One way people have improved the accuracy of measurements in the community is to build light integrating devices which give pretty consistent readings regardless of the beam shape or angle of the light being measured. For example, if a laser-like light gives the same numbers regardless of which direction it is aimed in an integrating device, then that device probably has good integration qualities. One person who has done a fair amount of work in this area is djozz, who commented above.
Another way people have improved the community’s accuracy is by distributing reference lights which have known amounts of output. Basically, measure a light in a high-quality reputable integrating sphere, note the numbers it produces, and send that light to anyone in the community who wants to calibrate their own sphere. Then measure more lights in the same original sphere, so that many people can calibrate their spheres in a way which agree with each other. Ideally, also measure a variety of both floody and throwy lights in the original high-quality sphere, so other people can make sure their devices are accurate regardless of beam type. Maukka has done a lot of this work, and greatly improved consistency in the community.
Similar to that last bit, some have also taken trips to officially-calibrated spheres used for standardization purposes, and/or purchased calibration reference lights from standards agencies, to ensure that our measurements are similar to the standards.
All of those things help, but for some types of measurements, none are actually necessary. For example, if someone tests a light before and after modifying it, the relative difference between measurements can be very precise even if the absolute numbers might be skewed. So we can get numbers for things like “this lens allows 99.5% of the light through, but that other lens only allows 97.7“. Or ”this ZL SC52 makes 5.5 more lumens than my other ZL SC52”.
Perhaps more to the point, it’s also possible to detect things like when one brand uses a different definition of a lumen than another brand, because their numbers are consistently higher or consistently lower than expected. There are at least two different known “lumen” values in use by flashlight companies, and they differ by about 30% or 40%.
… and there are also rather a lot of flashlight brands which make up numbers for marketing purposes, with little or no relation to reality. Those are usually easy to spot, even without measuring anything. Like, if a light uses a Cree XR-E emitter and claims to make 800 lumens, it’s simply not true.
For a true 100,000 lumen light though, all of those so far have quite a few LEDs. This actually makes things more precise and consistent, since the individual variation tends to average out to produce a much more predictable result than one could get with a single LED. There are still plenty of other sources of variation, like the driver components, the battery type and state, the amount of dirt on the contacts, and even the temperature… but at least the LEDs can be a bit more consistent.
I hope this answers some questions.