Title case is not one convention
There is no single correct title case. The major style guides disagree in specific, consequential ways. Chicago capitalises all major words and lowercases articles, coordinating conjunctions and prepositions regardless of length — so "Through" stays lowercase. AP capitalises prepositions of four letters or more, so AP writes "Through" capitalised and Chicago does not.
APA capitalises words of four letters or more, MLA capitalises all principal words regardless of length. All of them capitalise the first and last word whatever it is, which is the one rule they share.
Sentence case — capitalising only the first word and proper nouns — has become the dominant convention for interface text and increasingly for headlines, because it is easier to read at a glance and has no edge cases to get wrong. If you have no house style, sentence case is the defensible default.
Case is not a simple character mapping
Uppercasing looks trivial and is not. German ß uppercases to SS, so a three-letter word becomes four characters and the transformation is not reversible — lowercasing SS does not restore ß. Turkish has dotted and dotless i as separate letters, so uppercasing i correctly yields İ in Turkish and I in English, and getting this wrong has broken real software in ways that were hard to diagnose.
Greek final sigma changes form depending on position in a word, and some ligatures expand on uppercasing. Locale-aware case conversion exists precisely because the correct answer depends on language, and a naive per-character mapping produces text that is wrong for a substantial number of users.
For programming identifiers there is a related subtlety: converting between camelCase, snake_case and kebab-case requires deciding how to treat acronyms. Does parseHTTPResponse become parse_http_response or parse_h_t_t_p_response? Conventions differ, and round-tripping between styles is not always lossless.
Where all-caps hurts and helps
Text set entirely in capitals is measurably slower to read in continuous prose. Lowercase letters have distinctive ascenders and descenders that give words a recognisable outline, and capitals reduce every word to a uniform rectangle, removing that cue. The effect is small for a two-word label and substantial for a sentence.
It is also a problem for assistive technology. Some screen readers interpret all-caps words as acronyms and spell them letter by letter, so a heading in capitals can be read out as individual letters. The robust approach is to write text normally and apply text-transform: uppercase in CSS, which leaves the underlying text intact for screen readers and for anyone copying it.
Capitals do work for short labels, buttons and small amounts of display type where the aim is emphasis rather than reading. In that context, adding letter-spacing helps, since capitals are designed to be spaced more loosely than lowercase and look cramped at default tracking.