A cache-busting query on a page does not bust the CSS and JS it links

Fact

The stock static servers used during development — Python's http.server, most serve-style one-liners — send no Cache-Control at all. Browsers then apply a heuristic freshness lifetime and serve the file from cache without revalidating.

The reflex fix, appending ?v=<timestamp> to the page URL, only creates a new cache key for the document. The <link> and <script> URLs inside it are unchanged, so their cached copies are reused. Reloading gets fresh HTML stapled to a stale stylesheet and a stale script.

Why it matters

The symptom is that an edit did not take effect, which reads as a bug in the edit. New markup appears while the CSS that styles it does not, so the page renders as unstyled fragments; or the HTML updates while the old script runs against it and throws on elements it does not know about. Either way the investigation starts in the code that was just changed, which is correct.

It also poisons automated checks silently: a test drives the previous build and reports on behaviour that no longer exists in the working tree.

How to apply

• Serve development assets with Cache-Control: no-store. For Python that is a four-line subclass overriding end_headers; most other dev servers have a flag.
• Confirm what is actually being executed before debugging a change that "did not apply" — read back a string or symbol the edit introduced, rather than trusting the reload.
• Version the subresource URLs, not the document, when a query string is the only lever available.
• A hard reload fixes it once; it does not fix the next edit, so fix the headers.
• localhost and 127.0.0.1 are different origins with different caches, so switching between them serves the same files past an already-poisoned cache — useful when the tooling driving the browser offers no hard reload.