This app note, discussing the LTC1052 chopper-stabilized op amp, includes a wealth of good advice for designing and building high-precision circuits. (The box section on the final two pages discusses chopper theory in more detail.) In particular, the discussion of the thermocouple effect starting on page AN9-2 is very valuable. Figure 5 is worthy of careful study to understand the deleterious effects of thermal EMFs on precision circuitry. This kind of attention to detail separates man from mouse in the microvolt regime. A thermal gradient across parasitic thermocouples created by copper connections to solder or Kovar can be a significant source of error. (What's Kovar, you ask? It's a magical alloy of nickel that makes vacuum tubes, light bulbs, and other glass packages for electronics possible. While you're looking it up, look up the other magical alloys of nickel, like invar, elinvar, constantan, and mu-metal.)
More vintage hardware: Figures 2 and 7 show strip-chart recordings of the low-frequency noise. A strip-chart recorder is another one of those valuable tools (particularly for low-frequency noise measurements) that I wish I had room for. Figure 9 shows a voltage reference using several components from Julie Research Labs, including a saturated-cell voltage reference and a Kelvin-Varley divider. Julie Research Labs : when you absolutely, positively, have to exceed your research budget.
Figure 15 shows another HP oscilloscope photo (also, Figure 22). Heresy.
Figure 18 shows how to use the LTC1052 to correct for offset and drift in an Analog Devices AD650 monolithic voltage-to-frequency converter. Figure 19 shows a discrete voltage-to-frequency converter (that improves upon the specifications of the AD650 part) that is very similar to the architecture used in Figure 2 of App Note 7. Now we start to see the rest of the V-to-F dynamic range iceberg (see the so-called "best circuit" from App Note 7): Figure 19 has a dynamic range of 120 dB, Figure 21 has a dynamic range of 150 dB ("This is by far the widest dynamic range and highest operating frequency of any V-to-F discussed in the literature at the time of writing."), and even more is promised in App Note 14. See the footnote on page AN9-14.
Figure 25 shows a 16-bit analog-to-digital converter, this time using a sigma-delta architecture.
For the best circuit, I'm going to have to go with the thermometer in Figure 27, which feels like a wicked prank. This application points out that the offset drift of an AD547J (the competition's op amp, now obsolete) is so bad that it can be used as a temperature sensor. (In reality, the max drift was only 5 uV/C, but that is one hundred times worse than the LTC1052 discussed here.)
Best quote (page AN9-2): "Any connection of dissimilar metals produces a potential which varies with the junction's temperature (Seeback effect). As temperature sensors, thermocouples exploit this phenomenon to produce useful information. In low drift amplifier circuits the effect is probably the primary source of error."