OK, I now pumped data from that table to an table without any constraints:

CREATE TABLE TAB_GRAAFIK_TEST
(
UID INTEGER,
VAHETUS INTEGER,
TOOLEPING INTEGER,
MUSTAND INTEGER,
KUUPAEV DOM_KUUPAEV,
CHANGED DOM_TIMESTAMP
);

and then tried

select count(uid), uid from tab_graafik_test group by uid order by 1 desc

and it appears that there is indeed 2 uid which aren't unique!

UID VAHETUS TOOLEPING MUSTAND KUUPAEV CHANGED
==== ======= ========= ======= ======= ========
139972 1 24 08.08.2007 26.06.2007
11:12:00
139972 1 24 08.08.2007 26.06.2007
11:16:45
139973 1 24 09.08.2007 26.06.2007
11:12:04
139973 1 24 09.08.2007 26.06.2007
11:16:49

Note that only difference is in "Changed" field.
However, when running this query against original table (TAB_Graafik),
there is no dublicates! So it seems that data from an old transaction
is somehow has survived, but it is visible in only in some special
condition?

Database header page information:
Flags 0
Checksum 12345
Generation 13048450
Page size 4096
ODS version 10.1
Oldest transaction 13048431
Oldest active 13048432
Oldest snapshot 13048432
Next transaction 13048433
Bumped transaction 1
Sequence number 0
Next attachment ID 0
Implementation ID 16
Shadow count 0
Page buffers 0
Next header page 0
Database dialect 3
Creation date Jul 5, 2006 12:07:26


TIA
ain