Thanks again to Jack for his very helpful analysis of the variables in gretl's libset.c. These variables seem to fall into 3 groups: * Vars that are already settable via the GUI. * Vars that are irrelevant to the GUI. * Vars that are (at least potentially) relevant but that can't currently be set via the GUI. The last group includes the PRNG seed, the parameters associated with time-series filters, and the parameters associated with iterative estimation (tolerance, max iterations). Here's how I think these elements should be dealt with; comments are welcome. 1. The PRNG seed should be visible and settable in the dialog for defining a random variable. 2. Filters: There's no GUI mechanism for invoking these at present, so putting the parameters into a GUI preferences dialog seems odd. What we should have is an item in the time series portion of the Variable menu, "Filters", which includes Hodrick-Prescott and Baxter-King (and maybe for good measure some simpler ones too, e.g. moving average and centered moving average). When you select one of these items you get a little dialog where the parameters are configurable. User-selected parameter values are saved for the duration of the current session with the current data set, not permanently. 3. Iteration parameters: all estimators that use iteration, and where convergence is a potential issue, should have a little Options dialog associated with the model specification dialog (not sure of the best GUI design yet) where you can set tolerance and max iterations. Thus, my general idea is that these variables should be settable "in situ", rather than in a dialog under /Tools/Preferences. Allin.