One of my favorite choices for metals is this eraser type grit impregnated sanding blocks. Comes in 3 grits. Will take rust spots on cast iron out, all the way to polished. Good stuff.
I think the main purpose of R6 is to isolate the MCU supply from noise/inductive spikes/ringing etc passed into the main supply rail from the heavy FET switching transients, LED board lead inductance, etc. In combination with the standard 100nF decoupler on the MCU it acts as a high frequency filter. Adding a big bulk decoupler can probably only improve this, depending on high frequency characteristics of the type of cap. used.
The chip ceramic 100uF Flashy Mike has used looks well suited here.
Using instead e.g. a small leaded electrolytic, might actually make the filtering effect poorer, despite being good for holding up the MCU.
Reducing the value of R6 too much might also reduce the effectiveness.
I seem to remember that adding R6 turned out to be quite important during the development, and solved some strange issues that can afflict earlier FET drivers which don’t have it, particularly when tuned to the max. driving very heavy LED currents. The Attiny13 also seemed to be more tolerant than the later versions in this regard.
Is there any internationally available NO-OX-ID equivalent?
Dunno, sorry. It's a conductive grease, so sure there are equivalents out there but think they can be costly.
For that small 4.7/4.99 resistor, according to DEL and what he saw on the scope, that was the most important change to get the ATtiny85 working reliably. Our old standard circuit for the 13A did not work reliably with 25/45/85's. Richard with his 25 based Mtn boards went in a slightly different direction, still works but I can't argue which is best. Just with DEL's, he's posted all the evidence on what it did to cleaning up the spikes, signals, etc.
For a while I was doing cut/jumper mods and tail-standing components to jury rig up Mtn boards to have that resistor where it should be, before the diode, but then OSHPark boards came out in all sorts of configs and sizes.
Plan changed. Below is a photo of a coil of a be-cu spring pair over the existing springs. Near perfect fit. After cutting the coils from the pair I’ll bend the ends to align with the straight ends of the existing springs and apply a bit of solder at the base and tip.
edit: I checked on ebay hoping to acquire more and post a link but rhey’re no longer available. Fortunately I have 2 pairs.
Quick and dirty analysis, FET datasheet has total gate charge 21.8 nano Coulombs.
At 15 kHz, that’s 327 uCoulombs x2 per second to drive the gate high and low i.e. about 0.7mA (which is lower than I would have guesstimated).
Add on current drain of MCU when firing on all cylinders, dunno, first figure I saw on the datasheet is 300 uA at 1 Mhz. As I thought it’s the FET gate drive that dominates.
The switch LEDs aren’t on, so ignore them.
So maybe we have as much as 1mA total current drain to hold up through a jolt to the torch (nice round number, and not sure I trust my 700 uA calculation).
Are we using Attiny 85 or 85V (lower voltage). Or is e.g. brownout detection in use, I don’t know. Lowest votage might be between 1.7 and 2.8V, I’ll choose 2.25, and cell voltage of 3.0, minus Schottky drop, say 2.7 V.
So max. allowable voltage drop during interruption = 0.45V
Thus therefore, (Capacitance = Coulombs per Volt, Coulombs = Amps x Seconds)
Draining at 1mA the 100nF decoupler can only hold up the MCU for about
100nF x (0.45V / 1mA) = 45 microseconds 
No wonder it doesn’t like bumps.
Whereas a 100uF cap. can hold it up for 45 milliseconds. 
That seems pretty clear to me, as a SWAG, even if I’ve got the numbers badly wrong, we are comparing tens of milliseconds (good) with tens of microseconds (bad).
Use a diamond sharpening rod at a 45 degree angle, I do that to all my lights on the inside edge of the tube where batteries are inserted…
I think the “magic ingredient” of things like NO-OX-ID might be something that cuts through pre-existing tarnish and maybe keeps on doing so.
Otherwise anything that seals the contact surfaces from the air, like car battery terminal grease, plain silicone grease, even Vaseline, will be a great help in simply keeping out air from the mating surfaces and minimising the chance of later oxidation, both of which solder and bare aluminium are prone to. I think any of these would be better than nothing, long term.
Pics of the production driver have been posted - pretty sure its not the 85V version, and definitely brownout detection is enabled.
Ah well. I think my SWAG makes the point reasonably clear. PS: I might have made a few little edits to my post after you quoted it, sorry.
I’ll be bunging on at least 100 uF.
Edit, what level does brownout trigger, and what do you do when it does ? You won’t have much time left to do anything, if interrupted and only 100nF. Writing to flash ?
A Attiny85 is used in the Q8, no V variant. We had this discussion earlier, somebody told me there is not much of a difference. 
I’m using my own firmware in this light, and BOD is enabled at 1.8V. Perhaps too low, but I expected the same mcu in this light as in the D4 where I saw a Attiny85v if I remember right. (D4 was the first test object with Attiny85 for my firmware.)
Only tested bump resistance with full cells, will do more tests when they are low.
You don’t do anything, on brownout detection the mcu goes to power off and comes back with reset when voltage is back.
No worries then. 85 vs 85V, heard the discussion that they are the same die, just binned at wafer probe, and in reality all good, just 85V can be sold at a premium (edit: to those who need to be able to show a worst-case analysis, for critical applications). Makes sense to me.
As documented in the source code header for NarsilM:
So brownout is set to 1.8V. There is a legit reason, at the time, for turning on brownout. Early on with a cheesy prototype, tightening the battery tube caused intermittent power hits, and it would lock up the MCU. Turning on brownout cleared that all up.
Okey dokey, where does LVP kick in on this ?
Let me know, and if you like I’ll re-run the numbers assuming Vbatt just above LVP, minimum voltage = brownout 1.8V, Schottky drop 0.3V
Summing up something different to the 0.45V I guesstimated.
But I think the end result won’t change much.
I should say, for others watching, that this is only a concern if wanting your torch to reliably stay on after some bumps and squeaks, not an issue for most.
What strobe frequency is used in the Q8? I found different values. 16 or 18 kHz
Hz, not kHz
Was mentioned in Q8 thread. It is so long that almost impossible to find. I think we need 100 pages thread limit.
Not Hz, definitely kHz.
Strobe? How do you imagine this? Your eye is hardly recognize anything over 30Hz.
…depending on what ‘strobe’ refers to