Linear power supply was not a BBC engineering requirement by the time I joined Designs Dept in late 1981. Perhaps it was dropped well before actual production; I don't know if that was pushback from Acorn or experience with a prototype.
But by late 1981, shipping some with linear PSUs was an unhappy compromise caused by : (a) demand much higher than expected and (b) designing the SMPS and getting it into production would be a couple of months late (and miss the pre-Christmas launch altogether).
So, to keep the overheating down a bit, the first production was Model A, with half the memory, and slightly less power consumption.
I never saw a BBC Micro with 8" floppies though they were ubiquitous in the BBC at the time, as well as standard for the "Richard Russell Board" (his own home computer design, Z80 based, which grew up to a 64k CP/M system. About 100 of these were made, mostly by BBC engineers. I built one, instead of getting a BBC Micro).
Also the "disk connector shall be a 34-way type" in the hardware spec (linked in the other answer) strongly hints that 5.25 inch floppies were intended; 8" drives generally used a 50 way connector.
One aspect which (if I remember correctly) was a BBC requirement - probably Richard's own - was that BBC Basic had to have a useful block structured syntax, as far as could be achieved while still being BASIC. The aim was, for an educational machine, to teach best practice at the time.
You could program in a Pascal-like style that was a considerable advance on other BASIC dialects around then, and it's probably why BBC Basic still has its fans today. This was probably the most important and far reaching BBC requirement in its impact on users, who mostly did at least some programming - and perhaps on the UK as a whole.
(Source : Richard Russell was my project leader at the time).
I've been asked to expand on the Richard Russell Board.
It was probably unknown outside the BBC, and slightly pre-dated the BBC Micro, it was fairly new when I joined in 1981.
Sadly, I left mine in an attic about 15 years ago during a house move. Mine actually turned out to be a bootleg - apparently Richard loaned the PCB masters to Research Dept and they made some in house, one was left over. It wasn't actually used at the BBC per se, it was more of a homer "home office" in BBC jargon) project, but it inspired the ZEUS modular system which ran a lot of the BBC transmitters - and the broadcast clock and other graphics at the time. ZEUS modules configured for S/W development were known as ZELDA - Z80 Loader/Developer/Assembler - I think that (1981)was before Zelda was a game character.
The BBC had a fairly enlightened attitude to "home office" projects, as long as they were kept out of hours and didn't drain expensive components. They gained a lot out of the process, like the ZEUS (Z80 Universal System) following the RTR Board.
It was a SBC with 32K DRAM (4116, the original 3 rail type). These used +12V for the storage, +5V for the I/O and logic, and -5V at a ridiculously low current to bias the substrate. It was important that the -5V supply came up before the +12V, or all 16384 transistor turned on at once, with fatal consequences. I fitted a relay powered by -5V to connect the 12V supply.
Later, a second board had 32K more, either for monochrome high res bitmapped graphics, or 64K CP/M. But the main board used Z80, I think 8251 SIO, 8255 parallel I/O, and the ubiquitous 6845 for video. Two 64-way connectors along one side gave you everything. I may still have the schematic somewhere but can't find it just now.
Most had 8" floppy drives (dual was luxury!) and it booted to BBCDOS in under a second. BBCDOS was pretty simple, but robust. It used 6.1 filenames (6 character name, 1 character extension) and a robust chained sector file system. The last two bytes of a 128-byte sector were a pointer to the next sector. No interleaving, so a 2.5 MHz Z80 could read 6 tracks/second while the original IBM PC was throttled to about 1 track/second.
And yes, he wrote his own BBC Basic interpreter for it. It wasn't 100% compatible - naturally, the built-in assembler understood Z80 instructions rather than 6502. That went on to sell as a CP/M version, the ... wait for it ... Sinclair (!) Z88 portable, Tatung Einstein, and eventually ported to x86 for MS/DOS (still available).
Mine spent most of its time running CP/M, where I had the FTL Modula/2 compiler.