Two of our calculators had a results table that never filled. No error, no red console line. The cause was one repeated word.
<input id="years" placeholder="10">
...
<tbody id="years"></tbody>$("#years").innerHTML = rows; // $ is document.querySelectorquerySelector("#years") returns the first element with that id, which is the input. Setting innerHTML on an input is allowed and does nothing. The table body, further down the page, stays empty.
We found it by parsing all 2,865 calculator pages and checking things the existing audits did not: script syntax, label for targets, ARIA references, getElementById names, and repeated ids. Only the last turned anything up: four pages. Two had the empty-table bug. One worked by luck, because its selector went through a descendant (#rounds tbody). One sent an in-page link to a button instead of the section it named.
We have not screenshotted the broken pages; the diagnosis comes from reading the selector and the document order, so treat "the table stayed empty" as the likely behaviour, not a recorded one. Our unit tests missed it because they test the maths and the markup, not what the script does to the page.
The fix is four renames. The lasting part is check_duplicate_ids.py, which fails the build if any page repeats an id. The next step is a smoke test that loads each page, presses the main button and asserts that the result area is not empty.
Write-up with the full table of checks: docs/reports/2026-10-21-duplicate-ids.md.