I've been sidetracked by BOINC-related matters recently, but have nonetheless made some progress:

• In the last run, I noticed that most birdie signals didn't have any real signals nearby in time/freq space. Eric pointed out the reason for this: the telescope pointing history (used to generate birdie signals) goes back to 1999, but we've kept only signals since the ALFA receiver started operation, on 6 July 2006. I changed birdie signal generation to only use pointings after that, and that fixed the problem.
  
• There were conceptual errors in the code for detecting drifting RFI. This produces triangles in time/freq space that have a statistical excess of signals. We do RFI removal for spikes first, and save the resulting list of triangles. When processing Gaussians, we flag the signals that like in one of these spike triangles, as well as those in Gaussian triangles. My code stored both of these in a single list, which was supposed to be time-sorted but wasn't, and chaos ensued. All better now.
  
• The new RFI removal program (rfi2.cpp) had a drift window size twice too small, so the negative part of the window was chopped off and nothing got flagged.
  
• A couple of tape records had negative (invalid) beam numbers. This caused RFI removal to crash, but only after several hours of computing. Figuring this out was a bear.
  
• I've been working with Shiyu Li to get Nebula working with SERENDIP6 data again.
  

So, all this got done and I just finished re-running the pipeline.
If you browse around in the results, you'll see a couple of apparent issues:

• Birdies are getting detected, but as a bunch of 2-signal multiplets rather than
  one multiplet with a lot of signals (which would be much higher scoring,
  and thus we could detect fainter birdies).
  
• 95% of pulses are getting flagged as RFI.
  

And probably lots of other things; feel free to create a bookmark and email me
if you see something odd.

-- David