My quick notes on watch rechargeable batteries.

This type of watch batteries are generally split into three categories:

Manganese L
Vanadium L
Basic old-fashion Li-Ion
Manganese:

1000 charge cycles at around 20% discharge depth
3V output
Large capacities up to 50mAh
Charge voltage around 2.8V to 3.2V
Vanadium:

1000 charge cycles at around 20% discharge depth
3V output
Large capacities up to 100mAh
Charge voltage around 3.2V to 3.4V\
Winner here is ML2020, at 45mAh, 1000 cycles @10%, 100 cycles @50%, 20mm x 2mm, due to the small size, special overlap of charge and discharge voltage allowing for some very tricky dual-power solar charging scenarios (else needs a special harvester) and: 

At 45uA consumption this gives about 20 days at 50% discharge without charge, and 2000 days / 5+ years life.
At 45uA consumption this gives about 4 days at 10% without charge and 4000 days / 11+ years life.
At 4.5uA consumption this gives about 200 days at 50% discharge without charge, and 20000 days / 54 years life.
At 4.5uA consumption this gives about 40 days at 10% without charge and 40000 days / 110 years life.
Collapse
Clearly the truth is somewhere between these two numbers, and having a watch that would clearly outlive you and your children, anywhere between 5 years and 100 years, let's say we hit realistically 10uA @ 20% effective discharge, which will give us about 1000 hours/37 days at 20% discharge without re-charge, 4500 hours/187 days to full discharge, and lifespan between 500000 hours/57 years or 180000 hours/20 years definitely sounds vey attractive.

We could definitely hit the 20 year life of the watch, since the STM32U575 is pulling  20uA/MHz at full throttle, 4.3 µA Stop 3 mode with full SRAM and 510 nA at standby with RTC, so assuming a once a second wake-up to perform a simple operation of updating the screen and say one an hour or even once a day wake up the BLE to pull the time, that's a winner. That would probably average around 10µAh as I'm pulling that average number out of my ass but those seem quite reasonable.

TBD check the BLE and e-paper consumption in low-power mode as that seems rather tricky.

The Nordic BLE pulls <20uA continuously while waiting for a connection and ~65uA while connected and sending/receiving data, so doing this once an hour for say a few seconds to adjust/pull the time is very reachable.

The E-Paper is a lot trickier to calculate.

Deep sleep is 2µA / no RAM retain
Sleep 35 µA / RAM retain
8 mA image update
280 ms partial update
680 ms full update
Therefore we are bound at 2uA on the low side and nearly 10mA on the high side.

Assuming a full image update once a minute, and partial image update every second.

There is a display library https://github.com/ZinggJM/GxEPD which lowers the update to about 6mA.

Also there is a rather interesting discussion about power consumption on a coin battery, definitely worth the read. https://hackaday.io/project/134018-coin-cell-powered-temperaturehumidity-display

Full image update would give us 8mA * 0.68s / 60s / 60m = 1.5 uAh, say 2uAh? Not sure, that seems too low.

Partial image update would be a lot trickier, I cannot do it now. TBD.

The e-paper is definitely the power hog here. May have to change it perhaps to something else?

More discussions here: https://www.pervasivedisplays.com/why-e-paper-displays-will-run-for-22-years-in-coin-cell-powered-iot-applications/

More to come. Best would be make the setup and measure. I might be just full of hot air and all this is not attainable or has some basic errors.