But what degree of variation is there within examples of each revision ? If machines vary by +/- 1.25 dE you might just be seeing 'normal' tolerances of the machines.
Let me explain. Multiple (at least 4) Rev C's were compared to each other, they all correlate very well. Max dE (using an ECI2002 so a lot of patches) is less than half a dE (dE 2000). That's good and expected. There is '
noise' in the data even if you used one iSis and measured the same ECI two times in a row, you'll never got 0.00dE so we do see a fraction of the noise above. 4 iSis Rev C's are less than half a dE apart. That's good.
The same is true of Rev E! Multiple units correlate with each other as expected.
Rev C and E
don't correlate and the differences are in worse case around 2 dE and change, and found on white. I explained that X-rite changed the hardware. Which is correct, Rev C or E? As you'll see, we don't know, it's not really possible and in the end, it doesn't matter expect in one super important case (using a mix of units to measure data to correlate process control).
OK, onto BCRA tiles. Based on the above, they don't help (one could suggest in many cases they don't help in this analysis). Why? Say you have 100 BCRA tiles's and they are all identical, 0.00dE. But wait, what did we use to measure them to get to that result? If it were a Rev C, that be one thing. If it were a Rev E, that be another. Or a Rev C iSis and a DataColor Spyder. Or an iSis and some Photo Research Spectrophotometer. The tile didn't change, the instrument, how it measures, it's illuminant and a lot of hardware specifics could alter the measurement data. Yes, in a prefect world, every manufacturer's Spectrophotometer's would work the same way and provide identical values. You can see that in just one example of a single company, making a single device, that isn't the case.
What does X-rite (or any other company) use to measure the BCRA tiles? It has to be a known, more 'accurate' (higher end, more expensive, better spec's?) unit. Let's say in this case X-rite has a half million dollar lab grade Spectrophotometer they use to produce the spec's for accuracy that they do. It's used for recalibration for customers. It's used as their ultimate device for correlating all other devices they build. Is this high end device identical to DataColor's or Photo Research's super duper reference lab grade Spectrophotometer? Probably not. At least X-rite is attempting to bring multiple insturment manufacturer's onto a more level playing field with their M series of measurement protocols.
This comes down again to my pet peeve about discussions of accuracy. "
I want my raw to be accurate to the scene". "
I want the most accurate Spectrophotometer". A lot times, it's akin to those questions about which pin has more ICC profiles dancing on it. When rubber hits the road, all the debate hinges on some pretty small details we and other's can't control.
IF you use a Rev C to build an ICC profile and you verify by printing images it works to your satisfaction, that's one thing. That's a fine first step. Now you can and should use the
same instrument for trending and process control. When I talk about Color Aim (a CHROMIX Maxwell term), I'm referring to the numbers we decide define accuracy. In this case the profile was built with an instrument we picked. We need to use that same instrument to measure the color aim. The color aim can be any target the instrument can read, designing one for specific kinds of process control and trending is a complex and different subject. But let's say we built a CMYK profile. Now take some RGB TIFF target file that needs to be converted through our profile, we print that then measure it. We compare what the profile predicts (the color aim which is RI specific) to the actual colors just measured. We can now gauge QC and trending. The instrument may have a dE of 2 compared to another instrument and as you've seen, this is not only possible, the scenario exists today. Accuracy in this context demands we use the same instrument or an instrument we know correlates as the group of Rev C's do to each other. If we didn't do this, our analysis would show a dE 2 in paper white and we'd probably blame the paper manufacturer. That would be a huge mistake.
I've built profiles with Rev C and Rev E that are excellent. But they are different. Which one is more accurate? The instrument and it's spec's do not tell us this. Using the instrument and comparing what it measured to build a profile and what the print system actually produces does allow us to know if our process is accruate (to the device), not if the device is accurate to something else. And in the end, does that matter? How would we control it?