I think the analysis of the ''interested but non-expert observer' (and tbh, many not well phonics trained teachers) can be affected, though, by the fact that children can APPARENTLY learn to read by learning words as wholes while what they are ACTUALLY doing is creating their own 'inferred phonics system'.
DS learned to read before he started school, primarily by himself. He has a very retentive memory, and would recite the books he had had read to him to himself, turniing over the pages and 'following' thee words. From that recitation, he matched words to their written representations, which allowed him to recognise those words in other contexts but also recognise 'bits of words' - graphemes - in other words.
I observed this - partly because i thought he seemed keen to learn to read, bought the Phonics handbooks... and discovered he could already read. However, from really minute observation, it wasn't that he could only read words that he knew 'as wholes', he could use inferred phonics knowledge to decode new words too.
Obviously it is much more efficient and direct to teach the pghonic code directly, rather than ask all learners to infer it. But I do wonder how many 'my child learned to read through whole words' parents / teachers actually had children who worked out the phonic code and really used phonics to read.