Hi Ralf, I’m disturbed by odd results on REINDEXing against the fake RMNOCASE collation (a renamed NOCASE) extension you made for me. In a sample database, running: REINDEX; PRAGMA INTEGRITY_CHECK; 1.9.1 returns 19 rowids missing from 3 indexes 1.9.3 returns 32 rowids missing from 4 indexes If I REINDEX using one version and do the INTEGRITY_CHECK with the other: 1.9.1 reports 32 ... 1.9.3 reports 19 ... So it looks like REINDEXing behaves differently on the two versions. I received a CDECL type extension of a RMNOCASE-like collation that works with the command line shell sqlite3.exe and with another Delphi-based SQLite manager but not with SQLIteSpy. REINDEXing with either of these or with the original application results in OK from INTEGRITY_CHECK on any of the three. REINDEXing with the original application: 1.9.1 & 1.9.3 report 20 rowids missing from 4 indexes So: 1. Why does a REINDEX by one version of SQLiteSpy result in a different set of results from INTEGRITY_CHECK compared to REINDEXing by another version? Given the same collation sequence, they should be the same. 2. Why should a REINDEX result in any missing rowids from indexes when INTEGRITY_CHECK is executed by the same SQLiteSpy version? Tom