Beautiful work, and thank you for documenting it so openly. I noticed one possible issue in the lux-meter readings: the cone test appears to show AUTO ×10, while the direct-DC test appears to show AUTO ×100. If those are range multipliers, the corrected readings would be about 18,540 lux for the cone and 53,800 lux for direct DC. That would reverse the efficiency conclusion, so I think this point is worth checking before drawing the final interpretation. Much respect for sharing the experiment — this is exactly the kind of detail that helps refine the work.