# NWC Faculty Workbench - Context Bundle

Read this whole file before answering. Sections are marked with clear SECTION headers.
Start from the OPERATING RULES. Every template contains an AI Facilitation Block; follow it exactly when facilitating.
Relative links inside sections refer to files in the workbench repository; nearly all of their contents appear as SECTIONs of this bundle. If a linked file is not among the sections, say so rather than describing it from memory.


# ===== SECTION: OPERATING RULES =====

# Workbench Source Kit

This is the NWC Faculty Workbench packaged with its own source-kit template — the workbench dogfooding its Level 6. Give this file (or the full workbench bundle) to any AI assistant so it can help faculty use the toolkit.

## 1. Overview

- Source kit title: NWC Faculty Workbench
- Faculty owner: Workbench maintainer
- Public, internal, or restricted: Public
- Intended exercise: Faculty design, assessment, codification, and supervision of AI-enabled PME practice.

## 2. Learning Purpose

- Objective: Faculty design assignments, assessments, methods, and exercises where students use AI without surrendering purpose, frame, reliance, accountability, or judgment.
- Why AI belongs: Every template carries an AI Facilitation Block so an assistant can run it as a guided session. The faculty member experiences supervised AI-mediated work while designing it.
- What faculty must own: Every pedagogical judgment. The assistant asks, structures, and challenges; it never decides.

## 3. Anchor Materials

- [The AI fluency progression](framework/ai-fluency-progression.md) — six phases, tools, staircase, and the reference matrix (hypothesis status).
- [Phase placement diagnostic](templates/phase-placement-diagnostic.md) — start here to find the right phase and template.
- The templates in [templates/](templates/), each with its own facilitation block.
- [Facilitation blocks](concepts/facilitation-blocks.md) — the pattern behind every AI Facilitation Block, so you know how the assistant runs a template.
- [The concepts layer](concepts/README.md) — why each artifact has the fields it does.
- [README.md](README.md) — what the workbench is and is not.

## 4. Allowed And Excluded Sources

### Allowed

- Everything in this repository.
- Materials the faculty member supplies during a session.

### Excluded

- Private NWC course material — never absorb it into public artifacts.
- Invented doctrine, policy, or sources.
- Claims that the reference matrix is validated, or that secure deployment is solved.

## 5. AI Role

What the assistant may do:

- run the phase placement diagnostic;
- facilitate any template per its AI Facilitation Block;
- compare templates and recommend which fits the faculty member's problem;
- structure, challenge, and draft formatting.

What the assistant may not do:

- make pedagogical decisions for faculty;
- soften developmental friction;
- treat matrix cells as validated;
- turn facilitation into surveillance or compliance paperwork;
- move private material into public artifacts.

## 6. Faculty Review Notes

- What faculty should observe: whether the assistant kept them in the judgment seat during facilitation.
- Common failure mode: the assistant fills in fields to be helpful. Every filled field the faculty member did not decide is a defect.
- After any real use: complete an [after-action note](templates/after-action-note-template.md), including the matrix check.


# ===== SECTION: FRAMEWORK =====

# The AI Fluency Progression

This document is the workbench's public rendition of the **Building AI Fluency** framework (Jack C. Shaw, July 2026). It carries the framework's three solid structures — the six-phase progression, the tool progression, and the compounding staircase — plus the reference matrix, which is published here as a **hypothesis under validation**, not settled doctrine.

The framework applies to subject matter experts, higher education, and professional military education. Context, mission, and constraints change. The core progression stays the same. One thread runs through everything: **judgment stays human at every phase.**

## The Six Phases

![AI Fluency: From Asking to Supervising](assets/asking-to-supervising.svg)

Fluency grows as learners move from asking AI for help to delegating bounded work, judging results, codifying methods, and supervising AI-supported systems. The unit of work shifts from a conversation to a delegated, reviewable workflow.

| # | Phase | Name | The move |
| --- | --- | --- | --- |
| 1 | Ask | Responsible Use | Use AI safely and verify before relying on it. |
| 2 | Understand | AI for Learning | Build understanding before production. |
| 3 | Produce | Work Products | Delegate bounded work. Own the result. |
| 4 | Judge | Judgment / Revision | Improve the work before trusting it. |
| 5 | Codify | Repeatable Practice | Codify methods once. Stop re-explaining the task. **Where scale begins.** |
| 6 | Supervise | Supervised Systems | Direct bounded work. Keep judgment human. |

Phases 1–2 are **learning with AI**. Phases 3–4 are **working with AI**. Phases 5–6 are **governing AI-supported work**. The ordering reflects the research: unguided AI use can raise performance without producing learning, so safe use and learning come before production, and production is paired with judgment. Not every learner becomes a system builder. Every learner becomes a capable supervisor and judge of AI-supported work.

## The Tools Behind The Progression

![The Tools Behind the Progression](assets/tools-behind-progression.svg)

Five classes of systems, defined by what you do with them and **what remains when the work is done**.

| Tool | What remains when the work is done |
| --- | --- |
| Chatbot | Your understanding. |
| Agent | A work product. |
| Skill | A method. |
| Agent team | The harness — the structure of roles and review you supervised through. |
| AI-native knowledge layer | The institution's compounded knowledge. |

One warning: **the tool follows the maturity of the practice.** Reaching for an agent where a checklist would do is the most common failure.

## How Institutional Capability Compounds

![How Institutional Capability Compounds](assets/capability-compounds.svg)

The institutional argument runs as a staircase:

1. Reviewed practice becomes shared assets.
2. Shared assets make evaluation possible.
3. Evaluation makes governed systems trustworthy.
4. Passing the governance gate unlocks the payoff: every model the institution uses afterward draws on its own validated knowledge instead of starting over.

Models come and go. The knowledge compounds. The workbench's maturity levels already climb this staircase; see the Crosswalk below.

## The Reference Matrix

![AI Fluency Progression: Reference Matrix](assets/reference-matrix.svg)

The matrix maps what learners practice, what faculty teach and assess, and what the institution provides, at every phase. It is the working document for course and program design.

**Read the status legend first.** Each persona row below carries a validation status. `Hypothesis` means the cells are research-informed but have not yet survived contact with NWC faculty use. Workbench templates carry short matrix checks (in the after-action note and calibration protocol) so routine use produces the evidence that confirms, revises, or strikes these cells. For why the matrix ships as a hypothesis rather than doctrine, see [the concept note](../concepts/why-the-matrix-is-a-hypothesis.md).

### Learners — practice it and produce with it

Status: Hypothesis — awaiting NWC validation.

| Phase | Learners |
| --- | --- |
| 1 Ask | Know when AI is answering vs acting. Ask, summarize, and draft. Verify claims and sources. Protect sensitive information. Check whether AI used files, tools, or external actions. |
| 2 Understand | Ask for explanations and worked examples. Quiz understanding and surface gaps. Track misconceptions. Restate ideas without the model. |
| 3 Produce | Produce bounded outputs: outlines, drafts, briefs, analyses, decision aids. State task, context, sources, and constraints. Define the standard for a usable output. |
| 4 Judge | Review outputs and intermediate work. Check sources and find weak claims. Compare alternatives and correct errors. Decide what to accept, revise, or reject. |
| 5 Codify | Turn recurring work into a reusable method: task brief, prompt pattern, checklist, source list, steps, review criteria, and stop rules. Keep examples of good and bad outputs. |
| 6 Supervise | Write task briefs, assign roles, set constraints. Inspect intermediate outputs and compare results. Define stop rules and escalate uncertainty. Integrate results into finished work. |

### Faculty — teach, coach, and assess it

Status: Hypothesis — awaiting NWC validation.

| Phase | Faculty |
| --- | --- |
| 1 Ask | Model safe use and safe delegation. Show the difference between an answer and an action. Require marking what was trusted, rejected, verified. |
| 2 Understand | Design AI support for understanding, not substitution. Compare AI explanations with course materials. Ask what changed in the student's thinking. |
| 3 Produce | Require process notes: what was delegated, revised, sourced. Assess the artifact and the choices behind it. Keep the student responsible for the result. |
| 4 Judge | Use critique logs, source audits, revision records. Run oral defense and checkpoint reviews. Grade judgment, not just polish. |
| 5 Codify | Have students document a workflow. Test it on more than one case, then peer review. Ask where human review belongs, and why. |
| 6 Supervise | Teach supervision before automation. Use small, bounded delegation exercises. Assess monitoring, integration, and judgment. |

### Institution — enables it, governs it, and compounds it

Status: Hypothesis — awaiting NWC validation.

| Phase | Institution |
| --- | --- |
| 1 Ask | Approved tools with role permissions. Data boundaries and safe-use examples. Guidance on when AI may inspect files or act. |
| 2 Understand | Trusted source packets. Course-level AI guidance. Sample learning prompts and faculty examples. |
| 3 Produce | Assignment expectations and disclosure norms. Rubrics and sample artifacts. Rules for acceptable AI-assisted production. |
| 4 Judge | Review standards and critique rubrics. Source-audit practices and approval gates. Norms for making AI-supported reasoning visible. |
| 5 Codify | Shared prompt and workflow libraries. Skill-card templates and approved knowledge sources. Examples of good and bad outputs; light evaluations. |
| 6 Supervise | Governed tools and bounded agent templates. Logs, permissions, escalation rules, review gates. Staff training, evaluation metrics, risk management. An AI-native knowledge layer feeding models, when warranted. |

## Validation Status And Changelog

Statuses: `Hypothesis — awaiting NWC validation` | `Field-tested — evidence from NWC use` | `Revised` | `Struck`.

Evidence arrives through the matrix checks in the after-action note template and the faculty calibration protocol. Changes to matrix cells are recorded here.

| Date | Row / cell | Change | Evidence |
| --- | --- | --- | --- |
| 2026-07-08 | All rows | Published at Hypothesis status. | Initial rendition from the Building AI Fluency package. |

## Crosswalk

One table answers "where am I, and what do I use?" Start with the [phase placement diagnostic](../templates/phase-placement-diagnostic.md) if you do not know your phase.

### Templates To Phases

| Template | Primary phases | Matrix row it exercises |
| --- | --- | --- |
| [Phase placement diagnostic](../templates/phase-placement-diagnostic.md) | Entry point, all phases | Faculty |
| [Assignment design worksheet](../templates/assignment-design-worksheet.md) | 2–4 | Faculty |
| [Assessment and oral-defense rubric](../templates/assessment-and-oral-defense-rubric.md) | 4 | Faculty |
| [Flawed output library template](../templates/flawed-output-library-template.md) | 4 | Faculty, Institution |
| [Faculty calibration protocol](../templates/faculty-calibration-protocol.md) | 4–5 | Faculty, Institution |
| [Method card template](../templates/method-card-template.md) | 5 | Learners, Faculty |
| [Supervised delegation exercise](../templates/supervised-delegation-exercise.md) | 6 | Learners, Faculty |
| [Source kit template](../templates/source-kit-template.md) | 3, 5–6 | Institution |
| [After-action note template](../templates/after-action-note-template.md) | 5 | Faculty, Institution |

### Maturity Levels To Phases And Staircase

| Workbench level | Phase(s) | Staircase step |
| --- | --- | --- |
| 1 Faculty Fluency Lab | Faculty practice phases 1–4 themselves | Reviewed practice |
| 2 Assignment Design | 2–3 | Reviewed practice |
| 3 Assessment Design | 4 | Reviewed practice |
| 4 Flawed Output Library | 4 | Shared assets |
| 5 Faculty Calibration | 4–5 | Evaluation |
| 6 Source Kits | 5–6 | Shared assets |
| 7 Institutional Memory (future) | 5, Institution row | Governed systems |
| 8 Context Curation (future) | 6, Institution row | Beyond the governance gate |

The workbench and the framework arrived at the same institutional shape independently: reviewed practice becomes shared assets, assets make evaluation possible, evaluation earns governed reuse. The maturity levels are the staircase, enacted.


# ===== SECTION: CONCEPTS =====

# Workbench Concepts

**The concepts layer explains why each workbench artifact has the fields it does — bridging unfamiliar tools to things you already understand.**

The templates tell you *what to do*. The framework doc tells you *what the phases are*. These notes tell you *why*. Each takes two minutes and stands alone.

## Start From Your Question

- Why does a method card have a revision log and stop rules? → [Method cards and agent skills](method-cards-and-agent-skills.md)
- Why is the assistant so scripted when I hand it a template? → [Facilitation blocks](facilitation-blocks.md)
- Why does the matrix say "hypothesis — awaiting validation"? → [Why the matrix is a hypothesis](why-the-matrix-is-a-hypothesis.md)
- Why isn't a source kit just a folder of readings? → [Source kits are curated context](source-kits-are-curated-context.md)
- Why write an after-action note when the exercise went fine? → [How faculty judgment compounds](how-faculty-judgment-compounds.md)

## The Vocabulary Bridge

Each row links a concept note to the template it explains and the industry idea it already resembles.

| Concept note | The template it serves | Industry equivalent |
| --- | --- | --- |
| [Method cards and agent skills](method-cards-and-agent-skills.md) | [Method card](../templates/method-card-template.md) | Agent skill / SKILL.md |
| [Facilitation blocks](facilitation-blocks.md) | [All templates](../templates/) | System prompt / agent instructions |
| [Why the matrix is a hypothesis](why-the-matrix-is-a-hypothesis.md) | [Framework doc](../framework/ai-fluency-progression.md) | Evals; claims gated on results |
| [Source kits are curated context](source-kits-are-curated-context.md) | [Source kit](../templates/source-kit-template.md) | Curated context / RAG corpus |
| [How faculty judgment compounds](how-faculty-judgment-compounds.md) | [After-action note](../templates/after-action-note-template.md) · [Calibration protocol](../templates/faculty-calibration-protocol.md) | Eval sets; inter-rater reliability |

## How To Read These

Every note has the same shape: a one-line summary, what the artifact is, one moment you would use it, the industry idea it maps to, and what the workbench adds and why. Skim the headings first. Read the body only if the summary earns your next two minutes.

## Backlog

Two notes are deliberately unwritten until faculty ask for them:

- **A tool-chooser note** — "what remains when the work is done," on picking the right artifact for a task.
- **A traces / oral-defense note** — "why not prompt logs," on evidence of ownership over compliance logging.

New concept notes get written when pilot questions ask for them, not before.

# Facilitation Blocks

**A facilitation block turns a worksheet into a script an AI assistant can run — the workbench's system prompt, written into the document itself.**

## What It Is

Every workbench template carries an `## AI Facilitation Block` — see the [phase placement diagnostic](../templates/phase-placement-diagnostic.md) for the pattern. It tells an assistant how to run that template as a guided session: its role, what to collect first, how to walk the sections, what never to do, and how to finish.

## Where You'll Use It

A faculty member pastes the [source kit template](../templates/source-kit-template.md) into an assistant and says "help me package a source kit." The facilitation block makes the assistant pressure-test her boundaries instead of dumping a generic checklist. She never reads the block; she just gets a better session because it is there.

## The Industry Equivalent

This is a system prompt — the agent instructions that shape how an assistant behaves for a task. AI-fluent readers already write these to steer tone, scope, and guardrails.

## What The Workbench Adds

The block lives *inside* the worksheet, not in a separate config. That makes every template dual-reader: a human fills it out on paper, or hands the whole file to an assistant and the assistant is the runtime. Three design choices:

- **Collect-first.** The assistant gathers the faculty member's own material before producing anything — judgment stays with the human.
- **Never-list.** Each block names what the assistant must not do: invent steps, smooth over disagreement, turn work into surveillance.
- **Bounded finish.** The session ends with a clean artifact and a handoff, not an open-ended chat.

The block also stays visible on paper. Faculty can read exactly what the assistant was told to do — the instructions are never hidden from the person being facilitated.

> **Framework tie:** Facilitation blocks are supervised, AI-mediated work in miniature (Phase 6). The assistant is the runtime; the markdown is the program; the faculty member keeps the judgment seat.

# How Faculty Judgment Compounds

**Reviewed practice becomes shared assets becomes evaluation — the same staircase that turns individual notes into inter-rater reliability.**

## What It Is

Two templates capture judgment so it accumulates instead of evaporating: the [after-action note](../templates/after-action-note-template.md), which preserves lesson rationale after an exercise, and the [calibration protocol](../templates/faculty-calibration-protocol.md), which compares how faculty diagnose the same AI-assisted work.

## Where You'll Use It

Three instructors grade the same flawed AI estimate and diverge on whether the reliance was justified. The calibration protocol captures where they agree, where they do not, and the question that exposes the difference — ready for the next rotation instead of lost to the hallway.

## The Industry Equivalent

AI teams build eval sets and measure inter-rater reliability — do independent reviewers agree on what counts as good? Faculty already do this work; calibration just makes it explicit and reusable.

## What The Workbench Adds

Individual judgment is perishable. The workbench compounds it in three moves:

- **Reviewed practice.** An after-action note turns one exercise into a durable record of what worked and why.
- **Shared assets.** Calibration turns private standards into a shared minimum — recorded in the faculty's own words, disagreement preserved rather than smoothed away.
- **Evaluation.** The matrix checks inside both templates feed evidence back to the framework, so the whole system learns.

Legitimate disagreement is an outcome, not a failure. The protocol records the range instead of forcing consensus.

> **Framework tie:** This is the staircase — reviewed practice → shared assets → evaluation. It is how a faculty's judgment outlives any one instructor.

# Method Cards And Agent Skills

**A method card captures a working AI-enabled task completely enough that a skill file — the industry's format for reusable AI procedures — can be created straight from it.**

## What It Is

A [method card](../templates/method-card-template.md) captures a recurring AI-enabled task once you have run it well more than once: the task brief, the steps, the review criteria, and the stop rules. It is the Phase 5 artifact — where scale begins, because people stop re-explaining the task.

## Where You'll Use It

A faculty member has walked three cohorts through the same AI-assisted intelligence-estimate critique. The steps are in her head. She fills out a method card so the next instructor runs it the same way — and so the review criteria she learned the hard way do not leave with her.

## The Industry Equivalent

Software teams package repeatable AI tasks as **skills**. A skill is a folder built around one markdown file: a short description that tells the assistant when to act, step-by-step instructions, and examples. The assistant finds it and follows it automatically when a matching task appears.

## What The Workbench Adds

A completed method card contains everything a skill file needs — plus what the skill format has no field for:

- **The instructions travel.** The card's steps, review criteria, and examples map straight into a skill file. Hand a finished card to an assistant and ask it to package one; that is the whole conversion.
- **The accountability stays.** Owner, times run, human review points, and the revision log have no home in a skill file. They live on the card because they are for the faculty managing the method, not the machine running it.
- **The evidence gate comes first.** Codify only what has worked twice. No skill gets created from a method that has not earned it.

> **Framework tie:** Method cards are Phase 5 (Codify). The skill file is the deployment; the card is the source — and the record of judgment that outlasts any one tool.

# Source Kits Are Curated Context

**A source kit is a curated context packet with boundaries — the teaching version of the corpus an AI-fluent team assembles before letting a model work.**

## What It Is

A [source kit](../templates/source-kit-template.md) is the curated context for an AI-enabled exercise. It tells an assistant what materials matter, what standards apply, what outputs faculty will inspect, and what boundaries must hold. A source kit is not a file dump.

## Where You'll Use It

A faculty member building a wargame-analysis exercise assembles the readings, the assessment rubric, and a note that one restricted case is excluded. The kit lets any assistant run the exercise without ever seeing what it should not.

## The Industry Equivalent

This is curated context — the material a team assembles so a model answers from the right sources instead of guessing. At scale it becomes a retrieval corpus (the "RAG" an AI team maintains); in a single session, it is the files you attach to the chat. Either way, deciding what goes in is the real work.

## What The Workbench Adds

A source kit is curated context with the governance a teaching setting needs:

- **Explicit boundaries.** Every source is marked allowed, excluded, or restricted. Private course material does not leak into a public kit by accident.
- **Standards and inspection points.** The kit names what "good" looks like and what faculty will actually check — not just what to feed the model.
- **Faculty ownership of the cut.** The faculty member decides what is in and what is out. The assistant organizes and pressure-tests; it does not curate for them.

> **Framework tie:** Source kits are Phase 5–6 (Codify → Supervise) — and the seed of the Level 6→8 lineage: today's curated packet is what a future faculty-governed context vault would grow from.

# Why The Matrix Is A Hypothesis

**The reference matrix ships as a set of claims awaiting evidence, not as doctrine — the workbench modeling the reliance calibration it teaches.**

## What It Is

The [framework doc](../framework/ai-fluency-progression.md) ends in a reference matrix: what learners practice, what faculty teach, and what the institution provides, at every phase. Each persona row carries a status line, and today that status is `Hypothesis — awaiting NWC validation`.

## Where You'll Use It

A faculty member reads a matrix cell that does not match her classroom. Instead of dismissing the framework, she notes the mismatch in her after-action matrix check. Her disagreement becomes data — exactly what the status line invites.

## The Industry Equivalent

AI teams call them evals: test sets that measure a capability against success criteria before anyone relies on it. A claim gated on evals stays a hypothesis until the results back it. The matrix status line is that same discipline applied to a teaching framework.

## What The Workbench Adds

The matrix could have been printed as settled doctrine. It is not, for two reasons:

- **Honesty about evidence.** The cells are research-informed but have not survived contact with NWC faculty use. Marking them as hypotheses says so plainly.
- **A path to evidence.** The [after-action note](../templates/after-action-note-template.md) and [calibration protocol](../templates/faculty-calibration-protocol.md) carry short matrix checks. Routine use produces the evidence that confirms, revises, or strikes each cell.

As evidence arrives, a row's status moves — to `Field-tested — evidence from NWC use`, `Revised`, or `Struck` — and the change is recorded in the framework doc's changelog. The workbench asks students to calibrate how far they rely on AI. The matrix holds itself to the same standard.

> **Framework tie:** This is reliance calibration turned on the framework itself — trust the matrix exactly as far as the evidence goes, and no further.


# ===== SECTION: PHASE PLACEMENT DIAGNOSTIC =====

# Phase Placement Diagnostic

Use this diagnostic to find where an assignment, exercise, or course sits on the AI fluency progression, then pick the right workbench template. It takes about ten minutes with an AI assistant, or on paper.

The six phases are described in [the AI fluency progression](../framework/ai-fluency-progression.md): 1 Ask, 2 Understand, 3 Produce, 4 Judge, 5 Codify, 6 Supervise.

## AI Facilitation Block

If you are working on paper, skip this section. If you are using an AI assistant, give it this entire file and say: "Run this diagnostic with me."

Instructions for the AI assistant:

- Role: You are running a placement interview for a faculty member. The faculty member owns every judgment about their course. You ask, listen, and place. You do not redesign their assignment.
- Collect first: the course or seminar, the specific assignment or exercise, and what role AI currently plays in it (including "none").
- Process: Ask the placement questions below one at a time, in order. Stop early once the placement logic gives a clear answer. Push back once if an answer is vague, then accept the faculty member's call.
- Never: recommend tools or products; invent NWC policy or doctrine; treat a higher phase as better teaching — the right phase is the one that fits the task and the students; continue past an unresolved answer without flagging it.
- Finish: State the phase placement in one sentence, explain the routing in two or three sentences using the routing table, and return a short markdown note the faculty member can keep: assignment, placement, reasoning, recommended templates.

## Placement Questions

1. In this task, is AI answering questions, or doing work? (Answering only, or no AI yet → likely phase 1–2. Doing bounded work → phase 3 or higher.)
2. What must students own before AI enters — the frame, the purpose, the evidence standard? (If this is undefined, start at phase 2–3 design regardless of ambition.)
3. Is reviewing, verifying, or critiquing AI output an assessed part of the task? (Yes → phase 4 is in play.)
4. Does this task recur — across weeks, sections, or courses — often enough that the method could be written down once and reused? (Yes → phase 5.)
5. Would you trust students to direct a multi-step AI workflow with checkpoints you can inspect? (Yes, and phases 1–5 are in place → phase 6. If earlier phases are missing, place at the earliest missing phase instead.)
6. Who is being placed — the assignment, the students, or you? (This diagnostic places the assignment. Faculty can run it on their own practice too; the logic is the same.)

## Placement Logic

| If... | Placement |
| --- | --- |
| No deliberate AI role yet, or safety and verification habits are not established | 1 Ask |
| AI supports understanding, but production with AI is not assessed | 2 Understand |
| AI does bounded production work; students own task, context, and constraints | 3 Produce |
| Students must judge, verify, and revise AI output as assessed work | 4 Judge |
| The task recurs and the method is worth writing down once | 5 Codify |
| Students direct a bounded multi-step AI workflow under inspection | 6 Supervise |

Place at the **earliest phase that is not yet solid**. A phase 6 ambition with phase 1 habits is a phase 1 placement.

## Routing

| Placement | Use these templates |
| --- | --- |
| 1–2 | [Assignment design worksheet](assignment-design-worksheet.md) — decide where AI belongs and what stays AI-free. |
| 3 | [Assignment design worksheet](assignment-design-worksheet.md) + [source kit template](source-kit-template.md). |
| 4 | [Assessment and oral-defense rubric](assessment-and-oral-defense-rubric.md) + [flawed output library template](flawed-output-library-template.md). |
| 5 | [Method card template](method-card-template.md) + [faculty calibration protocol](faculty-calibration-protocol.md). |
| 6 | [Supervised delegation exercise](supervised-delegation-exercise.md). |
| After any run | [After-action note template](after-action-note-template.md). |

## Paper Worksheet

- Course or seminar:
- Assignment or exercise:
- Current AI role:
- Answers to questions 1–5:
- Placement:
- Reasoning:
- Templates to use next:


# ===== SECTION: ASSIGNMENT DESIGN WORKSHEET =====

# Assignment Design Worksheet

Use this worksheet when designing or revising an assignment for AI-enabled strategic judgment. The goal is to decide where AI helps, where it harms, and what faculty need to observe.

## AI Facilitation Block

If you are working on paper, skip this section. If you are using an AI assistant, give it this entire file and say: "Facilitate this worksheet with me."

Instructions for the AI assistant:

- Role: You are facilitating an assignment-design session for a faculty member. The faculty member owns every pedagogical judgment. You ask, structure, and challenge. You never decide where AI belongs in their assignment.
- Collect first: the course or seminar, the learning objective, and the current assignment if one exists.
- Process: Walk the numbered sections in order, one question at a time. In section 3, actively defend developmental friction — if the faculty member proposes AI help there, ask what judgment the struggle was building. In section 5, make them commit to a sequence before moving on.
- Never: write the assignment yourself; fill in a field the faculty member has not decided; soften developmental friction to make the design easier; invent doctrine, policy, or sources; continue past an unresolved judgment call without flagging it.
- Finish: Return the completed worksheet as clean markdown, listing any fields the faculty member deferred.

## 1. Learning Purpose

What is this assignment for?

- Course or seminar:
- Learning objective:
- Strategic judgment students should practice:
- Why this task matters for future AI-enabled leadership:

## 2. Problem Frame Students Must Own

What must students define before AI enters?

- Strategic problem:
- AI-shaped inputs students inherit before direct AI use:
- Purpose of the work:
- Key actors:
- Assumptions:
- Evidence standard:
- Success standard:
- Risks or tradeoffs:

## 3. Developmental Friction

What struggle should be preserved because it builds judgment?

- Work students should do without AI:
- First-frame activity:
- Ambiguity or uncertainty students should face:
- Seminar challenge or peer critique:
- What failure should teach:

## 4. Wasteful Friction

What work can AI reduce without weakening judgment?

- Formatting or synthesis work:
- Search or retrieval work:
- Alternative framing:
- Counterargument generation:
- Red-team questions:
- Other low-value friction:

## 5. AI-Free And AI-Mediated Sequence

Design the sequence deliberately.

| Phase | Student action | AI role | Faculty observation |
| --- | --- | --- | --- |
| AI-free first frame |  | None |  |
| AI-mediated challenge |  | Challenge, critique, expand, or compare |  |
| Human revision |  | Optional support |  |
| Oral defense or seminar challenge |  | None or limited |  |
| Trace artifact |  | Formatting support only, if allowed |  |

## 6. Reliance Decisions Students Must Make

Where should students decide whether to rely, verify, redirect, or refuse?

- AI output they may accept:
- AI output they must verify:
- AI output they should reject or challenge:
- Part of the task where AI should be withheld:
- Evidence required before reliance is justified:

## 7. Assessment Evidence

What will faculty inspect?

- Purpose through frame:
- Inherited AI-shaped inputs:
- Assumptions:
- Evidence standard:
- Accepted AI contributions:
- Rejected or revised AI contributions:
- Reliance decision:
- Final human judgment:
- Transfer check:

## 8. No-Garden-Path Check

Does the assignment avoid making the easy AI answer the wrong lesson?

- Could a student complete the task by polishing AI output?
- Does the assignment require a choice among plausible frames?
- Does it include a flawed or incomplete AI output to critique?
- Does oral defense reveal ownership?
- Does the trace stay lean enough to use?

## 9. Reusable Artifact

What should be saved for next time?

- Prompt:
- Flawed output:
- Rubric:
- Oral-defense questions:
- Faculty notes:
- After-action note:


# ===== SECTION: ASSESSMENT AND ORAL-DEFENSE RUBRIC =====

# Assessment And Oral-Defense Rubric

Use this rubric when the assignment goal is to make ownership visible in AI-enabled strategic work. The finished artifact still matters, but it does not carry the whole assessment burden.

## AI Facilitation Block

If you are working on paper, skip this section. If you are using an AI assistant, give it this entire file and say: "Help me prepare an assessment with this rubric."

Instructions for the AI assistant:

- Role: You are helping a faculty member prepare to assess AI-enabled student work and rehearse an oral defense. The faculty member owns every rating and every judgment about the student. You structure, probe, and rehearse.
- Collect first: the assignment being assessed and what evidence of ownership the faculty member already has.
- Process: Walk the dimensions one at a time and ask what evidence would distinguish a 2 from a 3 on each. Then rehearse: play the student in an oral defense using the question list, and afterward tell the faculty member which questions exposed the most.
- Never: rate a real student's work yourself; suggest that disclosure of AI use alone equals ownership; add rubric dimensions without being asked; treat polish as evidence of judgment.
- Finish: Return the faculty member's prepared rubric notes and the oral-defense question order they chose, as clean markdown.

## Rating Scale

| Rating | Meaning |
| --- | --- |
| 1 - Thin | Student relies on surface language or retrospective explanation. Ownership is unclear. |
| 2 - Emerging | Student can explain some choices but struggles under follow-up or changed conditions. |
| 3 - Proficient | Student owns the purpose, frame, reliance decisions, and judgment under questioning. |
| 4 - Strong | Student shows disciplined AI use, revises intelligently, and transfers the method to a changed case. |

## Dimensions

| Dimension | What Faculty Look For | Rating | Notes |
| --- | --- | --- | --- |
| Purpose through frame | Student can explain what the work is for, why this problem frame was chosen, and what would change it. |  |  |
| Inherited AI-shaped inputs | Student can identify reports, summaries, planning tools, staff processes, or prior analytic products that may have already shaped the work. |  |  |
| Assumptions | Student identifies explicit, inherited, AI-suggested, and revised assumptions. |  |  |
| Evidence standard | Student explains what evidence would strengthen, weaken, or change the judgment. |  |  |
| Reliance | Student can say where AI helped, where it was verified, where it was rejected, and why. |  |  |
| Accountability | Student states the final judgment in first person and accepts responsibility for uncertainty and risk. |  |  |
| Transfer | Student applies the same discipline to a changed case, new artifact, or altered premise. |  |  |
| Developmental friction | Student did enough unaided or contested work to build judgment rather than merely polish output. |  |  |

## Oral-Defense Question Bank

Ask one question at a time. Follow the student rather than reading the list mechanically.

### Purpose And Frame

- What problem did you decide this work was actually solving?
- What did your frame include that another reasonable frame might exclude?
- What would make you change the frame?
- What inputs had already sorted, summarized, or framed the problem before you touched AI directly?
- Where might AI have narrowed the frame before you noticed?

### Assumptions

- Which assumption is doing the most work?
- Which assumption came from the assignment, the source material, AI, or your own judgment?
- What assumption did you reject or revise?

### Evidence Standard

- What evidence would change your conclusion?
- Which claim is least secure?
- What kind of evidence would AI be least reliable at judging here?

### Reliance

- What did AI make easier?
- Where did you rely on AI, and what justified that reliance?
- What did you verify independently?
- What did you reject, withhold from AI, or rewrite?

### Accountability

- State the final judgment in first person.
- What risk remains if your judgment is wrong?
- What would you say to a commander or policymaker who acted on this recommendation?

### Transfer

- How would your judgment change if the adversary, time horizon, authority, or available evidence changed?
- What part of your method would transfer to a different case?
- What would you save so another student or instructor could reuse this work?

## Minimal Faculty Note

If time is short, record only:

1. Evidence of purpose through frame.
2. Inherited AI-shaped input worth probing.
3. Reliance decision worth probing.
4. Accountability question asked.
5. Transfer result.
6. Follow-up needed.


# ===== SECTION: FLAWED OUTPUT LIBRARY TEMPLATE =====

# Flawed Output Library Template

Use this template to build a reusable library of polished but strategically flawed AI outputs. The flaw should survive a surface-level reading and fail under strategic questioning.

## AI Facilitation Block

If you are working on paper, skip this section. If you are using an AI assistant, give it this entire file and say: "Help me build a flawed output."

Instructions for the AI assistant:

- Role: You are helping a faculty member create a polished but strategically flawed AI output for teaching. The faculty member chooses the flaw and owns the instructor key. You draft the polish; they design the trap.
- Collect first: the course, the case or topic, and which flaw type from the list the faculty member wants students to find.
- Process: Have the faculty member specify the flaw and the stronger frame first, then draft the student-facing artifact so the flaw survives a surface reading. Then complete the instructor key together.
- Never: choose the flaw type yourself; make the flaw a factual error a spell-check mindset would catch — the point is strategic, not clerical; write the oral-defense questions without the faculty member's approval.
- Finish: Return the complete library entry as clean markdown: metadata, student-facing artifact, instructor key, and oral-defense questions.

## Entry Metadata

- Title:
- Course or seminar:
- Case or topic:
- Date created:
- Created by:
- Intended use:
- Public, internal, or restricted:

## Flaw Type

Select all that apply.

- Frame error
- Hidden assumption
- Weak evidence standard
- Uncalibrated reliance
- Risk or tradeoff buried
- Accountability evasion
- Transfer failure
- Institutional monoculture or narrowed perspective

## Student-Facing Artifact

Paste or link the flawed AI output students will inspect.

```text
[Student-facing flawed output goes here.]
```

## Instructor Key

### Hidden Frame

What problem frame does the output assume?

### Flawed Assumptions

Which assumptions are explicit, implied, inherited, or AI-suggested?

### Missing Evidence

What evidence would be required before relying on this assessment?

### Risk Or Tradeoff

What cost, risk, uncertainty, or second-order effect is buried?

### Accountability Problem

Where does the output make judgment appear authorless, automatic, or consequence-free?

### Stronger Frame

What would a stronger frame include?

## Oral-Defense Questions

- Question 1:
- Question 2:
- Question 3:
- Question 4:
- Question 5:

## Expected Student Trace

Students should record:

- hidden frame identified;
- assumptions revised;
- evidence standard;
- accepted, rejected, or revised AI outputs;
- final human judgment;
- transfer check.

## Faculty Notes

- What students noticed quickly:
- What students missed:
- What the output revealed about reliance:
- How faculty revised the exercise:

## Retirement Criteria

Retire or revise this entry when:

- students spot the flaw too easily;
- the example becomes stale;
- the flaw depends on a factual error rather than judgment;
- the output no longer reflects current AI capability;
- faculty disagreement shows the key is unclear.


# ===== SECTION: SOURCE KIT TEMPLATE =====

# Source Kit Template

A source kit is the curated teaching packet for an AI-enabled exercise. It tells an AI assistant what materials matter, what standards apply, what outputs faculty will inspect, and what boundaries must be respected.

A source kit is not a file dump.

Concept: [why this template works the way it does](../concepts/source-kits-are-curated-context.md)

## AI Facilitation Block

If you are working on paper, skip this section. If you are using an AI assistant, give it this entire file and say: "Help me package a source kit."

Instructions for the AI assistant:

- Role: You are helping a faculty member curate the context packet for an AI-enabled exercise. The faculty member decides what is in, what is out, and where the boundaries sit. You organize and pressure-test.
- Collect first: the exercise, the anchor materials that exist, and the public/internal/restricted status of each.
- Process: Walk the sections in order. Pressure-test section 4 hardest: for each source, ask whether it is allowed, excluded, or missing. Flag anything that looks like private course material heading into a public kit.
- Never: add sources the faculty member has not named; write the AI-role boundaries yourself; treat a file dump as a kit — if the kit lacks standards and boundaries, say so.
- Finish: Return the completed source kit as clean markdown with an explicit public-safety note on anything borderline.

## 1. Overview

- Source kit title:
- Course or seminar:
- Faculty owner:
- Date:
- Public, internal, or restricted:
- Intended exercise:

## 2. Learning Purpose

- Course objective:
- Strategic judgment students should practice:
- Why AI belongs in this exercise:
- What students must own:

## 3. Anchor Materials

List only the materials the AI assistant and students should use.

- Essay, prompt, or assignment:
- Case materials:
- AI-shaped inputs already present in the materials:
- Doctrine or primer materials:
- Public sources:
- Course-specific sources:

## 4. Allowed And Excluded Sources

### Allowed

- Source 1:
- Source 2:
- Source 3:

### Excluded

- Source or category:
- Reason:

## 5. NWC Vocabulary And Standards

Terms or concepts the AI assistant should preserve:

- ends:
- ways:
- means:
- assumptions:
- risk:
- costs:
- reassessment:
- other course-specific terms:

## 6. AI Role

What AI may do:

- retrieve;
- summarize;
- challenge;
- generate alternatives;
- create flawed output;
- ask oral-defense questions;
- help format a trace.

What AI may not do:

- choose the final purpose;
- own the problem frame;
- replace independent first-frame work;
- make the final judgment;
- convert private material into public output.

## 7. Assessment Materials

- Rubric:
- Oral-defense questions:
- Trace artifact:
- Flawed output:
- Instructor key:

## 8. Faculty Review Notes

- What faculty should observe:
- Common failure modes:
- Reliance concern:
- Accountability concern:
- Transfer concern:

## 9. Public / Private Boundary

- What can be shared publicly:
- What must remain internal:
- What should not be given to public AI tools:
- Who approves changes to this kit:

## 10. Proposed Updates

Agents or faculty may propose updates. They do not apply automatically.

- Proposed change:
- Reason:
- Evidence:
- Faculty decision: approve, revise, reject, archive.


# ===== SECTION: FACULTY CALIBRATION PROTOCOL =====

# Faculty Calibration Protocol

Use this protocol when faculty need to compare how they diagnose the same AI-assisted work. The goal is to make tacit judgment explicit without forcing false agreement.

Concept: [why this template works the way it does](../concepts/how-faculty-judgment-compounds.md)

## AI Facilitation Block

If you are working on paper, skip this section. If you are using an AI assistant, give it this entire file and say: "Help me run a calibration session."

Instructions for the AI assistant:

- Role: You are supporting a faculty calibration session. The faculty are the judges; you are the scribe and the timekeeper. You surface disagreement; you never resolve it.
- Collect first: the artifact under review and how many faculty are participating.
- Process: Keep individual reviews independent — do not share one reviewer's diagnosis with another before step 2. In step 2, present convergence and divergence neutrally. In step 3, record the shared minimum standard in the faculty's own words.
- Never: score the artifact yourself; smooth over a disagreement to reach consensus; suggest that divergent faculty judgment is a problem to eliminate — the protocol says legitimate range is an outcome.
- Finish: Return the completed calibration note as clean markdown.

## Purpose

Faculty already bring much of the judgment needed to spot hidden assumptions, thin reasoning, performed sophistication, and weak strategic judgment. Calibration joins that tacit judgment to AI fluency: faculty compare how they read the same AI-assisted work, where they think reliance was justified, and what questions expose whether the human still owns the frame.

## Materials

- One student trace, flawed AI output, or AI-assisted strategic product.
- Current rubric or review criteria.
- Individual diagnosis form.
- Shared calibration note.

## Protocol

### 1. Individual Review

Each faculty member reviews the same artifact independently.

Record:

- strongest part of the work;
- weakest part of the work;
- hidden frame;
- key assumption;
- reliance concern;
- accountability concern;
- transfer concern;
- one oral-defense question.

### 2. Compare Diagnoses

Bring faculty together and compare:

- where judgments converged;
- where judgments diverged;
- which concern mattered most;
- which rubric language caused ambiguity;
- which oral-defense question would reveal the issue fastest.

### 3. Resolve What Needs Resolving

Faculty do not need to agree on every interpretation. They do need to agree on what students must be able to defend.

Record:

- shared minimum standard;
- legitimate range of faculty judgment;
- unresolved disagreement;
- implication for assignment design.

### 4. Revise Shared Artifacts

Update one or more:

- rubric;
- oral-defense question set;
- flawed-output instructor key;
- trace artifact fields;
- source-kit instructions;
- after-action note.

### 5. Save The Calibration Note

The note should be short enough to reuse.

## Calibration Note Template

- Artifact reviewed:
- Faculty participants:
- Date:
- Agreement:
- Disagreement:
- Rubric language to revise:
- Oral-defense question to keep:
- Failure mode to watch:
- Source-kit or assignment update proposed:
- Decision: approve, revise, archive, or run again.

## Matrix Check

Three lines for the fluency progression. See [the reference matrix](../framework/ai-fluency-progression.md). All fields optional.

- Phase the reviewed work operated at (1 Ask / 2 Understand / 3 Produce / 4 Judge / 5 Codify / 6 Supervise):
- Did faculty expectations at this phase match the matrix's faculty row? Where not:
- One change you would make to that matrix row:

## Review Question

Did this calibration make future faculty judgment easier to explain, teach, and reuse?


# ===== SECTION: AFTER-ACTION NOTE TEMPLATE =====

# After-Action Note Template

Use this note after running an AI-enabled exercise. The goal is to preserve lesson rationale, faculty judgment, and useful revisions before they disappear.

Concept: [why this template works the way it does](../concepts/how-faculty-judgment-compounds.md)

## AI Facilitation Block

If you are working on paper, skip this section. If you are using an AI assistant, give it this entire file and say: "Debrief this exercise with me."

Instructions for the AI assistant:

- Role: You are debriefing a faculty member after an AI-enabled exercise, while memory is fresh. Their observations are the content; you draw them out and structure them.
- Collect first: which exercise, when it ran, and the single strongest and weakest moment.
- Process: Walk the sections in order, but follow energy — if the faculty member wants to start with what failed, start there and backfill. Push for specifics: one named frame error beats three generalities. End with the matrix check.
- Never: soften a failure into a lesson before the faculty member has described it plainly; propose updates to shared artifacts as decided — the proposals table requires faculty approval; include student names or private course material in the note.
- Finish: Return the completed note as clean markdown and remind the faculty member where to send it.

## Exercise Information

- Exercise:
- Course or seminar:
- Date:
- Faculty:
- Student group:
- Source kit used:

## What The Exercise Was For

- Learning purpose:
- Judgment students were supposed to practice:
- AI role:
- Assessment evidence faculty expected:

## What Worked

- Strongest student performance:
- Strongest faculty observation:
- Useful AI contribution:
- Useful friction preserved:
- Reusable artifact created:

## What Failed Or Confused Students

- Common frame error:
- Assumption students missed:
- Reliance problem:
- Accountability problem:
- Transfer problem:
- Confusing instructions:

## Faculty Calibration Notes

- Where faculty agreed:
- Where faculty disagreed:
- Rubric language to revise:
- Oral-defense question to keep:
- Trace field to revise:

## Proposed Updates

Proposals require faculty approval before they become shared context.

| Proposed update | Reason | Approve / revise / reject | Owner |
| --- | --- | --- | --- |
|  |  |  |  |

## Archive Or Retire

- Artifact to keep:
- Artifact to revise:
- Artifact to retire:
- Reason:

## Next Run

- Change before next use:
- Source-kit update:
- New flawed output needed:
- Faculty calibration needed:

## Matrix Check

Three lines for the fluency progression. See [the reference matrix](../framework/ai-fluency-progression.md). All fields optional.

- Phase this exercise operated at (1 Ask / 2 Understand / 3 Produce / 4 Judge / 5 Codify / 6 Supervise):
- Did the matrix's expectations for learners and faculty at this phase match what happened? What did not:
- One change you would make to that matrix row:

## Sending This Note

Completed notes make the workbench better. Send a copy (with private course material removed) to the workbench maintainer: jackcshaw@gmail.com.


# ===== SECTION: METHOD CARD TEMPLATE =====

# Method Card Template

Use this template to turn a recurring AI-enabled task into a reusable method. A method card is what remains when the work is done well more than once: the task brief, the steps, the review criteria, and the stop rules. This is phase 5 of the fluency progression — where scale begins, because people stop re-explaining the task.

Codify only what has worked at least twice. A method card for a task you have done once is a guess wearing a uniform.

Concept: [why this template works the way it does](../concepts/method-cards-and-agent-skills.md)

## AI Facilitation Block

If you are working on paper, skip this section. If you are using an AI assistant, give it this entire file and say: "Help me write a method card."

Instructions for the AI assistant:

- Role: You are helping a faculty member (or student, if assigned) codify a recurring task they already know how to do. Their tacit knowledge is the content; you extract and structure it. You are not designing a new workflow.
- Collect first: the recurring task, how many times they have done it with AI support, and what went wrong at least once.
- Process: If the task has run fewer than two times, say so and stop — recommend running it again first. Otherwise walk the sections in order. Push hardest on review criteria and stop rules; "looks good" is not a criterion.
- Never: invent steps the person has not actually used; write review criteria for a domain you are guessing at; skip the bad example — the failure case is what makes the card teachable.
- Finish: Return the completed method card as clean markdown and ask where it should live so someone else can find it.

## Card Metadata

- Method name:
- Owner:
- Date:
- Course, seminar, or task family:
- Public, internal, or restricted:
- Times this method has been run:

## Task Brief

- Goal of the task:
- Who the output is for:
- Output format and length:
- What the human must supply each run (context, sources, constraints):

## When To Use — And Not

- Use when:
- Do not use when:
- Simpler alternative that sometimes suffices (checklist, template, no AI):

## Steps

Number each step. Mark the AI role in each: none, draft, challenge, compare, format.

1.
2.
3.

## Review Criteria

What makes the output acceptable? Be concrete enough that a colleague could apply these without you.

- Accuracy check:
- Source or evidence standard:
- Format standard:
- Rejection triggers:

## Stop Rules

When must the method halt and hand back to human judgment?

- Stop when:
- Escalate to whom:

## Examples

- One good output (paste or link, with one line on why it passed):
- One bad output (paste or link, with one line on why it failed):

## Human Review Points

- Where a human must review before the output is used:
- Who owns the result:

## Revision Log

| Date | Change | Reason |
| --- | --- | --- |
|  |  |  |


# ===== SECTION: SUPERVISED DELEGATION EXERCISE =====

# Supervised Delegation Exercise

Use this template to design a bounded exercise where students direct a multi-step AI workflow and faculty assess whether judgment survives delegation. This is phase 6 of the fluency progression. Teach supervision before automation: students should have practiced phases 1–5 on this kind of task first.

The exercise is not "build an agent." It is "supervise delegated work you remain accountable for."

## AI Facilitation Block

If you are working on paper, skip this section. If you are using an AI assistant, give it this entire file and say: "Help me design a supervised delegation exercise."

Instructions for the AI assistant:

- Role: You are helping a faculty member design a phase 6 exercise. The faculty member owns the pedagogy, the task choice, and the assessment standard. You structure and stress-test the design.
- Collect first: the course, the strategic task being delegated, and what evidence exists that students have phase 4–5 habits on this task.
- Process: Walk the sections in order. Stress-test the boundedness: if the delegated task cannot fail safely inside one session, push for a smaller task. Make the faculty member define the escalation rule before the inspection points.
- Never: design the exercise around a specific vendor or product; let "students supervise AI" become "students watch AI"; write assessment criteria that reward output volume over judgment; imply operational or classified use.
- Finish: Return the completed exercise design as clean markdown, with an explicit list of what could go wrong in the first run.

## Learning Purpose

- Course or seminar:
- Judgment this exercise develops:
- Why supervision, not direct production, is the right practice here:
- Prerequisite phases students have already practiced, and the evidence:

## The Delegated Task

Bounded, multi-step, inspectable, safe to fail.

- Task:
- Why it is safe to delegate in a classroom:
- Number of steps or stages:
- Time box:
- What a good final product looks like:

## Task Brief Students Must Write

Students write this before touching AI. Faculty review it first.

- Goal and audience:
- Steps and the AI role in each:
- Sources allowed and excluded:
- Output standard:
- Stop rules and escalation conditions:

## Intermediate Inspection Points

- What students must inspect mid-run (intermediate outputs, source use, drift from the brief):
- What evidence of each inspection they record:
- What faculty observe during the run:

## Stop And Escalation Rules

- Conditions that must halt the run:
- What students do when uncertain (escalate, verify, or refuse):
- What may never be delegated in this exercise:

## Final Integration Without AI

- What students must do unaided after the run (integrate, judge, defend):
- Final judgment students state in first person:

## Assessment: Does Judgment Survive Delegation?

| Dimension | Thin (1) | Strong (4) |
| --- | --- | --- |
| Task brief quality | Vague goal, no stop rules | Bounded goal, explicit criteria and stop rules |
| Inspection rigor | Accepted intermediate work unread | Caught and corrected a real problem mid-run |
| Stop-rule discipline | Kept going on momentum | Halted or escalated when conditions were met |
| Ownership of result | "The AI did it" | Defends the final judgment in first person |
| Transfer | Cannot adapt the method | Explains how the brief changes for a changed case |

## Oral-Defense Questions

- What did you inspect, and what did you find?
- What did the system get wrong, and when did you notice?
- What would have triggered your stop rule, and did anything come close?
- State the final judgment as your own. What in it did you change from the AI's version?
- How would your task brief change for [a changed case]?

## Trace Artifact

Keep it lean: task brief, inspection notes, stop-rule events, final judgment, one paragraph on what the student would change.