I know that it is not recommended to use filters, but I do have several old sequences I'm trying to restore, where I used a filter and removed it during the shooting - or in this case, I think I had an NDVAR and changed it by hand. I was still learning in those days!

This creates a jump in the curve without accompanying changes in the shutter speed, aperture or ISO, and I'm trying to work out the best way to deal with this because LRT doesn't seem to be able to cope very well in this scenario. If I enable 'optical HG keyframe detection', the HG wizard still constructs the HG Lum Levelling curve using the EXIF data, so it completely ignores the change in luminance due to the filter (see attached screen shot).

   

I wish there were some way of telling LRT that it should use the luminance differences either side of the filter change, instead of calculating the curve based on the EXIF data

What's the ideal way around this problem?

You could try to set two regular (4*) keyframes left and right of the filter removal and then just adjust the exposure in Lightroom on those 2 keyframes, until they match.

Thanks very much for that suggestion Gunther. I'm pleased to report that it works very well indeed!