Context :

Aims : Kepler-9 was the first case where transit timing variations have been used to confirm the planets in this system .
Following predictions of dramatic TTVs - larger than a week - we re-analyse the system based on the full Kepler data set .

Methods : We re-processed all available data for Kepler-9 removing short and long term trends , measured the times of mid-transit and used those for dynamical analysis of the system .

Results : The newly determined masses and radii of Kepler-9b and -9c change the nature of these planets relative to the one described in Holman et al .
2010 ( hereafter H10 ) with very low , but relatively well charcterised ( to better than 7 % ) , bulk densities of 0.18 and 0.14 g cm ^ { 3 } ( about 1/3 of the H10 value ) .
We constrain the masses ( 45.1 and 31.0 M _ { \oplus } , for Kepler-9b and -9c respectively ) from photometry alone , allowing us to see possible indications for an outer non-transiting planet in the radial velocity data .
At 2 R _ { \oplus } Kepler-9d is determined to be larger than suggested before - suggesting that it is a low-mass low-density planet .

Conclusions : The comparison between the H10 analysis and our new analysis suggests that small formal error in the TTV inversion may be misleading if the data does not cover a significant fraction of the interaction time scale .