Installed is not the same as importable or runnable

Trace the interpreter used by the failing command

A command launched by an IDE, scheduled task, shell alias or shebang may use a different Python than your terminal. First print sys.executable, the major/minor Python release and the location of the imported module. Use the same interpreter for -m pip; a bare pip on PATH can refer to a different installation.

python -c 'import sys,site;print(sys.executable);print(sys.version);print(site.getsitepackages())'
python -m pip --version
python -m pip show requests
python -c 'import importlib.util;print(importlib.util.find_spec("requests"))'

Separate four diagnoses

ModuleNotFoundError may mean a package is absent in the active interpreter, that your project shadows its name, or that an optional import path is unavailable. ImportError can arise inside an installed package. A native extension may fail only at load time if its architecture or ABI does not match. python -m pip check checks declared dependency relationships but cannot certify that binary imports or APIs work on the target platform.

Exercise with controlled shadowing

In a disposable folder, name a file json.py and attempt to run code importing Python's standard json module. Inspect the imported module's __file__ when available. Rename that local file and remove only its own stale cache; do not delete the entire system cache. This demonstrates why working directory and filenames matter beyond package installation.

Repair and acceptance

Choose the correct interpreter, use a project-local virtual environment where available, install declared dependencies from trusted sources, and run a minimal API call plus your project's tests. Capture OS, architecture, Python version and pip's reported version. A successful pip install is necessary for some repairs, but not proof the original application now works.