I have a good supply of XP-E2 R3 0D1 leds because BLF-member Mitko had (still has) them for sale cheap, and I even received a few for free extra. This made me think of not just using any led, but carefully choose the one with the best output. I understand that Vinh does that for his custom lights, but I never did that before. It is quite a lot of extra work, let's see if that pays off :-)
I have a Supfire F3 zoomie, decent quality (TomE did a review and mod on it), and the 27mm lens diameter is nicely in between the sk68 (18mm) and the UF-T20 (32mm). The size of the flashlight is therefore a bit more compact than the UF-T20:
(left Supfire F3, middle UF-T20, right UF-1405 shorty-mod)
Now, I did a sk68-mod some time ago using a dedomed XP-E2 driven to the max, that mod had a great throw: 39kcd. The theory predicts that with the same emitter and current in the Supfire I should get 272/182*39=88kcd. But I want 100kcd (just because..), so the emitter should be better. Enter: cherry picking the led :-)
I reflowed four of Mitko's XP-E2's on 16mm Noctigons. The leds were next in line on one strip of cut tape. Directly after reflow I dedomed them hot. I had the dedome set-up so well placed that at first I even dedomed them too hot resulting in the silicone being too 'blubbery' and much residu was left on the die. This costed me two of the leds. But after some practise I had 4 succesfully dedomed XP-E2's, not all completely silicone-free but I do not see a problem in that :
Led number four has a chip of phosfor broken off, but that should not matter much.
Then I tested all four of them in my (small) integrating sphere for voltage and light output. The results are the combined influence of led quality and reflow/dedome effects, but my guess is that any differences came from the specific emitters and not so much from the handling.
The differences appear not dramatic (no one will see it), but for what I want it is quite significant: the best led has 7% more output that the two middle ones, and 14% more than the worst one (you can be unlucky picking just that one 
). Led 1 has a suspicious drop after 2.2A, and after the test the output had decreased at the lower currents, and after an inspection it appeared the die was damaged. That tells me that XP-E2's driven so hard might not live long and happy lives 
.
Obviously led number 3 was chosen for the mod.
I used a NANJG AK-47 lineair driver with three extra 350mA chips stacked on top of the existing three, so 2100mA. The driver board in the F3 is 22mm, so I had to clear the stock board from components and let the AK-47 float on top of it: very functional (I measure exactly 2.10 A at the tail), and very ugly but I'm afraid I have no pictures of it 
. I also did some cosmetic touches on the output side of the light: sanded the brownish screw-in ledboard retaining ring (the new trend in chinese aspherics, see also the new Uniquefires) and painted it matt black with heater paint, and used a fine permanent black marker to blacken as much of the led around the die as possible without damaging the die and bond wires.
This resulted in a great clean beam when zoomed in.
I did an as accurate as possible throw measurement, with my Mobilux classA luxmeter at exactly 5 meter distance, with a freshly charged 18650 (charge status=not that important with a 2.1A lineair driver) exactly 30 seconds after switch-on. The number is 96 kcd. It hurts a bit that it was not 100, but it is 9% more than the 88 kcd that had to be improved. So I consider this little project: goal accomplished ! :-)
A beamshot together with the two modded flashlights from the picture above (left: this Supfire F3-mod 96kcd, middle: UF-T20 with dedomed XP-G2 R5 3C dedomed at 3A current 108kcd, right: UF-1405 shorty with dedomed XM-L2 T6 3D dedomed at 5+ A current 390kcd). Test tree is at 23 meters, in winter without leaves this does not really show the brightness differences, but it was fun making this shot 
:
Would I bother doing this amount of work again for a mod? Nah, not really
.