Independent research project — not affiliated with the South Portland School Department or City of South Portland. AI-assisted analysis with full source transparency. Learn more.

About This Project

What This Is

SP Budget Watch is an independent, volunteer analysis of the South Portland School Department's proposed FY27 (2026-2027) budget. The project translates the raw budget — meeting transcripts, presentations, spreadsheets, and committee discussions — into forms that residents can actually use.

The analysis is organized around 14 community stakeholder profiles (personas) representing the range of people affected: parents, teachers, taxpayers, board members, journalists, community advocates, and students from elementary through high school.

How It Works

Evidence Collection

Source materials are collected from public sources: the South Portland Schools website, meeting video platforms (Vimeo), the city's agenda portal (Diligent Community), and official budget documents. An automated pipeline checks for new materials daily.

Analysis Pipeline

  1. Collection: Meeting transcripts (auto-generated from video), budget documents, and presentations are downloaded and stored.
  2. Normalization: Raw files are converted to structured markdown for consistent analysis.
  3. Interpretation: Each meeting is analyzed through all 14 stakeholder perspectives using AI (Claude). The AI identifies what matters most to each persona based on their defined concerns and information needs.
  4. Cumulative Narrative: New interpretations are folded into running per-persona narratives, tracking how each stakeholder's understanding evolves across meetings.
  5. Briefings: Forward-looking briefings highlight what changed, what to watch for, and open questions — tailored to each perspective.

AI Usage

This project uses Claude (by Anthropic) for meeting interpretation, narrative synthesis, and briefing generation. The AI operates within structured prompts that specify persona characteristics, output schemas, and evidence requirements. All AI-generated content is clearly labeled.

The AI does not make editorial judgments or advocacy claims. It identifies facts, tracks positions, surfaces discrepancies, and frames information from each stakeholder's perspective. When evidence is insufficient or ambiguous, the briefings say so explicitly.

Limitations

How to Verify

Every briefing includes a "Where To Verify" section pointing to primary sources. Key verification paths:

What This Is Not

Open Source

The full analysis pipeline, evidence pools, and source code are available on GitHub. Contributions, corrections, and feedback are welcome.