Purpose
This site is a triage tool. It helps identify NPRMs that may merit closer review before comment deadlines.
Methodology version: 2026.03.28
Scoring Structure
Pass 1: Metadata/Title Signals
- Agency weighting
- Title lexicon matches (topic and governance signals)
- Broad-scope cues (national reach, scale, health, finance)
Pass 2: Full-Text Structural Analysis
- Section and heading detection
- Rule-type classification (technical/economic/legal/hybrid)
- Quantification, citation, alternatives, and threshold-language signals
- Heuristic risk-band assignment
Supporting Record Context
- Docket supporting-material metadata is collected and listed
- Attachment files are not downloaded for scoring
Two-Pass AI Summarization
- Pass A: a long-form analysis is generated from available rule text and docket context using a structured rubric.
- Pass B: that analysis is compiled into a shorter public-facing summary focused on practical comment opportunities.
- Prompting is type-aware (for example, NPRM vs RFI framing) to reduce category errors in generated output.
Display Conventions
- Presentation percentages are rounded to nearest 5%
- Reviewed overrides may adjust public-facing labels on specific dockets
- Closed-comment documents are hidden from the index
- Backend scoring outputs are static for a fixed scoring model. Time-based pass_3 display boosts are computed in-browser from docket dates (`posted_date` and `comment_end_date`) so ranking can reflect “newly opened” and “closing soon” timing without mutating saved backend scores.
- Comment sentiment shows cluster-weighted support/opposition shares (`+% / -%`) and a net chip. Net is computed as positive% minus negative% using log-weighted cluster counts (`log1p(cluster size)`), with gray shown for no data or net within ±3%.
Limits
- Scores are triage indicators, not legal conclusions
- A low structural-risk signal does not imply low public significance
- AI-generated summaries are informational only and are not legal advice; readers should verify claims in source materials.
Recent Changes
- 2026-03-28: Moved temporal pass_3 effects (freshness and deadline proximity) to client-side rendering; backend stored scores now remain static for unchanged inputs.
- 2026-03-21: Unified detail-page renderer and shared comment-table rendering.
- 2026-03-21: Added comment velocity and sentiment-based pass_3 contributions.
- 2026-03-21: Added structured metadata pages for review status and corrections.