This is very impressive work, i wonder how much this would be worth (materials plus time at reasonable wage)
There's three answers to the question how much it is worth:
1) materials+wages. It is a hobby so wages are not there, I enjoy the work that is in it. The materials: I guess about 150 euro, this is including 5 new-build test flashlights with constant output. If I ever have a lightsource officially calibrated for it, another 325 euro's is added. (for the luxmeter alone I payed 425 euro's second-hand, new and freshly calibrated this one goes for 1500 euro's. But I had the luxmeter already and use it for other measurements as well, so it was not purchased exclusively for this sphere).
2) the functionality. If this sphere performs close to the performance of a professional sphere, it could be worth over 1000 dollars. A professional sphere goes for many thousands of dollars (think 30), but it comes certified, which will never happen to my DIY-thingy, and it has all kinds of convenient computer-outputs and software.
3) relevance to what we do here at BLF. Well, probably it is worth close to any white-painted cardboard box with a cheap chinese luxmeter (so say: 20 dollars), meaning: if you get within 10 to 15% of the real lumen-value which is what such a box probably achieves if you build and calibrate it carefully (and more accurate if you measure only cool white flashlights with similar beam profiles), the difference is not noticable by eye and therefore good enough. My goal of getting within 1% for all beam profiles (from mule to thrower) and colour temperatures is purely for my own enjoyment (and I must admit that I do enjoy posting about it as well ;-) )
Thanks for the detailed reply 
I agree, its not about the money, its a hobby, but i do like to see what goes into a project, especially a hobby one (the watermark software comes to mind from another thread, the writer worked carefully on it instead of corporately produced which values production rate over refinement and bug catching)
You answered my next question about how much the meter cost, its far more then i expected but your right that its worth it for the repeatability and light colour correction!
How many hours has this project taken?
Looks like you are doing some great work here Djozz.
About the calibration. Using a single light from a single manufacturer as the holy grail for calibration seems to be a bit of waste of all the time, energy and money and you have put making the sphere and buying a good light meter to get accurate results. Most know that different manufacturers can have quite different numbers and some like to inflate their numbers or round them up. On top of that you cant get a good average even for the type of light you use. So even that light could be a bit off the average.
I know you don't buy a lot of premium lights. But you do buy a fair amount of emitters. Cree and other manufacturers have output data that you are familiar with so why not just use that data for calibration? With the amount of emitters you test you could average out the results over time and get a good calibration that you could fine tune as you test more emitters. Maybe remove strange results if you suspect one or several emitters are off. Over time I think that could give you a much better calibration without spending any more money. Not only much better compared to using a single light as the holy grail, but also a compared to using a bunch of supposedly ANSI calibrated lights.
Thing is that in this project I see the absolute calibration of the sphere the least interesting part to find out (my sphere project is bit a bit a case of "l'art pour l'art") because I can't really do a test on the end result, I can only pay a considerable amount of money and then get the real figure. Perhaps because I also use these spheres for emitter tests, the BLF-community deserves a better calibration than this, but taking a mean of the output of a bunch of unprecise light sources is highly unsatisfying to me, it will never get me to the 1% accuracy that is my (probably not realistic) goal for all other aspects of the light measurements . I will try to figure an affordable way to do a true calibration eventually but in the mean time my SWM D40A high output (it is my most stable output light, and I use it just for this purpose) is a very convenient 'working' calibration because I can alter any aspect of my sphere (or build entirely new ones) and the lightsource is always there to recalibrate it to the same standard. And if (at all) I arrange a true calibration in the end, all my results can be corrected with one correction number.
attention to details and determination on this new sphere for sure hold your signature Djozz. if i ever take the courage to make my sphere your work will be my main inspiration. hope to perform inside the hi-standard you allways seems to ask from your work.
as a side note, ouch before, that nice power suply than the exotic spectrometer now the precise light meter , this hobby together with those little crees seems to have plotting to sabotage your finances 
I never saw <6% lumen accuracy in any of LED datasheets,so your goal sounds optimistic
.
Your comparing apples to oranges, datasheets are not based on accuracy of lumen measurement, they are based on tolerance in bulk manufacturing.
You should read datasheets more carefully,for example:
Cree: "Cree maintains a tolerance of ±7% on flux and power measurements, ±0.005 on chromaticity (CCx, CCy) measurements and ±2 on CRI measurements"
Lumileds: "Philips Lumileds maintains a tolerance of ± 6.5% on luminous flux and ± 2 on CRI measurements"
Luminus: "Luminus maintains a +/- 6% tolerance on flux measurements"
Let me explain how manufacturing works, they produce millions of chips and sort them into bins which are labelled for sale. Each LED has a slightly different output so they sort the chips into output bins, each bin has a lumen output between the stated values. If you buy two XM-L2 U2 1A chips and tested them, they would fit into the U2 bin but would not have the exact same lumen output, but would be between the 7% difference stated.
Thanks,but you don't have to explain to me how manufacturing works,manufacturing tolerance and luminous flux measurement accuracy are two different things.
You said:"datasheets are not based on accuracy of lumen measurement, they are based on tolerance in bulk manufacturing";
Yes,datasheets are based on tolerance in bulk manufacturing,but how accurate you can measure those tolerances-depends only on accuracy of measuring equipment.
And only ToyKeeper can actually see the difference.
The rest of us mere mortals are immune to the 7% tolerances. 
(not said jokingly, but in utmost respect for the critical eye of a professional with talents)
...a slow going thread, but this work has no urgent practical use anyway :-) ....
A good integrating sphere integrates so well because the light is reflected multiple times with minimal loss on the inner surface of the sphere, so that the initial beam pattern of the light source 'drowns' in all subsequent reflected light. If on the inner surface the reflected percentage of the light is not the same for all wavelengths but slightly different, there is a problem that builds up, with every reflection the difference is increased, and if you realise that for the sphere's output 10+ reflections have a measurable impact, you understand that the reflection of the inner surface should be very close to the same for all (visible at least) wavelengths, or the sphere will 'favor' certain wavelengths. The crappier your sphere, the less this is a problem of course . But does a coated sphere do that any better than a bare styrofoam one? (to be honest: I can not measure 'better', I measure 'different', but that is already good information to me)
this weekend I did runtime measurements of six 'constant output' flashlights in the new sphere. I did this for two reasons, 1)to see how constant the outputs were (quite, but not really, but it is repeatable, that is even more important), 2)I will do the same measurements again when the inner surface of the sphere is coated, comparison will hopefully give a clue about what the difference is of the sphere's spectral output after integration, with the bare styrofoam, and after coating (does coating make a difference?).
To see a maximum difference of the sphere's 'response' for different parts of the spectrum I made a red, green and blue flashlight (host 602C, 1x7135-driver, XP-E2-leds), and to see how much of the (if any) differences are leftover in normal flashlight situations I made a cool white (Uniquefire S10 host, XM-L2 0D, 2x7135), neutral white (Fandyfire A10B host, XM-L2 3B, 3x7135), and warm white (Uniquefire S10 host, XM-L2 7D3, 2x7135) flashlight. All leds were on a DTP copper board and in case of a brass pill (Uniquefire and Fandyfire hosts) they were soldered to it. Batteries were freshly charged IMR batteries.
Here are the outputs of these six lights in the uncoated sphere, coating will be done later and then this will be repeated and compared:
So when made this way with good heat-paths to the shell and with only few 7135 current regulators, the flashlights have a fairly stable output but not 100%. For practical purpose, if these flashlights are used between 5 and 10 minutes, the output stays within 2%, some within 1%, for more constant output you can wait longer. BTW, I used no active cooling of the lights, they were left flat on the wooden sphere-box at room temperature (see picture below), the output may have been flatter with fan-cooling but that would make the lights very unconvenient to use. The clear winner btw is the blue flashlight, the output dropped just 2% for the entire runtime of 40 minutes, while for the red flashlight it was 10%. Differences are caused by battery voltage (even though they are supposed to deliver constant current, 7135 chips I found are slightly sensitive to input voltage) and by led temperature, while some leds are more temperature sensitive than others. . But for me the most interesting is that these runtimes are very repeateable (I showed that for the cfc-light in post #16), if I cool the lights down and recharge the batteries, these runtime graphs will be exactly reproduced.
An illustration of the problems you get when trying to measure in the 1% region is when I measured the red flashlight after 40 minutes, I noticed that during the measurement the output started rising again; it appeared that cooling of the light by touching it for the measurent was the cause: to check, I held it fully enclosed by my hand for ten seconds and that resulted in a 2% output increase, something very significant to account for when using this type of 'constant' lightsource 
. I have the impression though that the red XP-E2 led is more sensitive to temperature than the other leds.
2bcontinued..
Red,red-orange and especially amber LEDs have terrible output drop with temperature rise.As you can see,their output is much more temperature dependent than for white and other colors.
Some details of the sphere, still before coating.
The sphere half with the holes and the baffles. You can see some repairs, the inner surface had notches and polystyrene bits sticking out, I cut bits away and filled holes with lightweight wall filler.
Detail of the glued in (foamboard) baffles. BTW, not all glue goes well with polystyrene.
Here is the luxmeter-sensor. While the sensor of match's sphere was easy to clamp into the sphere repeatedly in quite the same way, with this one it was not so easy. I made a mounting solution for it with a black metal ring I had leftover and that accidentally exactly would fit the sensor within a fraction of a mm, making it possible to repeatedly mount the sensor in exactly the same position.
The ring on the inside of the box. The hole in the sphere fits around it with no play.
On the outside, with the clamping mechanism.
The sensor in position.
And how it looks on the inside of the box.
Ah, thanks, the specs back this up, also no surprise that the blue one is so nicely constant :-)
I started on the coating. For the first layer I used plain latex wall paint, to create a stable substrate for the 60%BaSO4/40%latex paint mixture of the subsequent layers. The mixture is not very great as a paint but it does stick and does not readily crack away when the styrofoam is lightly deformed.
Only when I started painting I discovered that this paint is not really white but slightly off-white, to the yellow/rosy side, just a bit, not a real problem because there will be more layers. For the mixture with BaSo4 I will use the other pot of latex paint I have that is much whiter, actually looks as white as the BaSo4 powder itself (and the styrofoam by itself also looks surprisingly white btw).
The off-white paint of the first layer gives an opportunity to show the cumulative effect of an integrating sphere. In the first picture I shine a flashlight (Nichia 219B) through the hole of one of the sphere halves onto the other half. But because the halves are separated, multiple reflections are not possible. (the greenish tint is from my limited camera). You can see that the reflected light from the paint is (almost) the same tint as the light reflected from the bare styrofoam on the outside of the sphere.
But with the halves connected, multiple reflections are taking place, leading to a much different tint (as seen through the hole), the yellowing effect of the paint is multiplied! :
Off course I should have shown this for the bare styrofoam too, but I forgot. I will get me another styrofoam ball one of these days and do the experiment still.
Thanks for the update.
… What?
I can see PWM pretty well, but I’m not very good at small differences in tint, and even worse at brightness. Especially at the red end of the spectrum, since I seem to have a bit of a blue shift in my vision.
I finished the coating and tested the results. The news is bad and the news is good. First some pictures:
On the left is the wrong paint I used for the first layer on the polystyrene, on the right is the latex paint that I used for two more layers, mixed with BaSO4, it is visually the same white as the BaSO4 powder in the middle. I made the 60/40 BaSo4/latex-paint mixture much thicker than for my first sphere, so I sufficed with two layers, the combined thickness was more than what is inside my first sphere.(for a background why I use this mixture, see my first 'Floating lumens' thread)
A picture of the BaSO4 powder next to bare polystyrene, also pretty much the same white visually (part of the polystyrene can be seen painted with the off-white latex).
After finishing the coating:
The sphere was put inside the wooden box, the holes matching the holes in the wood and the sphere firmly clamped in position by some cushioning leftover from flashlight packaging.
Then the lid was screwed in place.
I let the paint dry for a day and measured the reflectivity (the output of the cfc-light and my SWMD40A on high setting), then again two weeks later. In those two weeks the reflectivity of the sphere went up 1.02%, I hope that was still the paint drying for the last bit. Now it is two days later again and the measurements are exactly the same as two days ago. I will confirm (I hope) that is still the same a month from now.
The reflectivity of the coated sphere compared to the bare sanded styrofoam had gone down significantly (like in my first sphere) , by a whopping 37%. This is not nice because that reduces the integrating properties of the sphere. Luckily all my observations suggest that the integration is still pretty good. Because the total reflectivity is a combined result of a large number of reflections taking place inside the sphere, the decrease in reflectivity of a single reflection is much less of course. To get an idea, a few months ago, to get a feeling for my first sphere, I calculated the combined reflectivity after 17 reflections (adding more reflections to the calculation does not change the number much anymore) for two efficiencies for single reflection: 88% and 90%. After 17 reflections it comes to a difference of 13.6%. I could try to calculate what the difference for single reflection would be for the combined difference of 37%, but being a bit lazy, I estimate that on 6%. Ok, a bit less lazy, I just calculated it for 90% and 84% efficiency for a single reflection, after 17 reflections the difference is 30%. So the loss in reflection efficiency is even a bit worse than 6% 
.
So, if not for the reflectivity of the sphere, the coating must be good for something, right? At least the reflectivity for different light colours must have become more constant. A good integrating sphere must have the same reflectivity for all colours. Unfortunately I can not measure reflectivity directly, what I can do is measure the luxvalues inside the sphere of different coloured constant light sources before and after coating. If the percentual difference in luxvalue before and after coating differs for the various colours (sorry for this horrific sentence, any virtuosity in the english language really leaves me here), the coating has had an influence on the respective reflectivities over the colour spectrum (improved or gone worse I can not tell).
So I did the runtime tests again for the constant output flashlight that I measured in the uncoated spere as well. To get an idea in what analog age I am stuck:
Then I made the exact same graph that I made before coating of the sphere (see post #38), I only altered the scaling of the vertical axis, to set the runtime curves of the neutral white flashlight before and after coating on exactly the same level. So then it is easy to see if any curve has shifted (this only works if your display is wide enough to show the two sets of graphs next to each other):
Hmm, that difference is not great, let's flip the uncoated graph, so that differences are more easy to spot:
Still looks very similar, but there are changes, of course most clearly for the single colours. After the coating as compared to before coating, if the neutral white curves are defined equal (just to have an off-set), blue measures (about) 10%lower, red 10%higher, green 3%lower, coolwhite 0.7%lower, warmwhite 1.5%higher. Am I impressed by the changes in reflectivity over the colour spectrum that the coating causes? Not really. The extreme colours blue and red do have a clearly altered reflectivity, but not many will want to measure them very accurately (maybe me ;-) ). What matters in the real tough flashlight world is the different whites, and going from 3000K to 6500K the difference uncoated vs coated is only 2.2% (from neutral to cool only 0.7%).
I assume here, because I did some research on what substance to use, that this coating improves the constantness of reflectivity of the integrating sphere over the colour spectrum. But I did not find anywhere how good bare styrofoam already actually is in doing that (perhaps very good, who knows?). It is not unthinkable (I do not like that thought of course) that the difference that I find here are not an improvement but a change for worse. I just measure relative changes, not absolute reflectivities.
My conclusion about coating with BaSO4/latex mix? Because the effect is not that great, and because I do not even know for sure if that not so great effect is for the better, and because it is a messy and tedious job: just don't bother! And that is bad news because coating my spheres was apparently not really worth the effort, and that is good news because this makes all those uncoated styrofoam integrating spheres out there just a bit more trustworthy :-)