Honestly, I can't think of any reason to retain the current behaviour, nor how anybody could be relying on it. Sorry to disappoint. ;-) Anyhow, I compiled a version patched as follows: ``` --- src/shell.c.in +++ src/shell.c.in @@ -2913,11 +2913,11 @@ sqlite3_db_config(p->db, SQLITE_DBCONFIG_WRITABLE_SCHEMA, -1, &wrSchema); sqlite3_db_config(p->db, SQLITE_DBCONFIG_WRITABLE_SCHEMA, 1, 0); sqlite3_exec(p->db, "CREATE TABLE IF NOT EXISTS temp.sqlite_parameters(\n" " key TEXT PRIMARY KEY,\n" - " value ANY\n" + " value\n" ") WITHOUT ROWID;", 0, 0, 0); sqlite3_db_config(p->db, SQLITE_DBCONFIG_WRITABLE_SCHEMA, wrSchema, 0); sqlite3_db_config(p->db, SQLITE_DBCONFIG_DEFENSIVE, defensiveMode, 0); } ``` And it does not show the anomaly. Marginally related: Since I haven't found any documentation for how dot commands are interpreted, I consulted the source. As far as I can figure it out, the rules are as follows: You may quote an argument with single or double quotes. If double quotes, backslash escapes are interpreted. If not quoted, the argument ends at the next space. If quoted, the quotes are stripped, and the argument is the result. If the command is `.parameters set`, the value argument is first put verbatim into a `REPLACE INTO sqlite_parameters` statement. If that fails, the statement is run again, but now with the value argument quoted as a string. As a result the following dot commands will all insert the integer 7: ``` .param set :x 007 .param set :x '007' .param set :x "007" ``` while the command ``` .param set :x "'007'" ``` will indeed insert the string `'007'` (assuming the patch above). This is pretty hard to guess, so it might be good to document it.