Preset fidelity versus compilation freshness in post-upgrade Babel recovery

After a Babel major-version upgrade, project configuration files such as .babelrc or babel.config.js may reference presets or plugins whose APIs change, causing compilation errors or silent semantic shifts. The recovery decision hinges on two documented CLI mechanisms: --no-babelrc forces Babel to ignore project configuration and default to built-in presets, providing immediate isolation; --delete-cache-dir discards the compiled bytecode cache to prevent stale intermediate representations from persisting under new transform semantics, ensuring a fresh compilation at the cost of a full rebuild. Both mechanisms address recovery from a failed upgrade, but each introduces a distinct trade-off. The first risks omitting intended project-specific presets, while the second shifts the burden to the developer through a complete rebuild and may not remove all persistent configuration state in newer Babel releases.

When preserving project-specific presets is critical versus when ensuring a clean compilation state takes priority, which approach should guide the recovery strategy? Does the --no-babelrc flag reliably isolate the build without losing intended presets, or does --delete-cache-dir fully resolve stale representation errors at the expense of rebuild overhead? How should teams evaluate version-dependent risks when both mechanisms carry uncertainty in major-version upgrade scenarios?