The thing is, the lexicon often has lexical "rules" or generalizations.

For a lot of Greek words, stress goes three syllables from the end. Unless there's a long vowel (in Greek).

PO-ly-gon. Three short.

po-LY-go-nal. Four short syllables, stress is three from the right.

po-LY-ga-my. Three short (the final vowel can be long in some cases).

po-ly-HE-dron. That "he" is long, the syllable "heavy" (meaning it's long and/or closed).

This kind of "lexical rule" applies to English words, as well. The rules vary. Over time, words change classes and fall under different rules, word classes may vanish over time. It's the same with other parts of the (morpho)phonology. "Long lived" used to have only a short "i", making it from the past participle of "live," "lived"--someone who was long-lived just "lived long." Now many speakers assume that it's somehow derived from "long life" + -d --> long-lived with that "long i".

In some cases, roots have stresses and suffixes don't; but some roots may be stressless and other 'rules' take over and force stress on a suffix. For example, "-ation" tends to have stress on the second syllable from the end. This isn't just true for English. In other cases we seem to like having stress indicate part of speech: there's a decent list of disyllabic words that different in stress, with initial stress "PRE-sent" being the verb and "pre-SENT" being the noun.

In Russian there are words that seem to have stress on the stem. Others, stress on the prefix or suffix. Some roots are stressless and avoid the stress unless there's nothing after it to take it. Zaliznyak, years back, wrote a nifty book showing how to get from early Common Slavic to Russian in most cases by looking at the accentual properties of stems and suffixes in 700 AD and then making up combinatorial rules for them, how to resolve multiple stresses to a single stress or how to retract it from things that "carry" but won't accept overt stress. Surprisingly, this accounts for the vast majority of usual Russian words. However, Alderete and Crosswhite (et al.) did some research showing that the default stress in Russian depends on how "heavy" syllables are, rather like in Greek (same ideas, different implementation). In other words, Russian has a lot of memorized stresses that speakers make generalizations based on, but if they're given something novel to chew on they have a "default setting" that most speakers agree on. (Happily, in some ways this is similar to, oh, Spanish or Arabic, but again with some differences).

I'm saying "rule" because that's how I first encountered the idea decades ago based on even older Chomsky-Halle ideas of generative grammar, in which some processes were in the lexicon and others in the post-lexical morphology. This has been fought over a long time, and I don't take much of a stance on "where" in the grammar to plop these processes or generalizations. I figure part of the problem is that different speakers abduce slightly different grammars that are barely distinguishable from others', and which may even change over time, but which linguists tend to think of as "the same."

If you think there's something like a mental lexicon, that's where many of these processes happen. If you think that words are somehow generated from roots and other morphemes, then there's a fairly continuous grammar from the phonology, touching upon semantics, and producing words and sentences. Historically this kind of things tends to pop up, where words become slightly bound then totally bound morphemes. So if you compare, say, Lithuanian and Russian you get similar stems and prefixes in both for a number of words but in Lithuanian the negative prefix still comes in between the verb suffix (which is felt to be bound to the stem) and the stem itself. Either there's something building words or you have to say native speakers learn not just the verb but also the negative form of the verb.

And so on.

Personally, I can't decide between all the various arguments over this that I've heard. Each argument seems to crucially and painstakingly avoid certain data and prioritize others. I also tuned out of this over a decade ago when life got complicated (i.e., I was stay-at-home dad and working out of the home at the same time).

I learned that the stressed syllable of a word could be the ultima, the penult, or the antepenult. Classical Greek had pitch accents, but the classes I took substituted loudness for pitch. That was also the nature of the koine Greek, the language spoken in Hellenistic times (and the language of the Christian New Testament). It was also in Hellenistic times that people began to write the three kinds of accents (acute, grave, and circumflex) that we are familiar with in French. I learned the rules for assigning accents to Greek words, but I have forgotten those rules, just as I have forgotten conjugations and declensions.

It makes sense that words derived from Greek, like my examples, would have accents like those in Greek.

It's one of the things that makes learning this language quite tricky.

An ESL teacher told me that about half the verbs in informal speech are of this nature. Imagine trying to teach ESL students the meanings of take out, take in, take over, take up, take down, etc.