What I am saying is that your beliefs don't make sense because GSC should have been tested against Custodes which should reveal that one or both of them needed to be changed, same thing for Sororitas against AdMech and so on. It's not like the codexes are internally balanced either, internal balance hasn't been terrible either though.
I literally addressed both of those scenarios in my post and explain how they went awry as a result of
GWs rolling development methodology.
Sororitas were playtested against Ad Mech early on, but not after Ad Mech was finalized and sent to print. Sororitas ended up getting toned down because their
dev cycle (along with the next several books after) was far enough back that that they could account for some of the balance adjustments that were made post-release to the Necron/Space Marine/Death Guard books, etc. whereas Ad Mech (and Drukhari) were too far along in the
dev cycle to capture feedback from the live meta before release.
GWs designers aren't completely incompetent at their work - they understand that they need to make adjustments based on live feedback, but their
dev cycles and lead times don't necessarily allow them to do so.
Basically what I'm saying is that there was likely a point where Sisters, Grey Knights, and Thousand Sons were much more powerful than they were when released, but they got toned down based on live meta feedback prior to going to print, whereas Drukhari and Ad Mech probably started out less powerful than they were prior to release (i.e. they started out looking more like
SM, Necrons, and Death Guard), but got scaled up as a result of their exposure to early versions of Sisters/
GK/TSons which were at the time much more powerful books as they didn't yet have feedback from the released books to tone them down.
GSC on the other hand are an underappreciated book - they are
IMO more powerful than most of the books that came before them (barring Ad Mech/Drukhari), but eclipsed by Custodes who had more exposure vs the sublimely overpowered books that are Tau, Aeldari, and Tyranids. Note though that Custodes are extra-
OP due to day 0 points cuts and balance adjustments because they were perceived as not being powerful enough in playtesting against the armies that released afterwards. So yes,
GSC were playtested against Custodes, but the Custodes that they were playtested against are not the same as the Custodes that were published (which likely got some limited exposure to Aeldari/Nids prior to release), and definitely not the same as the Custodes that everyone has been playing since the day they released (which are even more powerful than they were as they were published.
You can kinda see this play out if you look at Codexes in "batches" in terms of performance - as they are eventually able to adjust based on feedback received and trends observed, just way further down the line than any of us would like or realistically expect it. If you play
40k games between two books that are within a given "batch", the relative power levels of those books tends to be much more consistent than the relative power level of books from across separate batches. I.E. a game between Necrons and Space Marines or between Grey Knights and Thousand Sons (post-release balance updates notwithstanding) will generally be more balanced than a game between Necrons and Adeptus Custodes or Space Marines and Drukhari.
These are the observed "batches" thus far:
1. Necrons, Space Marines, Death Guard
2. Drukhari, Ad Mech
3. Adepta Sororitas, Grey Knights, Thousand Sons
4. Orks, Genestealer Cults
5. Adeptus Custodes*, Tau, Aeldari, Tyranids
6. Chaos Knights, Imperial Knights
*Custodes are hard to gauge because of the aforementioned day 0
FAQ intended to make them even more powerful.
Batch 1 established the baseline, the balance between these books isn't perfect, but you at least feel like you're playing the same game when you play matches between them.
Batch 2 scaled up the power because while these books playtested against Batch 1 early on, they were also heavily playtested against Batch 3 (which were never playtested against Batch 1 themselves) which resulted in power creep prior to release.
Batch 3 scaled down the power - even though Batch 2 scaled up, Batch 3 was far enough back that they could eventually incorporate feedback from Batch 1 after those books released, and thus could be adjusted and balanced relative to live feedback, whereas Batch 2 was too far along for this to occur, and thus were sent to print before Batch 1 feedback could be taken into consideration.
Batch 4 gradually scaled up the power relative to Batch 3 because they received late playtesting relative to Batch 5 which trended them back towards being
OP, but not enough to push them over the edge as they were too far in
dev for Batch 5 to fully influence them and the balancing decisions ahd to be made within the context of their earlier playtest performance vs Batch 3 books. (i.e. - "Hmm these batch 4 books seemed decently balanced against the batch 3 books but are struggling against the batch 5 books now, we better give them a couple buffs to try to level it out, but we won't have time to take them back against the batch 3 books to see how they fare, so best be cautious and conservative with it"). Likewise, Batch 4 would have had the benefit of some of the feedback from Batch 2 and early Batch 3 which would have moderated some of the decisionmaking here.
Batch 5 is batch 5, they never were playtested against Batch 1, 2, or 3 properly, but they were far enough back in development that they could incorporate additional feedback from Batch 3 - these codexes were generally regarded as being middling and overbalanced books by the community. That feedback means that the design studio is being encouraged to be more aggressive with their rules writing and to ramp up the power relative to them in order to be more responsive to commuinty feedback (middling and overbalanced doesn't sell the way exciting and
OP does, after all), and so Batch 5 is born as a result of the designers pushing more aggressively in order to compensate for Batch 3 being too tame.
Batch 6 - too early to say how they'll perform, but I suspect that they will be toned down relative to Batch 5 as a result of early feedback from Batch 4. Batch 4 feedback (especially with regards to Orks) seems to have scared
GW. While Orks are (according to some) not particularly
OP they were able to demonstrate the worst failures and exesses of
GWs failed game design and balancing concepts in a very public manner by effectively tabling what was until then regarded as the most
OP faction in the game being played by someone considered to be one of the communities best players. I think that will translate to
GW downtuning the two knights books relative to the Batch 5 books which would have been too far along in
dev to benefit from the Batch 4 feedback. The fact that Knights could very easily run amok with the meta if
GW doesn't moderate their performance carefully is probably extra incentive for the design studio to end up overbalancing them a bit.
I don't know what books will constitute "Batch 7", or if there is another book that will fall into the "Batch 6" category, but I expect that Batch 7 will end up looking a lot more like Batch 3 as they incorporate further feedback from Batch 5 and the absolute debacle that those books have been.
Even within those batches you can also see trends, Space Marines were more powerful than Necrons, Death Guard were more powerful than Space Marines. That trend continued into Batch 2 with Drukhari being more powerful than Death Guard, but then Ad Mech were less powerful than Drukhari were as the balance issues of the previously released books became more readily apparent, etc.
You can also think of "codex creep" as being less of a "linear" process, and more of a "market" process. Rather than being a constant pattern of escalation, what you actually have is something more akin to a typical stock market with a bull-bear cycle or escalation to the point of correction. I.E. rather than power levels going 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, etc. ad nauseum, the way it actually works is more like 0, 1, 3, 4, 4, 2, 2, 3, 3, 4, 5, 8, 9, 9, 5, 5, 6, 6, 6, 7, 13, 15, 15, 7, etc. You get a couple of small incremental increases in performance, a sudden surge, and then a 50-60% crash, following by a return to small incremental increases, a sudden surge, and then another 50-60% crash again, etc.