In order to see the reliability of the online Mbta analysis, I rerun Mbta offline for the Hardware Injections performed in Virgo-Handford-Livingston during WRS13. The job uses the same grid parameters for all injections:

Analysis duration: 810 seconds

MBTA_MASSES  1.  34. 0. 200.

 MBTA_MTOT_MAX  35.

In total, 26 injections were done during WRS13 and the analysis result is given in the two first images.

In green lines we focuses our attention on injections which were found in the offline analysis and not in the online one or not analysed. 7 Injections correspond to this case. 2 injections are missed online due to a SNR too close to the Threshold (6), the 2 injections not analysed have been found "manually" in the MbtaOnlineHLV file. For the first case, the job which should have detect the event was crashed and for the second one, the job is ok and then the "non-detection" isn't understood. The last three injections reveal a Coincidence triggering failure: when an event is found in several detectors (double coincidence here), the individuals triggers aren’t all maintained and then missing in the event extraction. The missing events parameters have been extracted manually and added in bold type.

The red lines shows cases not found offline. Four cases correspond to bad data quality (<8) or missing data. We recalculate the expected SNR (noted updated) with the measured Horizon during these injections. Among the 7 cases remaining, 4 have an expected SNR inferior to the Threshold , one other was not found online but found manually in MbtaOnlineHLV file due to the same problem as before (lost of individual trigger in a double coincidence). The last two situations are not yet explained:
-for the same analysis parameters, the event is found online and not offline.
-the triple coincidence is not found online, and we check with FrDump command without success.





























The histogram below represents the distribution of the SNR difference between online detection and offline one.  The SNR RMS value is then estimated to 0.5 and gives a feeling of the SNR estimation. 










In conclusion among 26 injections, 9 were found online and 15 offline. The online/offline difference is due to the coincidence triggering labelling not present offline and events detected offline with a SNR very close to the Threshold. The “non detected” events correspond mainly to low expected SNR, bad data quality or missing data and again the coincidence triggering labelling. However one case(triple coincidence in dark red) has no explanation.