stellar quads

On the second day of the astrometry.net group meeting, Mierle, Lang, and I realized (or, more properly, bolstered our nascent view) that indexing all stellar quads (recall we are using quads of stars to get first guesses on image positions on the sky) on the entire sky is a very bad idea, for the two related reasons that (a) many stars tend to be involved in many quads, so the loss of a single star in the test image at test time can lead to a large loss in the number of expected matching quads, and (b) small n-tuples (like 5-tuples and 6-tuples) of stars can produce enormous numbers of stars, so we get false positive matches because there are many similar-looking 5-tuples on the sky. Yes, many! So we planned for Mierle and Lang to work tonight working on an index that contains no more than one quad for each star.