I used to do lab work for a living, so cynicism about precision comes naturally. Spent 4 months on a bit of work when it looked like someone was tankering really nasty liquid waste, loaded with cyanides, heavy metals and all sorts of other undesirable stuff and pumping it down residential sewers. People go to prison for those sort of tricks - can get up to 5 years for it.
I did the most painstaking analytical work I've ever done. Spent ages doing the statistics to check error boundaries. Checked the calibration of everything I could.
Got the cleanest error bars I've ever seen.
I'd have been ecstatic with 5% which is pretty good going for manual "wet" chemistry. I'd have been able to do enough handwaving to get away with 10%.
I got errors of under 1%. Didn't believe them. Got the boss who had a PhD in this stuff to go through all of it from the raw data. He got the same numbers. He watched me do a lot of the work all over again. He did some of it. We took each others' numbers and crunched them.
Still got the most precise work either of us had ever seen.
Right, somebody's going to jail over this. At which point all new sets of samples have to be taken - 3 of each in front of witnesses. You run one set of analyses yourself, you give one set to the accused and keep one set for independent analysis if the court tells you to.
27 hours straight spent throwing a bucket on a rope into sewage pipes did not count as one of my better days. Especially as this was the second time I'd had to do it.
The boss beat me about 300:0 at chess.
Took 4 sets of samples as neither of us really believed my results. Sent one set to a sister lab that had better gear for metals analysis (It was cadmium we were most worried about.)
I still got beautiful results - the sister lab gave us back random numbers. Turned out that the most expensive machine in our place, which had been calibrated a month earlier had a major problem. Like me, the guys who'd done the calibration had done them in a strict sequence.
The results from the spectrophotometer had nothing to do with the amount of cadmium in the sample and everything to do with how long the £$%£^&% thing had been running.
Precision is a very elusive thing sometimes.
Nowadays I'm not about to spend way more than the price of my house on gear for doing this stuff - just point out possible sources of error and try to keep it consistent. My test gear consists of a cardboard box painted white internally with a port for a lightmeter and a solar cell glued inside it. The solar cell cost about $2. A lightmeter that cost under $50, a data-logging meter that cost about $50 and a couple of other cheap meters.
Certainly not lab-grade kit. The cardboard box does have a diffuser (A piece of a plastic milk carton) and an internal baffle (White painted cardboard) so that light can't get directly from the torch to the solar cell, it has to bounce around a bit first, thus evening it out.