Running analysiscut2 on old data:
New ShowersI wanted to get some statistics on the errors MINUIT reports vs the errors calculated directly from the spread of the TDF. I saved the time error calculated for each site in the array for each shower in the best candidates file.
The points are grouped into lines, which I think represent individual showers with varying theta/phi/etc. Close to the core, errors are very small, much less than 1 us. They don't seem to take into account time errors caused by core location offset, etc. This causes good sites to be rejected. By 5km away, the TDF spread is larger than our shower window.
Both ways of calculating the errors give similar results:
The code with flat cutoff of 5 us kept 96 showers. The code using Minuit errors kept 91. The code using TDF spread kept 76. Many of the showers being lost had reasonable-looking data.
Although I have not run the TDF-spread code on the Other Possible showers, I think it would allow all of the high-energy, off-the-array showers which the flat cutoff removed.
I think a good argument can be made for having a single flat cutoff. On the low side, Minuit/TDF errors do not take into account the uncertainty in other reconstructed shower parameters. A typical core location error can be several hundred meters, or 1-2 us light travel time. On the high end, the very large spread at very large radii allows to many potential accidentals to sneak in and throw off the fit of the rest of the data. It also allows showers on the edge of the array to be moved outward so that the sites with bad timing fall far from the core where the time uncertainty is very large.