confirm the choice made in the first question (inf_Taches). If the choice 'Confirmer' is selected in this second question (this is the only possible choice), then the first question (inf_Taches) becomes read-only with the formula if(selected(${inf_Confirmer},'1'),'yes',''). In addition, as long as the confirmation is not given by the user to the second question (inf_Confirmer), all groups corresponding to each selected section remain read-only, by the formula if(selected(${inf_Confirmer},'1'),'','yes') applied to each group, preventing the user from entering data as long as there is a danger of accidentally deleting the groups.
With this solution there is still a small danger that the user uncheck the 'Confirmer' choice and then the first question becomes editable again.
Doug, I wasn't aware of the 'once' function. I tried it instead of the 'if' function in the readonly column of the 'inf_Taches' field so that it would become read-only as soon as the 'inf_Confirmer' question is answered (see TestUnselectOnce.xlsx). In this case, the 'inf_Taches' question remains read-only forever. I think that it will be a good solution: indeed, if the user has not made the right choice and has already confirmed his choice, he can simply delete the survey since this is the first question to answer, he will not lose his work.
If you have any other solution let me know, but I think that these solutions are viable to our team.
Thanks again !