KOTSUBU OS

The description states that KOTSUBU OS is a specification-driven, safety-first AI decision system. It evaluates requests through five deterministic gates before deciding whether to proceed with an answer:

• Input Validation
• Freshness
• Evidence Requirements
• Contradiction Detection
• Risk and Policy Constraints

It returns one of three outcomes:

• GO — sufficient evidence exists to proceed to human review
• WAIT — evidence is missing, stale, ambiguous, or contradictory
• STOP — continuing would violate a risk, safety, or policy constraint

The system includes:

• A readable audit trail
• Explanations for decisions (reason, rules applied, missing evidence, next action)
• Verification timestamp
• Human review requirement
• State transitions from WAIT to GO after evidence update

GPT-5.6 is used only for bounded tasks such as structuring input, extracting contradictions, and producing explanations; it cannot override the final decision state.

The system does not execute the final decision—it merely determines whether to allow a human to proceed.

Inference The system appears to be designed to prevent AI missteps by enforcing a structured decision boundary rather than allowing uncontrolled responses.

The description states that KOTSUBU OS was inspired by repeated failures in long-running AI interactions, such as:

• Rules disappearing between conversations
• Inconsistent answers under same conditions
• Stale information being treated as current
• Numbers misunderstood
• Confident responses generated despite missing evidence

It repositions the AI question from “Can an AI system answer?” to “Can an AI system decide when to answer?”

The core positioning is:

Most AI systems answer. KOTSUBU OS decides when to answer.

This reflects a shift toward safety-first design and human-in-the-loop decision-making, emphasizing trust, continuity, and risk prevention.

Inference The project positions itself as a solution to AI inconsistency and overconfidence, focusing on structured decision-making over raw response generation.

Not evidenced. The description does not identify any specific customer segment or ideal customer profile (ICP). It describes the system’s functionality but does not indicate who would use it or how.

Not evidenced. There is no mention of pricing, monetization strategy, or business model in the description.

The system was built using:

• Python, Flask, HTML/CSS/JS
• OpenAI SDK and GPT-5.6 (used only for bounded language tasks)
• Codex for engineering support during Build Week
• pytest for testing
• Jinja templates, JSON, REST APIs

Key technical features include:

• Separation of probabilistic language assistance from deterministic decision enforcement
• Deterministic core handling:
  • Gate execution
  • Freshness calculation
  • Evidence validation
  • Policy enforcement
  • State transitions
  • Audit facts

GPT-5.6 is used only for:

• Structuring natural-language input
• Extracting contradiction candidates
• Producing readable explanations of deterministic results

The system includes:

• A one-screen judge experience
• Seven synthetic demonstration scenarios
• 41 automated tests with zero failures
• Live verification of GPT-5.6 integration
• Clean-room startup verification
• Public repository without real user data or API keys

Inference The system shows a strong engineering foundation with clear separation of concerns and test coverage. It was built under time constraints but includes reproducible artifacts.

Not evidenced. There is no evidence of revenue, customers, usage metrics, or adoption beyond the hackathon submission.

Not evidenced. The description does not compare KOTSUBU OS to existing systems or identify competitors in the AI decision-making space.

1. No real-world validation: The system is entirely synthetic and has not been tested in production environments or with actual users.
2. Limited scope: The project was built for a hackathon, suggesting it may not be scalable or mature enough for commercial use.
3. Unproven human-in-the-loop model: While the system emphasizes human review, there's no evidence of how this would scale or integrate into workflows.
4. Dependency on GPT-5.6: The use of a proprietary and potentially unstable language model raises concerns about long-term reliability.
5. Single-person team: With only one member, there may be limited capacity for development, maintenance, or scaling.

1. What are the specific use cases where this decision framework could be applied beyond the synthetic scenarios?
2. How would you handle integration with existing AI workflows or platforms?
3. Are there plans to expand beyond the current five gates, and how would that affect consistency?
4. What is the long-term vision for maintaining and updating the policy constraints?
5. How do you plan to validate this system in real-world settings before full deployment?