I got the standard BTD library running, and found out it’s ridiculously easy to train and run these things. And I even know pretty well what they do! It seems vaguely like cheating, because it’s so easy. I’m working on lots of plots for next week’s meetings, but for now, here are some numbers:

So, starting from the same filter-level cuts I had before*, and defining the “signal” sample as upgoing and contained (so that’s 50k numu per year) I get (drum rolll.......)

__________________________________________

|   0.987 signal retention at 0.30 background    |

|   0.882                               0.10                        |

|   0.287                               0.01                        |

__________________________________________

Major remaining question (that I’m having trouble answering because my training sample mysteriously stopped working-- must be time for dinner):

Is that the fraction of

                    (background after/ background before)

                                                    or

                    (background after/ total after)?

It’s an important question there.

*”Filter level” cuts:    

1. (1)passing 1hit Deep core veto
   

2. (2)passing noise engine
   

3. (3)NCh>8 including some SLC: same cleaning as cascade analysis
   

4. (4)N(fiducial hits) >4 (same cascade cleaning)
   

5. (5)Nhits after static hit cleaning 7500 and dynamic cleaning 400 >3