As many as it takes, whatever the cost.

That said, much of the existing delta between CBA compliance and status quo could be caught by a good automatic review system/programs. Part of the rest could be significantly reduced by the DALPA scheduling SME's in the same rooms as the schedulers etc. The remaining could be easily reviewed by hand. If it can't be done within the existing budget, keep the 0.05% and apply it towards that.

Its unreasonable to expect 13,000 pilots to know the entire CBA in real time and properly flag everything including things that depend on a million other moving parts including other pilots's schedules and interconnected coverage sequences. PB/PR days are messed up all the time; the rolling thunder ninjas probably have that on lock, but not everyone does and some situations can get pretty esoteric. Beyond an individual pilot getting shorted, that creates in the blind errors no other pilot could possibly catch unless they are scheduling forensics experts. Its very hard to review something you should have gotten when you have no reasonable expectation to know you should have gotten it. That's 99% DALPA SME or programming review dependant.

Companies know this, and there is a built in default incentive to "deny deny deny" as there is no penalty and substantial reward for doing so. Its also the inevitable result of hard (over?) working schedulers sincerely doing their best while simultaneously trying to put out hundreds of interconnected fires at the same time. Its not all nefarious intent, nor is most of it. But errors can rapidly cascade into missed money/days off for numerous pilots down the line, directly and indirectly. If someone else doesn't get the proper PB days for example, 30 pilots down the line someone gets a trip instead of stayng home, or someone else misses a GS they had no ability to ever knew should have existed. All the errors add up to costing DALPA dues money too, so there's that.

100% compliance on RR, coverages, PB/PR and other major interconnected scheduling items should be the goal. These things should be automatically caught by computer review or manually looked at by SME's who are the ninjas that 13,000 line pilots can't ever be.