That's a great result Kevin
My suspicion - from my own experiments recently - is that the colour cells of the Bayer array can be somewhat transparent outside of the visual spectrum, since the visual spectrum is all a DSLR is designed to care about and they've usually got a UV/IR cut in front of the sensor.
Once it's modded to remove the UV/IR cut, the passband of the colour cells becomes more apparent, and since the pixels of the sensor still have some response to IR wavelengths, when the cut out is removed we end up with photons striking the pixels and we get signal.
From that, I'm guessing the red pixels are responding to Ha + IR, whereas the green and blue just IR. It's possible they could be responding to Ha too, it just depends on where the colour cells start letting photons through again. Does this mean it's "false colour" though