Alan-Jowett / CoPilot-For-Consensus / 20290803884

78%

push

github

fix: Generate citations from chunks instead of LLM output (#366)

* fix: Generate citations from chunks instead of LLM output (#365)

All production summarizers (LocalLLM, OpenAI, LlamaCpp) return empty
citations arrays. This fix generates citations from the actual chunks
passed as context to the LLM, ensuring citations reference real source
material and preventing hallucination.

Changes:
- Modified summarization/app/service.py to create Citation objects from
  all chunks in context before formatting citations
- Citations now include message_id, chunk_id, offset from actual chunks

Fixes #365

Signed-off-by: GitHub Copilot <copilot@github.com>

* test: Add coverage for citation generation from chunks

Addresses PR feedback from Copilot reviewer. Added comprehensive test
that verifies citations are generated from chunks in context, not from
LLM output (which may be empty or hallucinated).

Test validates:
- Citations are created from all chunks used as context
- LLM-returned citations are ignored
- Each citation has required fields (message_id, chunk_id, offset, text)
- Citations reference actual messages from context
- No hallucinated citations are included

Related to #365

Signed-off-by: GitHub Copilot <copilot@github.com>

---------

Signed-off-by: GitHub Copilot <copilot@github.com>
Co-authored-by: GitHub Copilot <copilot@github.com>

2 of 2 new or added lines in 1 file covered. (100.0%)

4287 of 5219 relevant lines covered (82.14%)

0.93 hits per line