As one of the few (sometimes) active PDP-1 coders around, I cannot stress enough how human and versatile this instruction set is. Amazing are also the constraints: All instructions are single-word and perform in a single memory cycle, where any memory access adds another cycle to this (as does any step of indirection in this addressing, of which there may be multiple) – that’s it.
Where the PDP-1 is not purely RISC, are the instructions that are altering the internal state of the CPU registers (found in the operate group), which are microcoded and microsequenced and may be thus combined into a single operation (owing to the TX-0) and the instructions of the shift-rotate group, which are again microsequenced, thus allowing for shifts and rotates by up to 9 bit positions (the sequence simply counting up high-bits in the less significant half of the instruction word) per instruction.
(However, even Sophie Wilson allowed the ARM to have a barrel shifter.)
A major exception are the operations of the automatic hardware/multiply option (Type 10), which performs a small internal program for each of the two operations, driven by its own micro-cycle clock (effectively halting the main CPU sequence). But this was a later extension and not part of the original design, which had just multiply/divide-shift steps over the combined two internal registers (AC and IO).
Regarding the single-word instruction format, this comes with a handicap, namely a fixed size of address space, which was 4K. The PDP-1, however, could be extended to up to 16 modules of 4K each, giving 64K. Apparently, inferring from the markings of the operator console, the first production prototype that went to BBN had an addressing scheme much like seen later on the PDP-8 (field selection for separate data and instruction fields, suggesting DF/IF bank switching — according to BBN-alumni, this was never used with Ed Fredkin opting for drum memory instead, using a custom designed drum, which wrote or read the entire 4K in a single revolution, which was essential for time-sharing on the PDP-1), while the production machines featured two other modes: one for extended addressing, where the indirection-bit featured as another address bit (thus losing the ability for indirection, which lets this appear rather unattractive), or mounting and switching of (named) banks.
(A second such drum was delivered to MIT, where it facilitated MIT’s own PDP-1 time-sharing project, which became operational in early 1963. Both projects were inspired by John McCarthy.)
P.S.: As a testament to the PDP-1 it can be said that I wrote an entire video game (a Computer Space Simulator) for the PDP-1, not even once having to resort to a debugger. (My emulation doesn’t implement an UI for the console typewriter or the console lights or most of the console switches, so it was really coding in the dark, as there isn’t any facility to output text onto the screen either – for which you have to implement a renderer in software.) Doing the same a year later for the 6502 on the PET was much harder and felt like work, while the principal logic had been worked out already.