Hello Andreas,
After looking at your data for the UPWx resistors, in my opinion the flaws of the resistor design are revealed in the measurements to some degree. The characteristics are not terribly flawed, at least within your measurement range nor are you subjecting them to any great stress, still the flaws are evident.
I would like to point out a couple of things here, first, they have a rather quaint way of specifying TCR, a 'typical' of 0 ±3 PPM/°C over a 0°C - 85°C embedded within the overall TCR spec of 0 ± 5 PPM/°C, by doing this they lead the customer into thinking that they will get the tighter TCR spec over the limited temperature range, which they very well may get if the definition of typical is adhered to (at least a majority of resistors fall within the spec), however a resistor can actually have a TCR anywhere within the 0 ±5 PPM/°C range and still be 'within spec', a bit tricky.
My second point concerns the tolerance of a resistor, in this case, we're looking at a ± 0.1% limit, technically, a resistor can fall anywhere within this range and still be considered good (note this measurement is considered valid at the time of manufacture only), usually the manufacturer will make use of a 'guard' band, this means they will 'pull in' the actual tolerance limits a little bit to compensate for whatever measurement errors exist on the production floor. This 'guard' band varies with the equipment in use in production and usually can be as large as ±0.01%, so the manufactured limits may be ±0.09% for instance. The measuring equipment at final QC should always be the most accurate on the production floor to insure the resistors going out the door are within the specified tolerance (±0.1% in this in instance).
Depending on the manufacturer, the resistors going out the door should be within the required tolerance and can be anywhere within that tolerance, even within the same batch. Due to inherent variations in the manufacturing processes, whether the resistors are made by an automated machine or by hand calibration. The distribution of values tend to be a bit different between machine and hand calibration, machine tends to be distributed around whatever the nominal setpoint of the machine is and the distribution tends to be more of a clump (note machine calibration in the case of PWW is only possible with larger wire sizes as the machines tend to be a bit more heavy handed than a human hand) while a human calibration tends to be more distributed within the given tolerance band.
Tight tolerances ?0.01% are always by hand and depending on the operator will have more of a scattered distribution of values within the tolerance band, some operators can be very good at 'hitting' the nominal value quite consistently but these people tend to be fairly rare.
TCR tolerance works somewhat similar to the value distribution, depending on the manufacturing process, of course the TCR should always be within the stated range, but the distribution of TCR varies somewhat with the manufacturing procedures, most manufacturers produce a wider variance within the tolerance band than others do, that is inherent to the process and design of the resistor.
In conclusion, very few manufacturers give any kind of numbers on how many resistors fall within those 'typical' (undefined) specs if they even mention a 'typical' specification at all. Frankly, I think this is a bit deceptive, if you are going to give a 'typical' spec, then you should clearly state (even approximately) what percentage of resistors tend fall within this range. Now to my favorite punching bag, Vishay, while these folks often do give a 'typical' spec on their TCRs, there is absolutely no statement as to how many of these resistors actually fall within the 'typical' range given. According to many engineers I've talked to (direct or indirect), they have found few if any of these resistors even come close to the 'typical' TCR range. That is not to say that these resistors are not within the stated overall TCR range, they are and that is perfectly fine, they are within the stated TCR spec but few are within the 'typical' spec and I find that misleading.....how many of these resistors are actually within your 'typical' spec? Vishay should know exactly since they have to perform the measurements on these parts, time to come clean. Oh, wait, Vishay doesn't intentionally do that. Just to make sure I'm clear on this point, these VHP, et al, resistors are within the general TCR spec, I'm not saying they aren't, they just don't tend to hit the claimed 'typical' mark very often and they don't say how many do, it appears that their 'typical' is just a calculated statistic. If Vishay has the actual numbers, they aren't sharing them, why is the question.
I do not accept any kind of statistical manipulation of the data, exact numbers are available which should give a good figure of what 'typical' is by simple averaging of batch results, of course there is going to be some variance to that number, that sort of thing is inherent, nothing is absolute. Typical does not guarantee that you will receive more than the 'typical' yield in your resistor order, you just might get more than the 'typical' number of resistors within the 'typical' band, you might receive less than the 'typical' too, by nature, typical is an average figure and a good manufacturer tends to produce more consistent results, that means more customers should receive 'typical' more often than not, it just doesn't guarantee that. No silly statistical manipulation, just simple actual results from actual measurements.