Back to where I started. I finally decided to face my original nemesis a third time and see if I'd learned anything in half a year. As told above, my very first attempt on that particular SE 8.4 puzzle failed miserably. As also told, my second attempt sort of solved it, but I wasn't at all happy about how it happened, so I never really counted it. It's been bugging me a bit, but I never looked at the puzzle again until now. Here's what I came up with (unedited p&p steps):

1. AIC (AHS): (6)r3c1 = r3c4 - r1c5 = (6-4)r4c5 = r8c5 - r7c4 = (41-6)r7c89 = (6)r8c9 => -6 r8c1

2. AIC-Loop (ANS, Grouped): (2)r8c4 = (2-4)r7c4 = r7c89 - (4=[9]2)r8c78 - loop => -35 r7c4, -2 r8c1; -9 r8c16, r7c89, r9c79

3. AIC: (2=7)r9c7 - (7=9)r3c7 - r8c7 = (9)r8c8 => -2 r8c8

4. AIC-Hydra (ANS): (#2)r4c8 = r7c8 - (2=94)r8c78 - (4)r8c5 = (#4-1|6)r4c5 = (61)r12c5 - (1=5)r2c9 - r6c9 = (#5)r6c7 => -4 r4c8, -2 r6c7

5. AIC-Hydra (AHS): (#5)r8c6 = (5-#2)r8c4 = (2-4)r7c4 = (41-3)r7c89 = r9c9 - (3=#8)r9c5 => -8 r8c46

6. AIC-Loop (Grouped): (8)r6c3 = r9c3 - (8=3)r8c8 - r79c3 = (3)r6c3 - loop => -2 r6c3; -3 r7c12, r9c2

7. AIC-Hydra: (#8)r4c4 = r6c4 - r6c3 = r9c3 - r9c5 = (8-4)r8c5 = (#4-6)r4c5 = r1c5 - (6=#3)r3c4 => -34 r4c4

8. Kite: (6)r3c1 = r3c4 - r1c5 = (6)r4c5 => -6 r4c1

9. AIC (ANS): (5=1)r2c9 - r2c5 = (1-6)r1c5 = (6-4)r4c5 = r8c5 - (4=92)r8c78 - (2=7)r9c7 - r9c9 = (7)r6c9 => -5 r6c9

10. Kraken Cell: (1378)r1c4 => -26 r1c2 (*a)

11. AIC: (7)r3c8 = r3c7 - (7=2)r9c7 - r7c8 = (2)r4c8 => -7 r4c8

12. Kraken Cell (Nested, ANS): (157)r5c6 => -7 r5c8 (*b)

13. AIC: (2=6)r1c3 - r1c5 = (6-4)r4c5 = r8c5 - (4=2)r7c4 => -2 r7c3

14. Kraken Cell (Grouped): (3467)r4c2 => +8 r6c3 (*c)

15. AIC: (6)r5c4 = r4c5 - (6=1)r1c5 - r1c7 = r4c7 - (1=4)r5c8 => -4 r5c4

16. AIC: (3)r2c5 = (3-6)r3c4 = (6-5)r5c4 = r8c4 - (5=3)r8c6 => -3 r8c5

17. Skyscraper: (6)r1c3 = r5c3 - r5c4 = (6)r3c4 => -6 r3c1, r1c5; stte

Details for (*a, *b, *c):

(*a)

- Code: Select all
`r1c4:`

|(7)-r1c1==============(7)r1c2--|-(26)r1c2

|(1)-r4c7=r1c7-r1c5==|(26)r1c35-|

|(3)-(3=6)r1c3-|r3c4=|

|(8)-(8=6)r4c4-|

=> -26 r1c2

(*b)

- Code: Select all
`r5c6:`

|(7)-----------------------------------------------------------------------------|-(7)r5c8

|(5)-r8c6=(5-2)r8c4=r8c7-(2=7)r9c7-r3c7=(7)r3c8----------------------------------|

|(1)-r5c1=|(7)-------------------------------------------------------------------|

|(6)-r3c1=r3c4-r1c5=(6-4)r4c5=r8c5-(4=92)r8c78-|(2=7)r9c7-r3c7=(7)r3c8-|

|(9)-r5c3=r9c3-(9=2)r9c2-----------------------|

=> -7 r5c8

(*c)

- Code: Select all
`r4c2:`

|(3)-r6c3==========================================|=(8)r6c3

|(4)-r4c5=(4-8)r8c5=r9c5-r9c3======================|

|(6)-r4c45=(6-5)r5c4=r8c4-(5=3)r8c6-(3=8)r9c5-r9c3=|

|(7)-r4c7=r9c7-(7=3)r9c9-(3=8)r9c5-r9c3============|

=> +8 r6c3

Looking back at my first two pathetic attempts (just found the notes and laughed)... yeah, I think I've learned something

For that I can thank a lot of people here, mostly the same ones who kindly replied to this very first post of mine and gave tips on how to move forward. Still a lot to learn.

The one thing that hasn't changed, however, is the pencil&paper solving system I developed last year. It's been stable since October, and I can't think of any more improvements (to the core at least). I'm very happy with it, except for the slowness. That's why my next ambition would be to turn it into software, but we'll see if that ever happens. Until then I'm stuck with p&p. (I'm not principled about that at all -- I'd switch to computerized solving immediately if any program had the features I want, but none do.)

The puzzle (again):

000900030004700600081054002005000000000020308000090060000070800017000000400106050