(I hope we haven't covered this somewhere...)
The 1949 EDSAC could boot, as we say nowadays, from a bank of uniselectors. It's unclear to me how program code was actually represented on the uniselectors.
The EDSAC had 18-bit words (17 usable!) so a naïve approach would be to have 18 uniselectors, with one word on each switch position. And given that Initial Orders 1 has been reported as being 31 words long, that naïve approach would require 31 switch positions.
The photos I can now find have 5 uniselectors. That, perhaps not coincidentally, corresponds to the 5-channel paper tapes that EDSAC used. So maybe the uniselectors were delivering the same thing. That equates to 7 uniselector positions for every 2 machine words, which works out at requiring over 100 positions total (for the 31-word Initial Orders 1). That seems infeasible, though I've not actually counted positions from the photos.
So how were the uniselectors coded?
Perhaps one position of the rotating switch could hold multiple values, selected by external logic, so a complete boot load required multiple revolutions?
Added: this YouTube video from the replica project says that 4 of the uniselectors contained the data, and the 5th controlled the operation of the other 4. That doesn't get me closer to the encoding, but it does shoot down the '5 channel paper tape similarity' hypothesis.