Skip to content

Bug: Processing step slider ignored — Analyze always runs full Answer pipeline #49

Description

@suung

Bug: Processing step slider ignored — Analyze always runs full Answer pipeline

Symptom: User selects Chunk or Map on the Processing Steps slider and clicks Analyze, but the app still runs question answering (LLM chat). Chunk-only runs can fail with chat model errors (e.g. Unknown model 'gpt-5.4-mini') even though answer LLM should not run.
Chunks only are not shown

Expected: Slider limits how far the pipeline goes — Chunk (split only), Embed (+ embeddings), Map (+ similarity/scoring), Answer (full LLM analysis). Chunk/Embed should work without selected questions. Chunks are always shown and loaded.

Root cause: processing_steps_slider in Streamlit was UI-only. run_analysis() / process_document() had no step limit — every Analyze always iterated questions and called _analyze_chunksllm.achat.

Fix: Pass max_processing_step from slider into process_document(); stop after chunk/embed/map; allow Chunk/Embed with no questions; add tests in tests/test_processing_steps.py.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions