Tony’s spreadsheets are a wrapper that helps generate the syntax for the M421 command. Where the data comes from is not terribly important as long as the measurements you take are at the same place the firmware thinks it is.
G42 can be used to move the head to mesh coordinates from 0…4 in I and J (firmware’s names, not mine). It you took manual measurements with a paper card, that’s great. If you took measurements with a dial indicator, that might be better.
I took measurements on my board with a dial indicator, and from the factory here’s what it looks like:
These surfaces are normalized at the center mesh value, which from ambient to 60C raised 0.05mm from thermal expansion.
In my opinion, this is beyond software leveling. Look at the deviation from one point to the next, and this is with a 10x10 grid. Watching the needle on the indicator move even from one point to the next was alarming, it could move up and down several thousandths of an inch before arriving at the next point. Even two adjacent points different by only 0.001", during the travel I saw variation as much as 0.005". The built in 5x5 grid cannot compensate for this level of variation, it would need at least a 10x10 grid, and that would have low and high spots still on my machine.
My solution is twofold: I have shim stock and I will be correcting the root cause, which is the bed webbing is not flat. Measurements will be taken directly on the threaded inserts and they will be brought in line.
After correcting that I’ll be putting a borosilicate glass build plate on top with a magnetic base and removable PEI spring steel build surface - which is similar to what the SM2 comes with and is why the inductive sensor can pick it up - the spring steel is just under the surface. I chose borosilicate glass because it does not experience the thermal expansion that aluminum build plates can exhibit.
When taking manual measurements, I prefer to have backlash compensation enabled because without it making fine adjustments has uncertainty unless you always measure from the same z direction. If you use friction on a card as your measurement instrument, and you command the nozzle to raise 0.02mm, it probably won’t. So you’ll have to go like 0.06mm, and then back down 0.04mm. With backlash compensation correctly set on my machine, commanding the nozzle to move 0.01mm will cause the nozzle to move, making the process much faster.