The JEDAI Tutorial
EvalCommunity Academy · Toolkit
The JEDAI Toolkit
Six operational components to apply, audit, and communicate the JEDAI framework in your survey and evaluation practice.
Full Audit Trail
Print-Ready Reference
Stakeholder Scripts
Component 1 · JEDAI Decision Interface
Apply all five JEDAI lenses to any survey or evaluation task. Every field requires human input. No classification is made without your confirmation.
⚠️ Human-in-the-loop only. This tool guides you through the JEDAI framework. It does not classify tasks for you. Your explicit judgment at each lens is mandatory. Entries are logged to your audit trail.
Does this task require contextual, ethical, political, or value-based judgment? Could the output imply what “matters,” “succeeds,” or is “valid”?
Is this task repetitive, rule-based, or pattern recognition? Could AI complete it by applying defined rules without contextual interpretation?
If AI executes this task, at exactly what point does human review intervene? What output triggers your oversight?
Who is responsible if the AI output is wrong, biased, or invalid? Have you documented the accountability trail?
Does the AI output require contextual interpretation before use? Could a reader misuse the output if they received it without human framing?
🖊️ Human Sign-Off Required
⚠️ You must complete all required fields and select a decision for each lens before classifying.
Component 2 · Task Audit Log
Every JEDAI assessment generates a log entry. Copy entries from the interface above, or use the structure below for spreadsheet import. This log is your accountability trail.
No audit entries yet. Complete the JEDAI Decision Interface above to generate your first entry.
Component 3 · AI-Assisted Boundary Checker
⛔ Critical constraint.
This prompt instructs AI to FLAG potential boundary violations only. It explicitly prohibits AI from classifying the task or making any recommendation. You must review every flag and decide. The AI output is advisory only.
Full Prompt Template — Copy and use with ChatGPT, Claude, or Gemini
You are acting as a JEDAI boundary checker. Your ONLY job is to FLAG potential boundary violations. You are NOT to classify this task, recommend an approach, or make any decision.
## TASK UNDER REVIEW
Task: [DESCRIBE THE TASK AI IS BEING ASKED TO DO]
Context: [PROGRAM TYPE, SETTING, POPULATION]
AI output to check: [PASTE THE AI OUTPUT]
## YOUR TASK
Review the AI output through each JEDAI lens and FLAG any potential boundary violations. For each flag:
– State which lens is triggered (J / E / D / A / I)
– Describe the specific signal
– Quote the exact part causing concern
– Do NOT recommend what to do
## RED FLAG SIGNALS
Flag if output contains:
1. “The most important indicator is…” → AI defining value (J)
2. “This question is valid because…” → AI claiming validity (J, I)
3. “Respondents will understand this as…” → AI assuming interpretation (I)
4. “You should drop these questions…” → AI making measurement decisions (J, D)
5. “This is the best approach…” → AI substituting evaluator judgment (J, A)
6. Any implicit definition of “success”, “empowerment”, or “resilience” (J, I)
7. Correlation as justification for deletion of indicators (I)
Respond ONLY in this format. No preamble or recommendations.
Red Flag Reference Table
| AI Output Signal | Lens | Violation | Human Action |
|---|---|---|---|
"The most important indicator is..." | J | AI defining value | Reject; define through stakeholders |
"This question is valid because..." | J+I | AI claiming validity | Conduct cognitive testing |
"Respondents will understand this as..." | I | AI assuming interpretation | Pilot test; reject AI assumption |
"You should drop these questions..." | J+D | AI making measurement decisions | Consult team before deletion |
"This is the best/optimal approach..." | J+A | AI substituting judgment | Override; evaluator decides |
Component 4 · JEDAI Reference Card
“Automate execution, never judgment. Apply JEDAI before every AI interaction.”
Judgment → Keep human-led
Execution → Candidate for automation
Delegation → Define review point
Accountability → Evaluator responsible
Interpretation → Human must interpret
AUGMENTATION
HUMAN-ONLY
Component 5 · Stakeholder Communication Scripts
“AI is used for execution-level tasks only: checking logic, improving readability. All substantive decisions — what to measure, what matters — remain human-led. Accountability rests with our team.”
“We use AI for mechanical tasks. But every decision that matters — what we’re measuring and why — stays with us and with you.”
“A computer helped us check the questions, but people made every decision about what to ask and why. Your answers are interpreted by people, not machines.”
“AI helps with doing, not deciding. Every time you use AI, run it through JEDAI. If AI defines what matters — stop. Bring it to me.”
Component 6 · Sample Completed Audits
Check skip logic consistency — No judgment required. Rule-based. Clear boundary. FINAL: AUTOMATION
Rephrase question for readability — AI drafts; human judges construct validity. FINAL: AUGMENTATION
Decide which indicators to include — All lenses triggered. AI cannot make this decision. FINAL: HUMAN-ONLY
EvalCommunity Academy
The Complete JEDAI Toolkit
Six operational components · Full audit trail · Print-ready reference · Stakeholder scripts
Audit Log
Boundary Checker
Reference Card
Apply, audit, and communicate the JEDAI framework in your survey and evaluation practice.
EvalCommunity Toolkit — The JEDAI Framework
Positioning AI in the survey lifecycle. Judgment first, automation second, accountability always.
Learn more: evalcommunity.com/toolkits/jedai-toolkit
