Hi! I think the explanation is: my previous tests were on an index of a db table of 300 documents; small index = 1 segment and doc at index level is equal to docID at LeafReaderContext. Anyway it is not recommanded to use doc as id of document, it is reserved for Lucene use. Next step, my index was of a 30000 documents and, then, I have 9 segments and doc is not equal to docID anymore. And it was my processing to set some score equal to 0 based on misleading docID in the CustomScoreProvider. Therefore non repetitive errors depending on delete, insert, update of documents or of reindexing. And therefore my confusion.
Claude Lepère claudelep...@gmail.com