Thanks!
I'm trying to get the main couple of non-standard schemes as good as they'll be, then clean things up a bit and hand out a download probably Wednesday evening.
Both are agreed on using the NAF as a response, differences are in the triggers and target substitutions.
For now I've used the names "crowd NAF" for your idea of switching when a system is firing into a crowd, and "NOC NAF" for going by these coordinate conflicts.
Starting with NOC NAF - a couple of small tweaks gives this an unbelievably good match.
The unbelievable part is how such a weird and generic switch condition makes almost no mistakes across all the other maelstroms!
The only (recent) non Oya problem is cycle 31 Indra choosing Arietis Sector DQ-Y c18, same as standard targets.
Tweak 1 from here-
At any given Maelstrom, let the coordinates of each target system use the Maelstrom as the spatial origin.
If there are any targets for which any one ordinate of the three coordinates equals the negative of any for another target, call it a Negative Ordinate Conflict (NOC).
Any-to-any here was causing an unavoidable problem at Hadad in cycle 33, but changing this to exclude x-to-x, y-to-y and z-to-z worked wonders!
Apparently the conflicts need to be between different dimensions
Tweak 2 from here-
Incidentally, including into the list the systems which are in respite has a positive effect for the NOC and NAF approach! Week 33 then works, and weeks 34–38 are unaffected, although week 39 still maintains an unlikely insistance that everything will be normal this time.
Amazingly that did improve 33 by avoiding a wrong prediction at HIP 21386, and without harming 34-38 in the process, although it was still choosing Aowicha over HIP 11111.
Allowing ineligible targets also causes quite a few mistakes elsewhere so I went with eligible only, sacrifice Oya 33 for now in return for everywhere else all the time (except Indra 31).
The standard targets prediction also looks right either way - there simply isn't the NOCs to trigger anything else so I presume that'll be the INIV guess this time (Lahua, ZE-A c8, BD, 21386, AF-A c22).
Looks like the most it has is FB-X b1-1 vs Bumbo as a possible NOC, but the initial targets don't reach as far as Bumbo and FB-X is offline this week anyway.
The crowd NAF needed close attention to your week 34 breakdown, that one example packs a lot of information for anyone scripting it!
Main tweak/clarification it needed was from here-
if you reach an attack from a control system which could also attack a further target, defer the attack and switch to NAF targeting
This needed to be "further inhab target" to get Oya 34 working, otherwise AF-A c9 decides it also likes the look of AF-A c21 and triggers NAF early rather than hitting DQ-Y b1.
That's then pretty good, it clears up a few similar cases and leaves 34, 35, 37 working.
36 has a problem though - it predicts 5 attacks, the extra being Liu Huang vs HIP 21386.
That happened because it deferred Daruwach vs HIP 3006 but resolved it later with Daruwach vs Aowicha.
I then tried it with only getting 5 attacks if the exact deferred attack gets used, but that broke 34 because Daruwach takes the place of HIP 10778.
38 also had a few problems, the root cause being how it goes for Ardhri vs Cephei Sector AV-Y b2 normally because further targets are unpop (by my earlier tweak).
I don't mind trying to make it allow further unpop targets when some other primary attackers are around to give them a place on the usual Narwhal list - not now though!
Smaller scheme corrections can be done tomorrow if I made a mess of it, though I'll focus more on making something downloadable
View of crowd NAF and NOC NAF together (no standard targets, although the alerts mostly match those anyway)-