P-Gates

What the company appears to be: P-Gates is a self-reported AI output auditing tool built by one developer (Chris Pang) for use in institutional workflows. It claims to detect structural gaps, authority drift, uncertainty loss, and boundary crossings in AI-generated content before human-facing release.

What changed: The project was submitted as part of an OpenAI 2026 hackathon. No prior version or evolution is described; this appears to be a single, self-contained development effort.

Single most important open question: Is there any evidence that P-Gates has been used in real-world AI workflows or tested with actual institutional users?

The description states that P-Gates is a PAL, PECAN, and PEA route auditor for AI-generated reports, recommendations, decision chains, and model handoffs. It separates observations, inferences, assumptions, and unresolved remainder (PAL), identifies consequential crossings, authority drift, provenance loss, uncertainty compression, and anti-backflow failures (PECAN), and surfaces affected people, burdens, proportionality, privacy, contest, remedy, and human governance (PEA).

It also produces a phrase-level trace showing the original phrase, detected problem, safer wording, and responsible framework layer. Results can be exported as JSON or Markdown.

The system is implemented as a FastAPI application with a browser interface, using Python, Pydantic for validation, and GPT-5.6 in live mode. Mock mode is deterministic and requires no API key.

Evidence: The author's own write-up.

Inference: This is a tool designed to audit AI outputs for institutional use, not to make decisions or create authority.

The project description states that P-Gates was built to make transformations in AI output visible before they produce real consequences. It aims to detect when AI language becomes more authoritative than the evidence supports — for example, changing “possible candidate” to “identified suspect.”

It positions itself as a tool for auditing AI outputs, not creating them or making decisions.

Evidence: The author's own write-up.

Inference: P-Gates is positioned as a guardrail or compliance tool for institutional AI use, not a generative system.

The description does not name specific customers or target industries. It implies use in institutional workflows, where AI outputs are processed through reports, summaries, and recommendations.

It is designed to be used by teams or individuals working with AI-generated content that must be vetted before public release.

Evidence: The author's own write-up.

Inference: Likely targets organizations using AI in regulated or high-stakes environments (e.g., law enforcement, healthcare, legal, government).

There is no evidence of a business model or pricing structure. The project is described as a single-person hackathon submission with no mention of monetization, licensing, or customer acquisition.

Evidence: The author's own write-up.

Inference: No commercial model is evident from the description.

P-Gates is built using:

• FastAPI
• Python
• Pydantic for validation
• GPT-5.6 (in live mode)
• HTML, CSS, JavaScript for frontend
• GitHub Actions for CI/CD
• Codex for scaffolding and development

It supports both mock and live modes, with mock mode being deterministic and not requiring an API key.

The system is designed to validate responses locally using Pydantic, and it uses server-side API keys with store=False to avoid storing credentials or private data.

Evidence: The author's own write-up.

Inference: The tool is built for local deployment and privacy-conscious use, with no indication of cloud-based SaaS delivery.

There is no evidence of traction, customers, revenue, or adoption. The project was submitted to a hackathon by one developer (Chris Pang). It has:

• 25 tests
• GitHub Actions CI
• Mock and live modes
• No production deployment or user feedback

Evidence: The author's own write-up.

Inference: This is an early-stage prototype, not a product in use.

The description does not mention competitors. It is unclear whether similar tools exist for AI output auditing or compliance in institutional workflows.

Evidence: The author's own write-up.

Inference: No competitive landscape is evident from the description.

• Single-person development: No team, no external validation.
• No traction or user feedback: Not tested in real-world environments.
• Unverified claims: The tool’s effectiveness is self-reported.
• No commercial model: Unclear how it would scale or monetize.
• Limited testing: Only 25 tests, no production use.
• Hackathon submission: Likely a prototype, not a mature product.

Evidence: The author's own write-up.

Inference: High risk of overstatement in claims and lack of real-world validation.

1. What specific institutional workflows or use cases have you tested P-Gates with?
2. How does P-Gates integrate into existing AI pipelines or decision-making systems?
3. Have you conducted any user studies or feedback sessions with potential customers?
4. What is the expected cost of deploying P-Gates in an organization?
5. Are there plans to support additional LLMs beyond GPT-5.6?
6. How does P-Gates handle edge cases or ambiguous outputs from AI models?