I'm not the only one to think this. The NY Times recently printed a review of the new Panasonic Oxyride battery, an alkaline disposable which is claimed to last 1.5-2x longer than regular premiums. The journalist not only did some tests, but actually called the manufacturer when they didn't live up to the hype. They laughed at him and his attempts at testing, but did explain some of the complications of trying to do it right. Keep reading, and you can laugh at mine...
I too have tried testing my batteries by seeing how long they run lights, with the results above. The colors are three different tests, solid lines are the voltage under load, dashed lines or dots are with the lights off, and the duration has been converted into Amp-hours output assuming that the lights have a constant resistance and produce exactly 20W at 12 VDC. The horizontal line is 1.75 V/cell, the number you're not supposed to go below without risking irreparable damage to the battery. There's an initial squiggle, which is the so-called "surface charge" coming off, and a long curve until the voltage falls off the end. I suspect very little additional charge can be gotten out even draining it to 0 V/cell, so I tried to stop where I was supposed to. The gap between open and closed circuit increases as the pack is discharged, showing increased internal resistance.
Really noticeable here is the difference between the black and green lines (first and third run) and the red line (second run). The first time, I measured open circuit voltage by turning the lights off, measuring the voltage, and turning them back on again. This would be fine, if the battery were actually an ideal voltage source in series with a resistor, but it's not. In fact, the voltage creeps up over seconds after the load is removed. So the second time, I waited two full minutes for the voltage to recover before measuring OCV. And, to ensure good data, I took a lot of points, one every 15 minutes, at least. So the battery got a fair bit of rest. While this time didn't count for getting current out, it did allow the battery to do much better than without it. Effectively, the average current drain was lower, leading to more capacity, though perhaps complete rest is even better than just a lower average. Who knows. You can see a similar effect in the green run, where there's a long break at around 7 A-h during which I took a few OCV points. When the load comes back, the voltage takes quite a while to come back down to where it was before, and if you squint at the lines, it seems clear that the break allowed more total energy to be extracted from the battery. It's this kind of thing that leads people to doubt the first law of thermodynamics.
The average of the green and black runs gives 8.7 A-h capacity, at C/7.7, which is around 80-85% of spec for this drain on a low end 12V 12 Ah SLA battery such as the Kung Long WP12-12. I've read that batteries are usually discarded when they reach 80% of new capacity, which seemed odd to me. 80% of a battery ought to still be worth 80%, right? However, it's just one sign of aging. Things like the internal resistance are also going up, so that the voltage under the same load at the same state of charge gets worse. Couple that to drive electronics which cut out at a fixed voltage to protect your oh-so-valuable battery pack, and you can't get even your 80% out.
ObRide report: I'm now attaching the battery pack with a bolt and wingnut. It works fine, except for having to stop and retighten them every couple of minutes. Now that its capacity is proven, the battery is working fine too, except for cutting out just when it's most needed. And yeah, it was windy today, too, but I'm getting tired of that excuse. 24.9V final.
Added. At least it made it home. Final voltage, 25.5, 27 min.