ProofPath AI YES — Track 1: AI Application Development

Problem

Students often encounter claims online without a clear method for checking whether the evidence they have actually supports, contradicts, or is insufficient for the claim. ProofPath turns that reasoning step into a small, transparent educational workflow.

Solution

Users enter a claim and a piece of evidence. ProofPath classifies the relationship as SUPPORTED, CONTRADICTED, or INSUFFICIENT and returns bounded probabilities. It is intentionally limited to educational evidence assessment and does not claim to be a truth detector or scientific validator.

AI and technical design

The application uses a lightweight evidence-classification model behind a single analysis path. The same bounded analyze(claim, evidence) function is used by the human-facing interface and the structured WebMCP tool. This avoids duplicated verdict logic and keeps the same safety limits across interfaces.

Evaluation and testing

The project includes automated regression tests for the model and UI plus a real-browser WebMCP runtime canary. The canary discovered the analyze_evidence tool in headed Chrome, successfully invoked a normal evidence case, and verified that blank evidence fails closed instead of producing a confident result.

Responsible AI

ProofPath uses an explicit claim ceiling: EDUCATIONAL_EVIDENCE_ASSESSMENT_ONLY. Missing required input returns a BLOCK state. The project is not presented as a medical, legal, scientific-validation, or general automated fact-checking system.

User experience

The app remains usable as an ordinary web form. Students can enter a claim and evidence directly, inspect the resulting classification, and see when the system refuses to classify because necessary evidence is missing.

Track 1 submission package

The submission includes a slide presentation describing the problem, architecture, evaluation, responsible-AI constraints, and impact; a video walkthrough/live demo; source code and a live deployment; and the required packaged deliverables.

Built and verified with

JavaScript, HTML, CSS, lightweight machine-learning classification, WebMCP, GitHub Actions, headed Chrome runtime verification, and ChatGPT-assisted implementation/test/documentation workflows.

Source and live demo

Public repository: https://github.com/lbkapcsolat-dotcom/ProofPath/tree/webmcp-challenge

Live deployment: https://elastic-cloud-7bddb2h.shipstatic.com

Video walkthrough: https://youtu.be/sN-UJNyZoL4

Built With

• chatgpt
• css
• github-actions
• html
• javascript
• machine-learning
• webmcp