Coverage snapshot
Product Workspace
Dashboard
Operational snapshot
See the whole SEC pipeline at a glance.
Track committee meeting coverage, proposal extraction, review backlog, and model readiness from one place before you move into detailed review or forecasting work.
Latest training
No training run yet
Create training runs from the backend or the upcoming training workflow.
Current label distribution
Outcome Mix
Recent training runs
Model Log
Recent forecasts
Prediction Log
Decision Intelligence
Compare SEC outcomes without losing the evidence.
Explore proposal volume, approval patterns, label coverage, and review maturity by year, specialty, proposal type, applicant, or outcome.
Reading the local proposal database.
Filtered comparison
Proposal count by year
Filtered records
Outcome mix
Exact values
Comparison table
Ingestion Command Center
Refresh new CDSCO recommendations in one run.
The full pipeline now scrapes live SEC metadata, resolves blocked CDSCO download wrappers, downloads missing PDFs, extracts text, rebuilds proposal rows, refreshes the database, and retrains the forecasting model.
One-click pipeline
Full Refresh
Use this when CDSCO publishes new recommendations.
Pipeline status
Latest Ingestion Run
Data quality
Extraction Completeness
PDF ingestion helper
Resolved Download Batch
Open the resolved URLs in your browser and save the PDFs into
data/raw/pdfs.
Integrated download
Resolve And Download PDFs
CDSCO session credentials are read from the server environment and are never entered in the browser.
Resolve targets first, then download new PDFs directly into the local raw PDF folder.
Download results
Latest PDF Download Attempt
Pipeline assets
Source Files
SEC metadata feed
Recent Meetings
Historical Recommendations
Search the CDSCO SEC archive
Results are sorted by latest meeting first and include PDF/source links where available.
Matches
Historical Recommendation Results
Proposal dataset
Proposal Rows
Each row is one parser-generated candidate agenda item from an SEC meeting PDF. It includes extracted drug, applicant, proposal text, recommendation text, and an inferred label.
Manual validation
Review Queue
The review queue is the subset of proposal rows that still need human validation, especially `unknown`, `low`, and `medium` confidence recommendations.
Focused review workspace
Edit Proposal And Finalize Label
Select a proposal from the review queue to edit the extracted fields and save the final review outcome.
Launch a model run
Training Control
Training currently uses only non-unknown medium/high confidence rows from the proposal dataset.
Training registry
Recorded Runs
Dataset readiness
What Training Can Actually Use
Run notes
Latest Training Summary
Forecast
New Proposal
Result
Outcome Forecast
Historical predictions
Prediction History
Comparable cases