Use cases
=========

Interactive evidence review
---------------------------

Use the Streamlit UI when you want to review evidence in a browser:

.. code-block:: console

   evidence-fetcher-ui

Search from the sidebar, mark results as ``more``, ``less``, ``irrelevant``, or
``unsure``, then rerank the current list or expand the search with a
feedback-refined query.

See :doc:`ui` for a step-by-step walkthrough of the sample app.

Text-based feedback workflow
----------------------------

Use interactive CLI mode when you want the same feedback loop in a terminal:

.. code-block:: console

   evidence-fetcher "retrieval augmented generation evaluation" --interactive

Interactive mode performs the initial search, then lets you mark displayed
results, request a local rerank, or explicitly expand retrieval with a refined
query.

Programmatic retrieval
----------------------

Use the Python API when you want normalized evidence in another application or
experiment:

.. code-block:: python

   from evidence_fetcher import fetch_evidence

   evidence = fetch_evidence("retrieval augmented generation evaluation")

Review caution
--------------

Interactive Information Retrieval is useful for condensing large search result
sets into a smaller set of promising evidence. The same feedback loop can also
drift into cherry-picking if users only reward sources that support an expected
answer.

Use reranking and expansion as prioritization aids, not as proof that the top
results are the whole story. For questions where balance matters, keep checking
contrary, ambiguous, and low-confidence evidence before drawing a conclusion.