Evaluation AI Faces Lv3 Hurdle; Meta-Evaluation AI "eval000" Launches Outsourced Evaluation Service "eval000 Essential"

TEN PROXY, Inc. (Headquarters: Tokyo, Representative Director: Masayoshi Budou) has launched its meta-evaluation AI platform "eval000 (Ebaru Triple Zero)" and, as its first initiative, has begun offering the outsourced evaluation service "eval000 Essential."

Evaluation AI: What Level Are We At? The Lv3 Hurdle

Automation of evaluations using generative AI has rapidly entered the implementation phase over the past 1-2 years. Cases of introducing LLM-based scoring and feedback generation are increasing in business plans, personnel evaluations, and corporate reviews.

However, when mapping the SAE levels of autonomous driving to evaluation AI, it becomes clear that many current systems are "stuck at Lv3."

▲ Lv3 Structural Problem: The Moral Crumple Zone

Scoring is automated, yet humans must continuously determine "whether the results can be trusted."

Even when using evaluation AI with confidence, the responsibility for decisions still rests with humans. Researcher Madeleine Elish calls this structure the "moral crumple zone (humans serving as a buffer for responsibility)."

eval000 responds to this problem with a meta-evaluation engine.

eval000's Lv4 Design: Automating in Three Layers

eval000's meta-evaluation engine breaks down evaluation into three layers. The design to complete Layer 01-02 without human intervention is the basis for Lv4 equivalence.

Layer

Role

Processing

Content

Responsibility

Lv3 (Reference)

Scoring Automation

Scoring + Recommended Judgment

Humans continuously judge the reliability of results

Fallback Wait

Layer 01

Sensor Input

Primary Evaluation

Scoring by AI Persona Judges

AI Persona Judge

Automation

Layer 02

Driving Behavior

Convergence Processing

Calculation of standard evaluation by meta-evaluation engine

Meta-Evaluation Engine

Automation

Layer 03

Destination Setting

Principle Setting

Final Judgment

Setting and final decision of evaluation objectives and policies

Human (Constant)

Constantly

Human

"The destination is set by the passenger, no matter how autonomous the vehicle. Automating driving behavior and automating destination setting are entirely separate questions."

From eval000.ai Technical Discussion Article "What Level Is Evaluation Autonomous Driving At Now? The True Meaning of HITL in Review AI and eval000"

This design signifies a shift from Human-in-the-Loop (AI's ad-hoc approval) to Human-in-Command (humans focused on principle setting and final judgment).

Overview and Features of "eval000 Essential"

Meta-Evaluation AI Platform

eval000 Essential provides the core technology of the eval000 meta-evaluation engine as an outsourced evaluation service, without requiring annual contracts or specialized knowledge. It delivers the value of meta-evaluation in a simpler, faster, and more accessible way to all evaluation scenarios, such as contest judging, grant reviews, and recruitment evaluations.

Key Features

Multi-faceted evaluation by persona judges: AI persona judges with different evaluation tendencies (e.g., innovation-focused, results-focused, balance-focused, collaboration-focused) evaluate in parallel. Convergence to a common standard evaluation is achieved through fixed-point convergence.

Individual feedback based on rubrics: Generates highly personalized feedback comments based on rubric evaluation criteria.

Initial formalization of evaluation policy (Layer 03): The judging objective and policy are formalized before commencement as "external principles." The standard (default) policy for the business contest field in Essential is "evaluate business potential that can reach the early stage equivalent to Seed to Series A within 4 years."

Simple start and clear pricing: NDA via electronic signature, as fast as the same day. Free estimates. For the business contest field, there is no setup fee, and a user-friendly pricing structure with tiered unit prices (¥1,800 to ¥5,000 per case).

Main Use Cases

Judging for business contests and accelerator programs

Internal new business reviews and intrapreneurship programs

Grant and subsidy selection reviews

Recruitment evaluation and internal award systems

Integration with award-of

"award-of (award-of.net)", operated by our company, is the official Japanese partner of Award Force, a global leader in award management SaaS adopted in over 50 countries and originating from Australia. By integrating eval000 and award-of, we achieve evaluation DX by centralizing entry management, standard evaluation calculation, and automatic feedback distribution.

Structural Challenges in Evaluation

Bias: Evaluation axes vary greatly depending on the judge's specialized field and experience, leading to divided opinions even for the same proposal.

Noise: Scores fluctuate even by the same judge depending on the timing and order of review. Kahneman et al.'s research confirmed an average fluctuation of 19%.

AI Personality: Even when using generative AI for evaluation, models have their own "personalities" in their assessments. In our PoC, we observed differences of up to 12 points (out of 100) between different models for the same proposal.

Errors: Designing rubrics (evaluation criteria) requires expertise and tends to become personalized or superficial.

Comment from Representative Director Masayoshi Budou

"Through the award-of business, I have been involved in numerous contests and came to believe that we needed to technically address the question of 'how to ensure fairness in evaluation.' eval000 is our answer to that question. eval000 Essential is the gateway to delivering that technology to more fields and more easily."

About eval000

eval000 is a meta-evaluation AI platform operated under a joint business operation agreement between patent holder Ryuichi Satoyoshi and TEN PROXY, Inc.

Patent Pending: JP 2026-096693 "Method, Apparatus, and Program for Calculating Standard Evaluation Based on Reconstructed Evaluation by Generative AI" (Ryuichi Satoyoshi)

Service Page: https://www.eval000.ai/service

Technical Discussion Article: https://www.eval000.ai/blogs/post/Lv4

Profile of Patent Holder Ryuichi Satoyoshi

Ryuichi Satoyoshi is a researcher and educator specializing in the Standard Evaluation Generation Principle. He is the inventor of the evaluation reconstruction algorithm using generative AI and the patent holder of the eval000 meta-evaluation engine.

Career and Research Achievements

Current Position: Part-time Lecturer at Tokyo University of Social Welfare (April 2024 - ) teaching subjects related to information ethics, information security, and multimedia. Previous Position: Information and Commercial Education Teacher at Prefectural High School (1992-2024, 32 years).

Degree: Master of Economics, Graduate School of Economics, Yokohama City University (1991).

Affiliations: Information Education Society of Japan / Ricardo Study Group (2000 - ).

Patents and Awards

Patent Pending JP 2026-096693 "Method, Apparatus, and Program for Calculating Standard Evaluation Based on Reconstructed Evaluation by Generative AI" (Pending)

Patent No. 3668491 "Self-Evaluation Ability Measurement Method, Apparatus, and Program" - Awarded Commissioner's Prize of the Japan Patent Office in 2005.

2024: Award of Merit from the Japan Commercial High School Association, Commendation from the Gunma Prefectural Board of Education.

Major Research and Publications

Book: "Human beings and generative AI" (Amazon, 2024).

Peer-reviewed paper: "Structural Analysis of Core Issues in Information Ethics Education" (Information Education 6(1), 2025), and 8 other papers.

Research Project: "Research on the Principle of Standard Evaluation Calculation Based on Reconstructed Evaluation by Generative AI" (Industry-Academia Collaboration, 2026).

Work/System: "Generative AI Evaluation Reconstruction System (Standard Evaluation Calculation Algorithm)" (2026).

International Conference Presentation Scheduled (July 2026)

IOTEIR 2026 — Bridgewater State University (Massachusetts, USA)

Conference Name: IOTEIR 2026 First International Conference "Reimagining Education for the Future"

Date: July 28-29, 2026 (Fully Hybrid)

Presentation Title: From Evaluation to Standard Generation: A Generative AI-Based Model for Reproducible Educational Assessment

Organizer: International Organization of Technology Educator & Innovative Researcher (IOTEIR)

Company Overview

Company Name

TEN PROXY, Inc.

Representative

Masayoshi Budou, Representative Director

Established

1995 (Over 30 years in business)

Business Activities

Operation of meta-evaluation AI platform "eval000.ai"

Operation of contest management platform "award-of.net" (Official Japanese Partner of Award Force)

IT, Marketing, and Business Development Support

Head Office Location

2-11-3-104 Shimouma, Setagaya-ku, Tokyo

URL

https://eval000.ai / https://award-of.net / https://tenproxy.io

Inquiries Regarding This Release

TEN PROXY, Inc. (eval000 Business)

[email protected] Weekdays 10:00 - 18:00 (Japan Time)

Phone: 03-3413-2267

Contact Person: Masayoshi Budou