>
bridged-T attenuator... goal would be 5k input impedance and output as low as possible.
Then ignore all the classic constant impedance pads.
You want a U pad. That gives highest ratio of input to output impedance possible. It also needs the least number of wipers/poles.
The 5K variable resistor could be a pot or a buncha resistors on a switch. Audio-taper pot is suitable.
With no output load, the input impedance varies from 5K at "silent" to 10K at "full-up". Minimum loss is 6dB voltage.
With 10K output load, the input impedance varies from 5K at "silent" to 8.33K at "full-up". Minimum loss is 8dB voltage.
With source impedance assumed to be much-less-than 5K, the output impedance varies from "zero" at "silent" to 2.5K at "full-up".
This pad gives "silence" by trying to short-out the two 2.5K resistors with the pot/switched-resistor. It is not really possible to reduce level to zero because there is always some stray resistance when you turn the pot or switch to zero resistance. Dirty wipers/contacts will give deafening blasts. When switching, make-before-break is essential in this rig.
A standard potentiometer, or equivalent switch, is probably a better plan.
If you actually must present over 5K to the source and 100 ohms to the load, without a transformer, the minimum loss must be at least 40dB voltage. You can not maintain voltage over such a radical change of impedance without an amplifier.
Why do you think you need a 100 ohm output impedance? No standard audio power amplifier needs anywhere near that low. For short cable runs, 5K impedance seen by the power amp input is usually fine.