Both stub the network. I can write this:
When I mock "**/api/banner" with JSON:
"""json
{ "text": "Welcome {{username}}" }
"""or I can tag the scenario @har:banner and record it. They look interchangeable from where I am standing. Is there a rule for which one to pick, or is it taste?
Not taste. They answer different questions.
@har:<name> is a snapshot of whatever the application actually sent, every request in the scenario, captured once and replayed verbatim. Use it when the payload is not the point and you just want the scenario to stop depending on a live service.Rough test: if you would have to hand-write more than one or two responses to make the page render, record a HAR. If you are asserting on a value you chose, mock it inline, because a recording will drift away from your assertion the next time someone re-records it.
We ended up using both in the same scenario and it works fine. Record the background traffic so the page loads offline, then mock the one endpoint whose shape the test is about. No trouble so far — the mock wins for its pattern and the recording answers everything else.
One thing to watch: tag the scenario @mock as well if it carries an inline mock, otherwise nobody grepping for stubbed scenarios will find it.
One more axis nobody has mentioned: layers.
The mock step is a browser step. It fulfils a page route, so it does nothing for an @api scenario. The @har:<name> tag drives both layers from the same tag — the browser side through Playwright route interception, the API side through the API client’s own recorder and replayer.
So for an @api or @hybrid scenario that needs stubbed traffic, HAR is not one of two options, it is the option.
Have a different question? Ask it.