Yejun Tak / ResearchGitHub repository

Evaluation methods / Engineering-handoff profile

Human-Centered AI
Deployment Readiness Protocol

Examine the gap between an interface that looks ready and the evidence needed to hand it to engineering.

Read the protocolGet the evaluation workbook
v0.1 release candidate · September 2026
A proposed method with a synthetic instructional example. External validation and measured effectiveness have not been established. This profile evaluates engineering handoff; it does not authorize production release.

The problem

Interface polish and functional completeness describe different properties. A readiness judgment needs explicit requirements, observable state transitions and recovery evidence.

The approach

Freeze the criterion, collect the evaluator’s judgment before showing the reference key, and compare that judgment with documented evidence. Preserve disagreements and missing information.

Five distinct measurements

01

Reference-set defect recall

How many frozen reference defects did the evaluator correctly identify?

02

False-ready acceptance

How often was a criterion-nonready artifact judged Ready? Missing judgments and abstentions are reported separately.

03

Expected-recall gap

How far did expected recall differ from observed recall, in percentage points?

04

Recovery coverage

How many applicable recovery scenarios were specified and walkthrough verified?

05

Requirements-omission recognition

How many predefined omissions did the evaluator notice? Artifact requirements coverage is measured separately.

There is no combined score. A high evaluator recall cannot override an unresolved critical issue in the artifact.

Run an assessment

  1. Freeze the artifact, task brief and handoff criterion.
  2. Evaluate and lock findings, expected recall and judgment.
  3. Reveal the reference and adjudicate matches.
  4. Calculate, reconcile evidence and record the owner’s decision.

Try the worked example

Inspect a fictional service-request specification with ten requirements, six recovery scenarios and a separate reference key. Follow the constructed findings through the completed workbook.

Open the fictional artifact

The example contains no real participant results or employer product data.

Materials

Protocol PDF8 pages · Scope, steps, definitions and limitations Evaluator scorecard3 pages · Judgment, findings and decision record Blank evaluation workbookFormula-driven session and batch calculations Worked example PDF3 pages · Reference records and explained calculations Completed example workbookSynthetic inputs with reproducible results Feasibility kit and feedback logTemplates for actual review, use and revision

MCP server + ai-ready Skill

MCP v0.1.0 and Skill v0.1.1 support protocol v0.1-rc.3. Prepare a review and calculate descriptive results from supplied records, keeping agent reviews separate from human observations.

Install and use the tools · Software release · Project website

Related design system

The Peter Tak Design System supplies reusable styles, tokens and component specifications for constructing reference cases. It is a separate implementation resource; external adoption and protocol validation have not been established.

Explore the design system · Read the connection and documented correction

Research context

The method relates explicit requirements and human judgments to evaluation evidence. Its NIST references provide context, not certification or endorsement.

NIST AI RMF 1.0
TEVV-Athlon initial public draft

Version and citation

Tak, Y. (2026). Human-Centered AI Deployment Readiness Protocol: Engineering-Handoff Profile (v0.1-rc.3).

Archived on Zenodo: 10.5281/zenodo.22667623.

Citation metadata