Lessons Learned So Far

In principle, predicted S/N of 5000 achievable, but systematics deadly.

Collect as many calibrator photons as target photons.

The telescope must move efficiently.  Palomar collects more photons than VLT!

Instruments must not be Q&D or "toys".  Clean arrays, no internal reflections, simple designs.

Systems must be robust to handle constant switching.

Automate everything.

Don't bother with quick looks.