sound shift
sound shift or root

See alphabetic-change: wholesale shift of phonetic sign interpretation

Universal vocal affinities independent of specific regional sound changes:

A : A E I O U H R l m n : ain air al aph pha va av agh
B : V F P MB W ᵠHUO
C : CH K Q Z S Š G ⇨ DJ ç ⇨ Tsh
D : T DZ DJ δ ⇨ S
E : A E I O U V W fe er el en we
F : V B PH P E W
G : CH K Z DJ Š C Q R
H : * A E I O U HR CH K G
I : A E I O U ain air il vi wi fi
J : I Y G DJ Š LL?
K : C CH G Z DG
L : R N Y J I A …
M : N MW MB lam
N : NG M L ain on vn nv
O : A E I O U H on or
P : B PH F V W
Q : KU CU G KV GV QR KR GR
R : A Ä E L GH air
S : Š Z C CH SH ⇨ G J … δ TH ⇨ T D
T : D TH TZ δ ⇨ S …
U : A E I O U V W phu bu
V : F B PH P E
W : V F B PH P E
X : C KS K CH GH
Y : H G! J Z : C S TZ K DJ ß ∫ : J Š SCH SH Ǧ

There are only a few clusters:
P PH F B V W U
Q C K G J ⇔ D T TZ ⇔ Z δ S Š CH C
M N NG ⇔ L R
all can be dissolved into H/Y/W => AEIOU => ø

Which leads to the wildcard fallacy: any word can be transformed into almost any other word if one just uses sound cluster analyses. This insight led linguist to adhering to only very strict 'formal' allowed sound changes in very narrow settings. Unfortunately, this myoptic approach is not the final solution to the messy nature of language: While it greatly reduces the number of false positives and 'folk etymologies', unfortunately it also creates an abhorrent number of false negatives and wrong rejections of ultimately truly related words. This problem is especially rampant in contact zones of unrelated or distantly related languages, because here borrowing usually does not follow strict formal laws, but rather the anything-goes principle: cucaracha => cockroach => Kakerlake.

There is no simple solution to this narrow path: All changes within the above clusters (and beyond) can and will be observed. Every single word in the dictionary will potentially create an endless source of debates for linguists, securing their profession ad infinitum.

Fortunately if one takes a step back and treats all potential word pairs as an un¬opinionated collections of probabilities of relatives (if a contact between languages is reasonable plausible) the situation is less dire: With the methods of mutational similarity, we might get better insights into relationship trees and can compare connected languages which are traditionally not considered to be related.

Application of the previous general observations to specific questions

The languages of the Neolithic homeland are of outmost interest: It seems very plausible that at the time of Gobleki Tepe, the tower of Jericho and Jerusalem, people in the fertile crescent (including Turkey, South Caucasus and Iran) spoke languages which had at least some impact on slightly later proto languages: Proto-Semitic, Proto-Caucasian and Proto-Indo-European.

Over the course of three millennia, most universal sound changes did occur in the Egyptian dialects.

There are some specific regional Egyptian sound changes (those which linguists adore) but most often it is best to think of sound distributions, in fact probability distributions, if one encounters an Egyptian sign. So instead of strictly reading 𓏏 as T, it is best to keep in mind that it also carried at least its natural partner distribution D T TH δ, if not phonetic values DZ DJ DS in certain dialects / epochs. D and T are the most stable.

In fact the Egyptian alphabet fell apart in very similar ways as the roman alphabet: Today the letter 'C' can be read as
K : curriculum as in kitty
S : center as in santa
Z : centaur as in tsunami (German reading)
Š : cevabcici as in short
SH,CH,GH : …

On the other hand some of the letters (Egyptian and Roman) which initially had different readings later collapsed:
C and K and Q have identical values in German (except for Fremdworte)
V and F have mostly identical values in German.

Some languages exhibit sound changes in transition, such as the ambivalence of μβ as m or b, or مطر as mbetar or mnepar or mlepar or mnetar or mletar

Partners

Less often indirect sound changes occur through partners.
N Ñ NG ⇨ G
R RW WR ⇨ W 𓍢 root,wurzel < WRZ/RWTS •wzerl
Q KW W ⇨ V (lux,lupo,wolf<georg) PIE peculiarity
G GH ⇨ F (enough) pheygh five
T δ ⇨ S
K C ⇨ S (centum/satem)

The more dirty a sound is (broad fourier spectrum), the more it acts as a wildcard for other sounds.

As random mutations all sounds can be changed to other sounds for various reasons, mishearing, reconnection, ommision, reinterpretation, etc etc …

Because of human lazyness and the tendency to save energy, the usual direction of sound change is from hard sounds to soft sounds.
Todo: link list of exceptions
Once a word is (almost completely) worn down, in extreme cases to a single vowel (aqua => eau => 'o'), it will commonly be recombined with specifications (eau-mineral) and ultimately replaced with specifications, grammatical variants, different roots or reset to older versions of the words roots. It can also be re-imported through (un)grammatical variants of neighbor languages.

ablauts

While ablauts in nouns are often free and devoid of semantic patterns, ä,ö,ü,y sometimes derive from clusters (ag/ar…)

kh θ th => real sound change!
BONE ոսկոր ʋɔskɔɾ voskeor (Fuß 𓃀 ʋɔtkʰ) ostheo ὀστέον