Floating: When your primary buffer becomes solved, rather than cycle breaking you can instead use another, unsolved buffer (which saves an alg). This is called a "floating buffer", and the act of using it is called "floating".
Full Floating: If you're able to effectively solve any 3-cycle on the cube regardless of which pieces it involves, you're said to know "full floating".
LTCT: Last Target Corner Twist. This is an algset that solves your parity corner target and a corner twist at the same time. It consists of 252 algorithms.
T2C: Twisted 2 Corners. This is similar to LTCT, but covers the cases where your main buffer is twisted in place, and there is also a twisted 2-swap of corners. It consists of 126 algorithms.
DAP: Dynamic Alternate Pseudoswap. This is a technique where you know several different algs for a given corner parity case (which solve different edge swaps) and you decide which one to use during your edge tracing, to take advantage of things like solved/pseudosolved edges, avoid cycle breaks, etc.
Sandwiching: When you have a string in memo such as ABCA (or any even number of letters between the two matching ones) it is considered a memo "sandwich", which you can solve with a floating comm(s) - in this case, the floating comm BC from buffer A.
Block comms: These are algorithms that cycle 3 edges and 3 corners at the same time, by moving blocks of pieces, rather than individual ones. A simple example is [r' D r, U]. These can be used when you have compatible cycles in corners and edges.
Full Parity: This is an algset that swaps your buffers (assumed to be UF/UFR) with any edge and corner target. Can be used to solve parity without pseudoswapping, or can be applied to give you many options for alternate pseudo swaps. It consists of 462 algorithms.
Full Floating Parity: Like the above, but not restricted to swaps involving your primary buffers. This would allow you to solve any 2E2C case with one algorithm. It is not considered to be learnable in full however, with 11,088 algorithms.
DLin Tracing: Stands for Dynamic Linking (or Daniel Lin) Tracing. A tracing method where instead of cycle breaking, you memo every cycle separately, making a note of the parity and orientation of each. You then decide on your cycle breaks during exec, ideally to give the highest number of floating opportunities.
2E2E: An algset that solves cases where you have two 2-swaps of edges (eg. [M2, U2]). With 5940 cases, it is not thought to be learnable in full. However knowing easy cases is useful for 3BLD, and the subset involving UF-UR swap is important for floating in MBLD.
2E2E': Like 2E2E, but each individual 2-swap is flipped (if you traced them, you would end each one on the opposite sticker from where you started. eg. [E, R U' R]). 2E2E' also covers 5940 cases, so is not thought to be learnable in full.
Credits to James for the accurate definitions.
