6.2.1 Input data from the upstream processing

The events producing a discontinuity in the wavelength calibration zero-point are defined as breakpoints. In the Gaia DR3 dataset, which includes the data acquired between 25 July 2014 (OBMT 1078.4; see Section 1.3.1 for the definition of OBMT) and 28 May 2017 (OBMT 5230.1), the breakpoints are the following: OBMT 1192.13 EPSL/NSL transition; OBMT 1317, 2330.6 and OBMT 4112.7 decontaminations (Section 1.3.3); OBMT 1443.9 and OBMT 2574.5 refocus. Then, in addition to these physical breakpoints, there are 3 additional breakpoints: OBMT 3904 and 5084 (related to a discontinuity in OGA3 due to a jump in the basic angles) and OBMT 4513.16, due to a discontinuity in OGA3 related to the change in the ELSF calibration in the Initial Data Update (IDU), which is expected to go away in DR4, when an improved AGIS version will be used. These 9 breakpoints are used to separate the dataset into the 10 trending epochs shown in Figure 6.2. The epochs separated by these last three breakpoints are named with suffix B and C.

The effective time covered by the trending epochs are reduced because of the time intervals where the data known to be of bad quality are removed. In particular, the duration of TE2, TE4 and TE6A are significantly reduced because of large BadDataIntervals, due to the cool-down time of the payload after the decontaminations (Figure 6.2).

The bad data intervals used by the RVS pipeline include the time during decontaminations, the refocusing, the commissioning of a new version of the on-board VPU (Section 1.3.3), the time intervals where the AGIS residuals are high, the gaps in the ObjectLogs, and are available for each CCD and FoV via the web page: https://www.cosmos.esa.int/web/gaia/dr3-data-gaps.

The filters applied to the input data and during Pre-processing are of two types: at source-level and at transit-level. The filters at source-level remove all the transits of the source, while the filters at transit-level remove only the relevant transits. The following filters are applied to the input data:

• •

  Poor or not existing AGIS coordinates (the good astrometric positions provided by AGIS are indispensable for wavelength calibration and thus for the radial velocity estimation): The sources are not treated by the pipeline if 1) the source coordinates are not from AGIS 3.1, or AGIS did not converge; 2) the source is flagged as duplicated by AGIS; 3) there are less than 5 observations used by AGIS; 4) the source has large astrometric excess noise astrometric_excess_noise $>20$ mas; 5) astrometric_sigma5d_max $>100$ mas. This filter is at source-level.

• •

  Bad field angles: the spectra for which the uncertainty of the field angle $\eta$ are $>$ 200 mas are filtered out. This filter is at transit-level.

• •

  Bad data intervals: The spectra acquired during the bad data intervals (Section 6.2.1) are excluded. This filter is at transit-level.

• •

  $G_{\mathrm{RVS}}^{\mathrm{ext}}$ (Section 6.1.2): The filters on $G_{\mathrm{RVS}}^{\mathrm{ext}}$ depend on the pipeline (see Figure 6.1). In the Scatter pipeline the border of the spectra of the faint sources are used, then the sources with are filtered out. In the Calibration pipeline, instead, only the bright sources are used, and the sources with $G_{\mathrm{RVS}}^{\mathrm{ext}}>10$ are filtered out. In the FullExtraction pipeline, to make deblending possible, there is no filter for the input sources: however, after the deblending has been applied, the sources with $G_{\mathrm{RVS}}^{\mathrm{ext}}>14$ are filtered out.

• •

  Truncated windows: The filters on the truncated windows depend on the pipeline. In the Scatter pipeline all the truncated windows and also the rectangular windows with AC size are filtered out (to avoid blended spectra). In the Calibration pipeline the truncated windows are filtered out, but the rectangular-truncated windows are kept (in order to permit re-blending). In FullExtraction, where the spectra are deblended, there is no filter on the truncated windows.