@svart-hvitt I just want to clarify my post a little further. I hope I don't come off as irritated or whatever
I just want to expand a bit on my previous post to hopefully add a bit of clarity.
When you measure a single driver on- and off-axis, you'll notice that even the best-performing and most linear drivers exhibit little response deviations of usually more than 0.5dB, even when moving the mic position just a few degrees.
Given that this is the case, getting your on-axis response to +/- 0.5dB (even if that level of accuracy were audibly preferable to e.g. +/- 1dB) would not necessarily result in the best (i.e. most neutral) frequency response.
This is because a driver might be e.g. +1dB at 2° relative to its response at 0°, and +1dB again at -3°.
Better then to take a lot of measurements close together (e.g. at 2° increments from +10° to -10°) and then use DSP to flatten the composite frequency response of all these measurements. Doing so will almost certainly result in some narrowband peaks and dips of more than 0.5dB at 0°, but the result will be a more neutral speaker than one that measures strictly +/-0.5dB on-axis.
EDIT: I've been trying to find a good example of this online. Unfortunately the best I can do is a speaker measured at 5° increments (it's the Genelec 8351A measured by 3D3A Labs at Princeton). It's clear though that you see plenty of 1 or 2dB peaks and dips from -5° to 0° to +5°.
Look at that on-axis c. 2dB dip @ 6KHz. They've apparently decided to have it there because at -5° and +5° the tweeter is once again flat. If the designer had chosen to bump up that 6KHz dip to flatten it, the outcome would be a less neutral speaker, even though it would now have less linear on-axis response on paper.
Ditto the more broadband on-axis dip throughout most of the top octave. There may be other variations not captured by the 5° measurement increment chosen by 3D3A.