It is indeed unforturnate that the stuff of many classical sudoku chains is poorly served by the definitions of strong and weak links. For this reason, I find it less ambiguous to use the terms strong inference and weak inference, or alternatively just OR and NAND.

It is relatively easy to write chains that use OR and NAND as resoltution rules. One merely needs to make the stuff that these chains consider "cell equivalents". A cell equivalent is any container such that its objects have the following property: Exactly one is true.

One need not restrict cell equivalents to natural sudoku containers. Cell equivalents immediately allow the use of ALS, HALS, and many other Almost patterns. In order to infuse some patterns into chains, one may wish to add semi-cell equivalents. Two types of semi-cell equivalents come to mind: One in which At least one item is true, and one in which at most one items it true. In most cases, the use of semi-cell equivalents is not required - but, for example, to chain an almost Y wing into a pattern, one can usually only define a container that has the strong inference quality.

In any event, sudoku solving is generally viewed as a form of entertainment, and personally I find the use of such concepts to be highly entertaining. The percent of puzzles solved by a particular technique or idea by a computer really is not relevant to the entertainment value of technique set.

The Easter Monster - I have taken an opening salvo at this beast, using precise resolution rules based upon exactly the type of cell equivalents described above. The following step is most properly similar to Denis' C-chains. Regardless of whether this step is viewed as proper or not, 'tis highly entertaining, and speaks volumes for the value of this type of non-adhoc logic.

To best find this chain independently, study the grid before entering the possibilities.

(2=7)r13c2=(2&7)r56c2 (this is a typical cell equivalent type argument, and makes the finding of such chains much easier.)

It is known that amongst the 4 cells, r1356, there exists one permutation of candidates 2,7 per possible solution. One can then partition that cell equivalent snippet as one sees fit. This is precisely analagous to the partitioning of a cell when using, for example, t-chains.

(2=7)r13c2=(2&7-1&6)r56c2=(1&6)r79c2-(1&6)r8c13=(1&6-2&7)r8c45=(2&7)r8c79-(2&7)r79c8=(2&7-1&6)r45c8=(1&6)r13c8-(1&6)r2c79=(1&6-2&7)r2c56=(2=&7)r2c13

The following is eliminated by this chain:

(7)r1c3, (2)r3c1, (1)r7c3, (6)r9c1

Also, the following cells are resticted to candidates 1267:

r56c2,r8c45,r45c8,r2c56

for a total of 13 eliminations. The puzzle is still a monster after this point, just a tad less angry!

The opening boolean argument, (2=7) is intentionally written that way, as it serves as the compression vehicle for the balance of the cell equivalents, or super cells. The compression for the rest of the chain is a bit different, as the chain itself allows one to consider each of the cell groups and pertinent candidate groups as if they were merely bilocation relationships between only two candidates.

Compression into cell equivalents can be done along any primitive strong inference, and also groups of same (although often within a common container).

Finally, the = and - signs are merely a convenience, as they illustrate the preferred partioning of the cell equivalents. Such a partioning is actually arbitrary and superfluous.

Relatively complexity is difficult to measure. This particular chain, to me, feels like a conjugate chain of length 8.