We turned the verification procedure "inside out": instead of treating index stars as dartboards and throwing image darts, we treat image stars as dartboard and throw index darts. This has the advantage that it's much easier to build a background model for index stars because we control them and by construction they have nice properties (mod some problems with USNOB).

This information comes from run 123, which uses index 53.

This index was built with a slightly different strategy (start with no reuse of stars allowed, making a full sweep through, then increase the number of reuses allowed) and a new range of scales: 4.0 to 5.6 arcminutes.

Phase	Field Objects	Code Tol	Fields Solved This Run	Fields Solved Total	Fields Solved Total (%)
R1	0-50	0.005	7654	7654	98.74%
R2	0-50	0.01	14	7668	98.92%
R3	50-80	0.01	30	7698	99.30%
R4	80-100	0.01	6	7704	99.38%
R5	0-100	0.015	0	7704	99.38%
R6	100-150	0.005	1	7705	99.39%

(Zero false positives, of course.)

If I turn the log odds ratio threshold to 10⁸, one more field solves. If I turn it down to 10⁷, I get five more fields solved but one false positive. At 10⁶ there are more than 10 false positives and only about three extra true positives.

Here are the histograms:

 http://trac.astrometry.net/attachment/wiki/NewVerify/fig1.png?format=raw

 http://trac.astrometry.net/attachment/wiki/NewVerify/fig2.png?format=raw

 http://trac.astrometry.net/attachment/wiki/NewVerify/fig3.png?format=raw

 http://trac.astrometry.net/attachment/wiki/NewVerify/fig4.png?format=raw

 http://trac.astrometry.net/attachment/wiki/NewVerify/fig5.png?format=raw

 http://trac.astrometry.net/attachment/wiki/NewVerify/fig8.png?format=raw