-- Per-char iteration over a non-trivial text with `at s i`.
-- Before the fix, `at s i` did `s.chars().collect::<Vec<char>>()` on every
-- call, making this loop O(n²) and the apparent culprit behind the 222k-token
-- "OOM" reports from NLP personas. Now `at` is allocation-free for the
-- in-range case and the loop scales linearly.
-- Count uppercase ASCII letters in a mixed-case word.
upper-count s:t>n;l=len s;n=0;@i 0..l{c=at s i;>=c "A"{<=c "Z"{n=+n 1}}};+n 0
-- Iterate the same word twice, once forwards, once with negative indices,
-- and confirm both paths see the same characters. For index i, the matching
-- negative index is i - l (so 0 maps to -l, l-1 maps to -1).
roundtrip s:t>n;l=len s;ok=0;@i 0..l{a=at s i;j=-i l;b=at s j;=a b{ok=+ok 1}};+ok 0
-- run: upper-count HelloWorld
-- out: 2
-- run: upper-count ALLCAPS
-- out: 7
-- run: upper-count nouppershere
-- out: 0
-- run: roundtrip abcde
-- out: 5
-- run: roundtrip x
-- out: 1