End-to-end pipeline orchestrator + filter/extraction quality bundle#28

Closed

smodee wants to merge 21 commits into

mainfrom

feat/end-to-end-orchestrator

Commits on May 28, 2026

Fix Excel-serial parsing in load_questions
smodee
committed
Add build_forecast_question factory and load_question_by_id
smodee
committed
Add orchestration/test_questions.csv with q7 + q12
smodee
committed
Add OpenAI price table and cost estimator
smodee
committed
Add bioscancast.orchestration package and persistence helpers
smodee
committed
Implement end-to-end pipeline orchestrator in bioscancast/main.py
smodee
committed
Add scripts/run_pipeline.py wrapper and gitignore data/runs/
smodee
committed
Handle sets and dated OpenAI model aliases (surfaced by live runs)
smodee
committed
Bypass heuristic scoring for dashboard-injected results (#14)
smodee
committed
Enrich dashboard titles and snippets with pathogen-specific text (#14)
smodee
committed
Lower heuristic_keep_threshold from 0.72 to 0.65 (#13)
smodee
committed
Add suspected_deaths to controlled metric_name vocabulary
smodee
committed
Require the quote field to contain the figure (extraction prompt)
smodee
committed
Adaptive retrieval_top_k when filter survivors are few
smodee
committed
Exempt dashboard-bypassed docs from cap_per_domain_and_type
smodee
committed
Add per-stage cost line-items to orchestrator epilogue
smodee
committed
Add case-insensitive layer 4 to the hallucination guard
smodee
committed
Add a relevance term to the search-stage score

smodee
and
claude
committed
Fix stale dashboard URLs and make pathogen routing tolerant

smodee
and
claude
committed
Add national/international news outlets to Tier 3

smodee
and
claude
committed
Add opt-in soft fallback for the no-LLM filter path

smodee
and
claude
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

End-to-end pipeline orchestrator + filter/extraction quality bundle#28

End-to-end pipeline orchestrator + filter/extraction quality bundle#28
smodee wants to merge 21 commits into
mainfrom
feat/end-to-end-orchestrator

Commits on May 28, 2026

Fix Excel-serial parsing in load_questions

Add build_forecast_question factory and load_question_by_id

Add orchestration/test_questions.csv with q7 + q12

Add OpenAI price table and cost estimator

Add bioscancast.orchestration package and persistence helpers

Implement end-to-end pipeline orchestrator in bioscancast/main.py

Add scripts/run_pipeline.py wrapper and gitignore data/runs/

Handle sets and dated OpenAI model aliases (surfaced by live runs)

Bypass heuristic scoring for dashboard-injected results (#14)

Enrich dashboard titles and snippets with pathogen-specific text (#14)

Lower heuristic_keep_threshold from 0.72 to 0.65 (#13)

Add suspected_deaths to controlled metric_name vocabulary

Require the quote field to contain the figure (extraction prompt)

Adaptive retrieval_top_k when filter survivors are few

Exempt dashboard-bypassed docs from cap_per_domain_and_type

Add per-stage cost line-items to orchestrator epilogue

Add case-insensitive layer 4 to the hallucination guard

Add a relevance term to the search-stage score

Fix stale dashboard URLs and make pathogen routing tolerant

Add national/international news outlets to Tier 3

Add opt-in soft fallback for the no-LLM filter path