How would you choose the right mix of frontend tests for a product feature without making the suite slow and brittle?

I start from product risk and known failure modes, then make ownership and determinism part of the strategy. Contract checks should detect when frontend assumptions drift from the API, critical browser tests should use controlled data and stable user-facing selectors, and flaky tests should be investigated rather than normalized through retries. I track suite duration and failure quality, quarantine only with an owner and expiry, and add regression coverage at the lowest layer that reproduces a production escape. Visual, accessibility, and performance checks are useful for specific risks, but I add them deliberately instead of building a large suite that produces noise nobody trusts.