I've been thinking a bit about artificial horizons, and I'm wondering whether this hybrid of mirror and liquid flat would work.
The goal is a highly reflective surface suitable for star sights, similar to mercury but without the viscosity, cost, and toxicity problems. The classic solution is a mirror and a spirit level. That is still the sensible solution, but here is a possible alternative.
Put a first-surface mirror in a watertight, stiff container, and cover with a few millimeters of clear oil. The container is provided with leveling screws. Now observe a bright star. There should be two images, one bright and one dim, separated by the misleveling of the mirror relative to the oil surface (which is level by construction). One could now manipulate the screws to make these images coincide, but that has the problem of judging the coincidence of a dim object swamped by the light of a bright one. Instead, get the images to within 10 or 20 arcminutes, then take two auxiliary star-star sights to estimate the leveling error. These would be between a dim star (observed via the sextant's index mirror) and a bright star's dim image (observed via the oil). Two such sights with reasonable geometry will suffice to estimate the mirror's zenith distance and azimuth (or xi and eta if you prefer).
Then take star altitude sights at will, and correct for the now-known mislevel during sight reduction.
This is similar to the Rayleigh water test for flats, and indeed one could theoretically search for fringes between the oil and mirror, but this would be next to impossible under field conditions (and fringe contrast would be very low).
Pro: oil is less costly and more robust than a spirit level. Con: messy oil handling; time spent with the two auxiliary star sights, which probably take longer than leveling up with a spirit level; loss of light from the oil reflection; extra computation.
Does this scheme hold water (if you'll pardon the pun)?