Feedback Auditor

The description states that Feedback Auditor is a tool for teachers to review feedback consistency in anonymous student work. It uses GPT-5.6 to extract structured evidence from submissions and applies four deterministic rules (R1–R4) to flag inconsistencies. These signals are presented to the teacher for inspection, confirmation, or dismissal.

The system includes:

• A frontend built with Next.js, React, TypeScript, Zod
• Use of OpenAI Responses API and Structured Outputs
• Separation of model extraction, rule engine, and human decision-making
• Public demo site with no API key exposure
• Two demo paths: a narrated video fixture and a precomputed Codex audit

Inference The tool is not an automated grader but a consistency-checking assistant.

The author claims that Feedback Auditor was built to address the need for a final consistency check in assessment workflows where anonymous student work and multiple marking passes are used. It aims to reduce token usage and avoid automating grades, instead focusing on helping teachers review feedback.

Inference The positioning evolved from an idea of AI re-grading to one that emphasizes human control and transparency.

The description states that Feedback Auditor is intended for teachers who use anonymous student work and rubric-based marking. It is designed for educators working in assessment environments where consistency checks are needed before results are released.

Inference The target customer is a teacher or instructional designer in an educational setting, likely at the secondary or post-secondary level.

Not evidenced. The description does not mention any pricing model, monetization strategy, or business model.

The system uses:

• GPT-5.6 for structured evidence extraction
• TypeScript rules for review pattern detection
• React and Next.js for frontend
• GitHub Pages for public demo
• Zod for schema validation
• OpenAI Responses API and Structured Outputs

It is described as having two demo paths:

1. A narrated video fixture with fixed workflow
2. A precomputed Codex audit that does not call the API

The author states that a self-hosted deployment can run live GPT-5.6 extraction using the owner’s own key.

Inference The tool is built for educational use and has a clear separation between model processing, rule enforcement, and human review.

Not evidenced. There is no mention of actual users, customers, revenue, or adoption beyond the author's own demonstration.

Not evidenced. No information is provided about competitors or market positioning.

• The tool is described as a hackathon submission with no evidence of real-world usage.
• It is not clear whether the author has tested it in actual classrooms or gathered feedback from educators.
• The public demo does not involve live API calls, which may limit understanding of its real-world utility.
• The lack of any mention of monetization, customers, or traction raises questions about scalability or commercial viability.

1. Has the tool been tested in real classrooms or with actual educators?
2. What is the source of the rubric and student data used in the demo? Are they representative of real-world use cases?
3. How does the author plan to scale beyond a single-person development effort?
4. Is there any interest from educational institutions or platforms (e.g., LMS providers) in adopting this tool?
5. What are the actual limitations of the current rule engine, and how might it evolve?