None of the companies claim you get 'x' output for 'y' time. They claim output and runtime separately.
No they don't, really. Random example from a manual:
Output and Runtime
80 lumens /35min
20 lumens / 8h
0.5 lumens /60h
Any
reasonable person would read that as "80 lumens for 35min, 20 lumens for 8 hours, or 0.5 lumens for 60 hours". So the "20 lumens" can't in any reasonable, sensible way be seen as a claim that's separate from "8h". But in fact:
ANSI FL1 runtime is when the light output is 10% of its initial output (measured 30sec after turning on).
There's the deceptive part of runtime claims. That is:
if this ANSI method is used. Many flashlight manufacturers (or sellers) don't even bother using a formal, reproduceable method.
Of course as you fire up a light, you can expect some dimming shortly after. But a good measure of performance is how long the bulk of that initial output is held. Or vice versa, what output you have for the bulk of the runtime. For example, runtime until output drops <50% of initial output, would be a better measure. Or say, 60%. Or 70%. For output dropping below 99% from initial, runtime would be near-0, that's a pretty useless measure. So you have to draw the line
somewhere. But let's be honest: 10% is just a
really, really low % compared to what you start with. Resulting in a big discrepancy between what's expected, and what really happens.
For a complete picture, you measure the output using a reproduceable / calibrated method, and plot that in a graph against time. Then it's easy to see whether output drops quickly or not, stays flat or drops all the time, or hovers just above that 10% mark for most of the time.
Surface area below that line is a measure for the total amount of light output, and that is what you should be comparing between lights. Using the same battery in same state of charge between runs.
That's where the more knowledgeable CPF members, integrating spheres, and (calibrated) lux meters come in...
Note that such runtime graphs are sometimes included in manuals. Whether those are accurate or not, is open to debate.