Nearly every failed print I've had traces back to the first layer, and nearly every bad first layer traces back to the bed. Not the slicer, not the filament, not the temperature. The bed. Once I accepted that, I stopped tinkering with five variables at once and started looking at the one that actually mattered.
The single biggest improvement cost almost nothing. I clipped a sheet of borosilicate glass over the stock aluminium bed. The aluminium had a crown in the middle you could feel with a straight edge, maybe a third of a millimetre, which is colossal when your layer height is 0.2mm. Glass is flat to a tolerance I'll never out-print, and it gives a glassy-smooth bottom on the part as a bonus. That one swap did more than any amount of thumbwheel fiddling.
The second was a routine, not a gadget. Level diagonally, in three passes, not one. Adjusting any corner tilts the whole plane and throws the others off, so the first pass is always wrong and the second is closer. By the third the changes are tiny and you're genuinely done. Every time I rushed and did a single pass, I paid for it with a print that lifted at one corner.
If your firmware and probe support it, mesh compensation is the proper answer: probe a grid, build a height map, and let the firmware ride over the dips during the first few layers. G29 then M500 to save it. The only rule is that a saved mesh is a snapshot. Re-tension a belt, swap the bed, nudge a spring, and the mesh is now a lie. Reprobe.
That's the whole thing, really. Bed levelling isn't a one-off you nail and forget, it's maintenance you do whenever something physical moves. A flat surface, three patient passes, and the discipline to reprobe after you've touched anything. Do that and the first layer mostly behaves. The war isn't won, but it's a manageable skirmish rather than an evening lost.