blue,

thank you for the quick evaluation again.

Since i wondered, that you got so much guesses on average in the line 2 list, i went through the puzzles with a simple solver.

Here bilocal guesses can be more effective. Similar to naked pairs the bilocals should not share a box plus a line, which naturally makes them "weaker".

I found (since i did it manually, there might be a mistake, but this would not change much), that with the following 4 rules, in all puzzles trying one pair of bilocals/bivalues is sufficient - with one exception.

1. If there is a bilocal for 1 in r4, take it

2. If there is a bilocal for 5 in c3, take it

3. Look at r4c467. If there is an ALS with a single 5 in a bivalue cell, and the 5 directly leads to another one, choose this cell (exception 427581936, where you should choose 18)

4. Take the bivalue cell r7c7

This way in the worst case 50+2*29=108 (or 110 with 2 more guesses) tries are needed.

412589736 5r48c3

415289736 29r3c3

417285936 23r7c7

417589236 39r7c7

419285736 5r48c3

421589736 5r48c3

425189736 1r4c36

427185936 58r4c4

427581936 18r4c4

429185736 1r4c36

429581736 1r4c34

451289736 29r3c3

452189736 1r4c36

457189236 39r7c7 (9 solves)

457281936 23r7c7

459281736 1r4c36

471285936 5r48c3

471589236 5r48c3

472581936 5r48c3

475189236 39r7c7

475281936 23r7c7

479185236 5r48c3

479581236 5r48c3

491285736 5r48c3

492185736 5r48c3

492581736 5r48c3

495281736 1r4c34

497185236 5r48c3

497581236 58r4c6

blue wrote:BTW: For line 4, I had 24 possible fills, not 20. Did I miss something ?

No, obviously my count was wrong.

PS i expect now, that i can solve every puzzle with pencil and paper on a weekend (note that uniqueness methods are allowed here for the invalid puzzles

). A copier would make this a bit less boring. And i hate the thought, that i could make a mistake in the "good" puzzle.