← Back to Grand Jury Labs
Grand Jury Gavelly
LiveOverview
Gavelly is an AI-powered compliance tool for California Grand Jury draft reports. Grand jurors upload their draft before publication, and Gavelly checks it against the official California Grand Jury Association style guide and statutory requirements.
It reviews each finding and recommendation for proper formatting, checks that recommendations are addressed to the correct responding parties, flags passive voice and vague language, and surfaces issues that might cause a report to be challenged or returned. Jurors get actionable feedback, not just a pass/fail score.
What Gavelly Checks
# Compliance checks performed
→ Finding structure (evidence basis, clear statement)
→ Recommendation format (imperative, specific, measurable)
→ Responding party identification and role verification
→ Response deadline requirements (90-day rule)
→ Passive voice detection and rewrite suggestions
→ Tone and neutrality (no editorial language)
→ Cross-reference consistency (finding ↔ recommendation links)
→ Finding structure (evidence basis, clear statement)
→ Recommendation format (imperative, specific, measurable)
→ Responding party identification and role verification
→ Response deadline requirements (90-day rule)
→ Passive voice detection and rewrite suggestions
→ Tone and neutrality (no editorial language)
→ Cross-reference consistency (finding ↔ recommendation links)
How It Works
# Step 1: Upload draft report
→ upload alameda-2024-2025-draft.pdf
→ parse Extracted 42 pages, 8 findings, 12 recommendations
# Step 2: Compliance checks
→ check Finding F-1: PASS — properly supported by evidence
→ check Finding F-3: WARN — passive voice in key sentence
→ check Rec R-4: FLAG — responding agency not identified
→ check Rec R-7: WARN — no measurable outcome specified
# Step 3: Detailed feedback
→ report 7 issues found (2 flags, 5 warnings)
→ suggest Inline revision suggestions for each issue
✓ Report ready for review
→ upload alameda-2024-2025-draft.pdf
→ parse Extracted 42 pages, 8 findings, 12 recommendations
# Step 2: Compliance checks
→ check Finding F-1: PASS — properly supported by evidence
→ check Finding F-3: WARN — passive voice in key sentence
→ check Rec R-4: FLAG — responding agency not identified
→ check Rec R-7: WARN — no measurable outcome specified
# Step 3: Detailed feedback
→ report 7 issues found (2 flags, 5 warnings)
→ suggest Inline revision suggestions for each issue
✓ Report ready for review
Tech Stack
Claude (Anthropic)
PDF parsing
Surya OCR
FastAPI
Python
PostgreSQL