- Always make an image first. This is mandatory if you have no idea of the structure of your source but this is also useful since the image will directly show you the calibration problems, which are in general much more difficult to detect directly in the visibilities.
- Use any knowledge you can have on your source. For instance, fitting a disk function is probably the most appropriate for barely resolved planets and satellites.
- If your source is barely resolved and you have no
*a priori*knowledge of the source structure, you should start with the simplest possible model (*e.g.*`POINT`) and complexify the model only gradually (*e.g.*`C_GAUSS`,`E_GAUSS`, etc...). Verify how the uncertainties on the fitted parameters evolves to decide which model is most appropriate. - Always be critical with the results. In particular, image and deconvolve the residual table. You should obtain just noise if the fit succeeded.

Gildas manager 2018-06-21