Hi,
Very inaccurate. And, how do you measure the 20M resistor with any certainty?
50nF is 'huge' for accurate timing. 100pF already introduces error for 1ms with reliable repeatability. A frequency counter for a 1 kHz RC clock should show anywhere up to 10% error at one ambient temperature. Imagine accumulated error over 24 hours...
You could read comparison charts of capacitor dielectrics regarding capacitance change with applied voltage and temperature fluctuations and roughly extrapolate notions of expected minimum error. Polystyrene, polycarbonate, PPS, silver mica are good timing choices, NPO is okay, new silicon-based caps also allegedly.
Air wiring would remove invisible enemies like stray capacitance and inductance.
In my opinion, 1 second is already unrealistic for an RC network used in a clock, especially with a BJT and its temperature dependence.
I'd make the tau as short as possible, a second is a long time to make timing repeatably accurate for any capacitor, and divide down with counters, etc.