ProofPath: Disaster Evidence Map
Inspiration
After a disaster, people may need to prove where they lived or what they owned while dealing with displacement, damaged records, and urgent recovery tasks. FEMA's guidance describes the kinds of ownership and occupancy records applicants may need. Reporting from the 2025 Los Angeles wildfires also showed survivors struggling to locate and upload insurance documents.
We wanted to explore a focused way to help people make sense of records they already have, without pretending software can replace a caseworker or decide whether a claim will be approved.
What it does
ProofPath is a proposed UGREEN AI NAS Skill that compares a checklist chosen by the user with documents in one folder they authorize. It aims to return a requirement-by-requirement map of candidate evidence, apparent gaps, conflicting details, and items that need human review. Each finding should point back to the source file.
“Not found” means not found in the selected folder and readable files; it does not mean the document does not exist elsewhere. ProofPath does not determine eligibility, certify that evidence is sufficient, or submit claims.
How we built it
We began with the user’s task and worked backward: choose a checklist, choose a folder, find candidate files, read relevant document content, and compare what was found with each requirement. The planned Skill uses the file and document capability families described in UGREEN’s public Hackathon Starter Kit, through supported UGOS operations and registered expert agents.
Our current deliverables are a Round 1 proposal and a pre-test SKILL.md package. We have not run ProofPath on a UGREEN NAS. The official developer archive still needs to be inspected so we can confirm exact agent names, commands, permissions, supported file formats, and validation requirements before describing the workflow as operational.
Challenges we ran into
The main challenge is balancing usefulness with accuracy. A file may look relevant while containing the wrong address, date, or name. ProofPath therefore needs to show the supporting source and mark ambiguity for review, rather than call a document valid proof.
The other challenge is technical verification. The Starter Kit gives a high-level capability list, but the downloadable developer package defines the actual supported environment. Until we can validate the Skill against that package and on the assigned NAS, we cannot claim the evidence-mapping workflow works end to end.
Accomplishments that we're proud of
We narrowed a broad “AI for documents” idea into a specific, consequential workflow with a visible result: a checklist coverage map linked to source files. We also defined strict boundaries around folder scope, read-only access, uncertainty, and human review. Our current proposal is 384 words, below the Round 1 limit.
What we learned
A useful AI workflow needs more than a convincing answer. People need to see what source supports each result, what remains uncertain, and exactly what the system searched. We also learned to separate documented platform capabilities from assumptions and to treat privacy claims as something to verify in the actual runtime.
What's next for ProofPath: Disaster Evidence Map
First, verify the official UGREEN Skill Developer Package and update the Skill to match its registered agents and supported operations. Then validate it on the assigned NAS with synthetic records covering exact matches, missing items, conflicting details, and unreadable documents. We will measure whether candidate matches point to the right sources and make every limitation clear in the demo.
Our goal is a small, dependable tool that helps people see what they have gathered and what still needs checking—without making promises about the outcome of a claim.
Sources
Built With
- accessibility
- ai
- ai-agents
- artificial-intelligence
- civic-tech
- disaster-recovery
- document-management
- document-search
- evidence-management
- information-retrieval
- local-ai
- nas
- ocr
- personal-data
- preparedness
- resilience
- ugos
- ugreen
- workflow-automation
