Attest — proof a human understood this

What the company appears to be: Attest is a self-reported tool that uses AI to assess human understanding in written work by turning essays into live oral defenses. The system records writing process (keystrokes, pauses, pastes) and uses this provenance data along with GPT-5.6 to identify weak points in the text — what the authors call "seams" — which are then challenged in a real-time voice conversation.

What changed: The project description states that traditional assessment methods have failed in the age of AI, and Attest proposes a new approach: instead of detecting AI use, it verifies understanding through live oral defense. This is presented as a shift from "did a human make this?" to "can a human stand behind this?"

Single most important open question: Is there any evidence that the described system has been used in real-world educational or professional settings? The description contains no information about actual deployment, users, revenue, or adoption.

The description states that Attest:

• Turns written assignments into short, adaptive oral defenses
• Captures writing process as an append-only event log with timestamps and edit metadata
• Uses GPT-5.6 to find "seams" — claims the writer is least able to defend
• Conducts live voice conversations over WebRTC for defense
• Provides remediation through Socratic questioning until understanding is demonstrated
• Generates scannable summaries showing what held up, what needed teaching, and transcripts

The system is described as having two main components:

1. Process capture (recording keystrokes, pauses, pastes)
2. AI-driven seam detection and oral defense mechanism

The description states that Attest positions itself as a response to the failure of traditional assessment methods in the age of AI. It claims:

• Traditional AI detectors are "snake oil" that falsely accuse honest students
• Schools' responses (banning AI or buying detectors) are ineffective
• The question "did a human make this?" no longer has a reliable answer
• The real question is "can a human stand behind this?"

The positioning evolves from:

1. A problem statement: AI broke assessment
2. A solution approach: verify understanding, not detect AI
3. A technical implementation: process capture + GPT-5.6 seam detection + live oral defense

The description states that Attest is designed for:

• Teachers who want to assess genuine understanding rather than just flagging AI use
• Students who need to demonstrate comprehension through oral defense
• Educational institutions seeking better assessment methods

The authors mention that the system works "for teachers" by allowing them to rewrite assignments to demand reasoning and examples, and generate oral-defense rubrics for grading.

Not evidenced. The description contains no information about pricing models, revenue streams, or commercial arrangements.

The description states that Attest was built with:

• Next.js 16 (App Router) on Vercel
• GPT-5.6 via the Responses API
• OpenAI Realtime API over WebRTC
• Web Speech API as fallback
• Neon serverless Postgres for data storage
• Vitest and Playwright for testing
• A dependency-free token-bucket rate limiter

The system is described as having:

• Deterministic evidence through TypeScript code
• Model reasoning over facts rather than invention
• Character provenance tracking through replay mechanism
• Auto-degradation capability when API keys are absent
• Live voice path that can't be unit-tested but has fallbacks

Not evidenced. The description contains no information about:

• Revenue or funding
• Customer base or adoption
• Usage metrics or growth
• Product maturity or iteration history
• Market traction or user feedback

Not evidenced. The description does not mention any competitors, market positioning relative to existing tools, or competitive landscape.

The description indicates several potential risks:

1. Unproven real-world deployment: No evidence of actual use in educational or professional settings
2. Technical limitations: Live voice path cannot be unit-tested and has known bugs with API integrations
3. AI model dependency: Relies on GPT-5.6 which may not be publicly available or stable
4. Scalability concerns: Single-person team (1 member) suggests limited development capacity
5. User experience gaps: The system auto-degrades when no API key is present, potentially limiting full functionality
6. Market validation: No evidence of demand or traction beyond the hackathon submission

1. Has Attest been deployed in any real educational or professional environments?
2. What specific evidence do you have that the oral defense mechanism actually improves understanding versus traditional assessment?
3. How does the system handle edge cases like collaborative writing or multi-author documents?
4. What are the actual technical limitations of the WebRTC implementation and how were they addressed?
5. Can you demonstrate any real-world usage data or user feedback from the hackathon or beyond?
6. What is the plan for scaling beyond a single developer team?
7. How does Attest handle different types of content (essays, code, math proofs) in its defense mechanism?