ALPHAFLOW_ISSUE_1.md

Mission: Unmask the Cursed Bias (Easy • 2 pts)

Minimal Reproducible Example (MRE)

1. Backend running:

uvicorn Jujutsu-Quants.app.adk.main:app --reload

2. Send sample request:

curl -s -X POST http://localhost:8000/api/v2/report -H "Content-Type: application/json" -d '{"question":"Test Tesla hypothesis"}' | jq

3. Expected: response JSON contains bias list. Current bug: entities are hardcoded; update detector to accept list via config or request.

Acceptance Criteria

• Add config-driven or request-driven entity list.
• Update tests and README with configuration example.

Task

Make BiasDetector usable by removing hardcoded TARGET_ENTITIES and allowing entities from:

• config (AGENT_CONFIGS['bias_detector']['entities'], optional)
• request-time input (method argument)
• fallback to simple title keywords

What to Implement (small scope)

• Add optional parameter entities: List[str] | None to detect(articles, entities=None).
• If entities is None, read from config; if still empty, derive from titles by picking capitalized tokens.
• Keep existing proximity logic unchanged.

I/O

• Input: articles: List[Dict{title:str, content:str}], optional entities: List[str].
• Output: existing bias result list (unchanged schema).

Example

articles = [{
  'title': 'Apple beats expectations',
  'content': 'Analysts say Apple showed unprecedented growth.'
}]

det = create_bias_detector()
out = det.detect(articles, entities=['Apple'])
# out[0]['entity_focus'] == 'apple'

Acceptance Criteria

• Hardcoded list removed; method accepts optional entities.
• If no entities provided, detector derives at least one from titles.
• Existing output shape unchanged; tests pass on simple sample above.

Hints

• To derive entities quickly: re.findall(r"\b[A-Z][a-zA-Z]{2,}\b", title) and lowercase them.

Labels

good first issue, agents, bias