I compared the canned Canon plat pro profile's accuracy against colors printed with the factory defaults and after calibrating. Attached are the plots of dEs (1976 and 2000) for 855 evenly distributed color patches.
I also ran a test to see if custom profiles made after calibrating were more accurate than custom profiles made with the factory settings. Opposite to my previous expectation, there was no material difference and the averages for both color patches and near neutrals were within .01 dE. Gamut sizes were virtually identical well. My conclusion is that if one can easily make custom profiles there is no advantage to calibrating.
OTOH, if one buys custom profiling services it's best to calibrate first. Future calibration should compensate for printer aging w/o the need to get new profiles.
Adding to the data. Similar results.
I've run similar curiosity-tests in late 2017 on the Pro-1000 and now some again. On smaller targets though.
Back then I compared Canon's OEM profile for Pro Luster with actual measurements, and confirmed that a Color Calibration brought the output closer to the expectations stated in their profile. So one can hope that at least Canon themselves assume the printer has been Color Calibrated when using their profiles, and one could hope other paper manufacturers do the same when issuing profiles, for best consistency.
Before (without) calibration I saw average CIEDE2000 errors of ~1.8 and peaks of ~4.7.
colverify -k -w -x target1-uden-cal.ti3 target1.ti2
Verify results:
Total errors (CIEDE2000): peak = 4.667316, avg = 1.825594
Worst 10% errors (CIEDE2000): peak = 4.667316, avg = 3.015105
Best 90% errors (CIEDE2000): peak = 2.493174, avg = 1.687279
avg err X 0.011928, Y 0.012853, Z 0.013387
avg err L* 1.430251, a* 1.170308, b* 2.192808
After calibration I saw average CIEDE2000 errors of ~0.8 and peaks of only ~2.2, and better results on "Worst 10%". Assuming profile is also smooth, this looks fine to me.
colverify -k -w -x target1-med-cal.ti3 target1.ti2
Verify results:
Total errors (CIEDE2000): peak = 2.177756, avg = 0.831824
Worst 10% errors (CIEDE2000): peak = 2.177756, avg = 1.779103
Best 90% errors (CIEDE2000): peak = 1.449286, avg = 0.721675
avg err X 0.003210, Y 0.003624, Z 0.006529
avg err L* 0.401963, a* 0.609139, b* 1.114283
That was back in December 2017.
As a sidenode, the OEM profile assumes high/highest printer driver quality setting. On standard it's a significantly different output.
Here around the start of 2020 / end of 2019 (around two years later) I ran a new Color Calibration on Canon's Pro Luster (which I use for generic Color Calibration - I haven't found use for custom Color Calibration). The change from before to after calibration was bigger than expected:
Average CIEDE2000 difference of ~1.3 and peaks of ~4.4.
Total errors (CIEDE2000): peak = 4.446026, avg = 1.299139
Worst 10% errors (CIEDE2000): peak = 4.446026, avg = 3.168163
Best 90% errors (CIEDE2000): peak = 2.559952, avg = 1.081811
avg err X 0.007117, Y 0.006639, Z 0.009679
avg err L* 0.632272, a* 1.046717, b* 1.642424
The total average is ok, but the peaks are a little bothersome to me, and the "worst 10%" are significantly off.
Even after re-calibration, I have a feeling my output is currently somewhat off compared to results from the earlier couple of years, as preliminary target tests after re-calibration doesn't seem to agree that well with old results (also after calibration) - which should be the whole point of calibration, to keep consistency. I'll dig in further. Perhaps something has unintentionally changed output slightly that even the calibration is missing, like inks (original), firmware and media types.