Course 02 Content Spec: First-Principles Problem Decomposition

1. Title and Source Files Used

Course title: First-Principles Problem Decomposition

Owned course folder: 02-first-principles-decomposition

Source files used as base truth:

• 02-first-principles-decomposition/curriculum.md
• 02-first-principles-decomposition/website-prompt.md

Purpose of this file: Convert the curriculum into a delivery-ready instructional specification that a facilitator, course builder, LMS operator, or AI-assisted grader can implement without needing to infer missing structure.

2. Design Decisions at the Top

1. Curriculum truth is preserved; structure is expanded.

The six 90-minute sessions, three-week pacing, core exercises, assessment weights, and central philosophy remain intact. This file adds sequencing, activity mechanics, outputs, facilitator guidance, and grading logic.

1. The course teaches a repeatable method, not a collection of opinions.

Students should leave with a reusable decomposition protocol they can apply to industries, systems, and personal decisions. Every session therefore reinforces the same method with different objects of analysis.

1. The pedagogy moves from visible assumptions to deep rebuilds.

Session order intentionally progresses from assumption spotting, to structured decomposition, to case analysis, to applied rebuilding, to personal transfer, to final synthesis.

1. Outputs are designed to be AI-evaluable.

Written and presentation deliverables use explicit structures so an LLM evaluator can distinguish between shallow critique and actual first-principles reasoning.

1. This is rigorous but accessible for students ages 15-25 with mixed backgrounds.

Domain expertise is not required. Scaffolds, templates, and facilitator prompts reduce intimidation while keeping intellectual standards high.

1. The website design prompt informs tone and visual implementation, not curriculum changes.

Terms such as "The Three Layers," "The Tearing Down Protocol," "What are you assuming?" and "Before/After" should appear consistently across slides, worksheets, LMS modules, and landing-page copy.

1. Assessment privileges plausibility over cleverness.

Novel ideas only score well when they distinguish real constraints from conventions and show a credible path from current state to rebuilt state.

3. Delivery Model Assumptions

• Format: Cohort-based short course
• Duration: 3 weeks
• Session cadence: 2 live sessions per week
• Live session length: 90 minutes
• Recommended cohort size: 12-30 students
• Recommended facilitation: 1 lead facilitator; 1 support facilitator or teaching assistant if cohort exceeds 20
• Modality: In-person, live online, or hybrid
• Default delivery assumption for this spec: Live synchronous seminar/workshop with asynchronous assignments between sessions
• Student profile: Ages 15-25; mixed academic background; no prerequisite expertise required; Course 01 recommended but not required
• Technology baseline: Shared slide deck, timer, breakout rooms or table groups, shared docs/slides/whiteboards, LMS or shared folder for submissions
• AI usage assumption: Students may use LLMs for brainstorming, but all submitted work must show explicit reasoning steps and personalized judgment
• Facilitator stance: Intellectual rigor, respectful disagreement, strong bias toward asking "Is that a law of reality or a habit?"
• Core instructional routines repeated throughout the course:
• Assumption harvest
• Physics vs. convention sort
• Current-solution reframing
• Rebuild from zero
• Minimum viable path back to reality

4. Detailed Course Content

Course-Level Throughline

Students learn one central operating sequence:

1. Name the problem as the current solution.
2. Surface assumptions hidden inside that solution.
3. Separate physics, economics, human nature, and regulation from custom and inertia.
4. Strip away convention.
5. Rebuild a new design from bedrock constraints.
6. Define the smallest credible action that moves from current state toward rebuilt state.

Module 1 / Session 1: What Are You Assuming?

Session outcome: Students can identify multiple types of assumptions and distinguish first-principles thinking from analogical thinking.

Session artifacts:

• Individual assumption audit sheet
• Group "law vs. habit" list
• Exit ticket defining first-principles reasoning in the student's own words

Lesson 1.1: The Assumption Audit

Timing: 35 minutes

Activity sequence:

• 0-10 min: Opening provocation
• Prompt: "What are you assuming right now about school, work, money, or success?"
• Students free-write individually for 3 minutes, then share one assumption in pairs.
• 10-20 min: Mini-lesson on assumption types
• Facilitator defines user, technology, market, institutional, and social assumptions.
• Blockbuster vs. Netflix is used as the anchor example.
• 20-35 min: Assumption spotting practice
• Small groups choose one industry: education, healthcare, or transportation.
• They list everything that feels "normal" or "obvious."

Facilitator moves:

• Keep pushing beyond surface complaints.
• Ask "What must someone be assuming for this design to seem normal?"
• Convert vague claims into explicit assumptions.
• Distinguish description from judgment.

Student outputs:

• Minimum 8 assumptions for the chosen industry
• At least 2 assumptions sorted into each of these categories: human behavior, technology, economics, social norms

Activity: The 5 Whys Applied to Industry

Timing: 20 minutes

Protocol:

• Each group selects three "that's just how it works" statements.
• For each statement, students ask "why?" or "who decided this?" five times or until they hit a bedrock constraint.

Facilitator moves:

• Interrupt when students confuse regulation with physics.
• Ask "If the law changed tomorrow, would this still be true?"
• Ask "If you had to rebuild this on a new planet, what would remain?"

Student outputs:

• Three traced chains from surface norm to underlying rationale
• One identified false constraint per group

Lesson 1.2: First-Principles vs. Analogical Thinking

Timing: 25 minutes

Activity sequence:

• 0-10 min: Direct instruction
• Analogical thinking defined as borrowing from precedent.
• First-principles thinking defined as reasoning from what must be true.
• 10-20 min: Comparison exercise
• Prompt: "How would most people improve school? How would a first-principles thinker approach education?"
• 20-25 min: Reading excerpt discussion
• Students respond to a short excerpt on first-principles reasoning.

Facilitator moves:

• Model one weak analogical answer and one strong first-principles answer.
• Make students name the constraint class they are using: physics, economics, human nature, regulation, convention.

Student outputs:

• Two-column comparison: analogical solution vs. first-principles approach
• Exit ticket: one sentence answering "What is first-principles thinking?"

Session 1 Close

Timing: 10 minutes

Debrief prompts:

• Which assumption felt hardest to challenge?
• What makes assumptions invisible?
• What is the cost of solving the wrong problem well?

Homework assigned: Assumption log

• Students record 10 assumptions they notice in daily life before Session 2.

Module 2 / Session 2: The Decomposition Method

Session outcome: Students can apply the three-layer model and the tearing down protocol to a real-world problem.

Session artifacts:

• Three-layer decomposition map
• Fresh water problem worksheet
• Individual protocol summary card

Lesson 2.1: The Three Layers of Any Problem

Timing: 30 minutes

Activity sequence:

• 0-12 min: Diagram instruction
• Facilitator introduces:
• Physics layer: absolute limits, energy, information, time, cost floors, human needs
• Component layer: people, infrastructure, capital, tools, institutions, regulation
• Design layer: current packaging of the solution
• 12-20 min: Guided example
• Example object: school system
• Class sorts elements into the three layers.
• 20-30 min: Paired practice
• Pairs sort a new system into layers.

Facilitator moves:

• Reject category drift; keep students precise.
• Ask "Is this a limit, a subsystem, or a design choice?"
• Stress that the design layer is the most disposable.

Student outputs:

• One completed three-layer map
• Three examples of design choices mistaken for constraints

Lesson 2.2: The Tearing Down Protocol

Timing: 25 minutes

Protocol taught explicitly:

1. State the problem as the current solution.
2. Remove everything that is not physics, economics, or fundamental human need/behavior.
3. Identify the true problem space.
4. Rebuild from zero.

Activity sequence:

• 0-10 min: Facilitator models protocol on "school"
• 10-25 min: Students apply protocol in triads to "How do we get fresh water to arid regions?"

Facilitator moves:

• Demand precise language: "The current solution is..."
• Ask "What survives if pipes, trucks, governments, and current vendors disappear?"
• Prevent magical thinking by requiring plausible energy, cost, and logistics assumptions.

Student outputs:

• One protocol worksheet with all four steps completed
• One "what survives?" list
• One rebuild concept sketch

Activity: Fresh Water Decomposition Challenge

Timing: 25 minutes

Prompt: Rebuild water access for arid regions without defaulting immediately to current infrastructure assumptions.

Required student deliverables during the activity:

• 3 physics constraints
• 3 economics/logistics constraints
• 3 assumptions to discard
• 1 rebuilt concept
• 1 sentence on why the rebuilt concept is better than the status quo under the chosen assumptions

Facilitator moves:

• Push for specificity on transport, storage, purification, maintenance, and incentives.
• Ask "Who maintains this?" and "What is the failure mode?"

Session 2 Close

Timing: 10 minutes

Debrief prompts:

• Which layer was easiest to identify?
• Where did your group accidentally smuggle convention back in?
• What does a strong rebuild need in order to stay credible?

Homework assigned: Mini decomposition memo

• 400-600 words on one system the student wants to challenge using the four-step protocol.

Module 3 / Session 3: Case Study Deep Dive

Session outcome: Students can perform a full industry decomposition and compare assumption-bound versus rebuilt designs.

Session artifacts:

• Transportation case notes
• School system case notes
• Group industry decomposition deck or one-pager

Lesson 3.1: Case Study - Transportation

Timing: 25 minutes

Activity sequence:

• 0-10 min: Brief lecture on the automobile in 1900 vs. 2025
• 10-15 min: Students identify what problem a car actually solves
• 15-25 min: Groups list which assumptions held in 1900 and which have broken

Facilitator moves:

• Keep attention on function over object.
• Ask "What is the transport job to be done?"
• Separate private car ownership from mobility itself.

Student outputs:

• Problem statement framed as movement of people/goods from A to B
• Assumption table: still valid / broken / uncertain

Lesson 3.2: Case Study - The School System

Timing: 20 minutes

Activity sequence:

• 0-8 min: Facilitator frames school as a bundle of functions: skill transmission, socialization, credentialing, childcare
• 8-20 min: Groups sort which functions are fundamental and which current structures are conventional

Facilitator moves:

• Ask "If AI changes instruction cost, what changes and what does not?"
• Ask "Which function are you optimizing for?"

Student outputs:

• Function map of schooling
• List of conventional design features that may not survive an AI-era rebuild

Group Exercise: Industry Decomposition

Timing: 35 minutes

Group options: Healthcare, housing, finance, education

Required group deliverable structure:

1. Current system definition
2. Top 10 embedded assumptions
3. Three-layer decomposition
4. Rebuilt system concept
5. Minimum viable path from today's world to the rebuilt design

Presentation format: 4-minute share-out plus 2-minute questions

Facilitator moves:

• Require teams to name their most important non-negotiable physical or economic constraints.
• Challenge teams that merely propose software layers on top of old assumptions.
• Ask "What does your rebuild stop doing?"

Student outputs:

• Group decomposition artifact suitable for grading

Session 3 Close

Timing: 10 minutes

Debrief prompts:

• What did your group think was fundamental that turned out to be conventional?
• What makes a rebuild feel uncomfortable or unrealistic?
• How do you know when you are being rigorous versus merely contrarian?

Assignment launched: Industry Decomposition submission

• Group artifact due before Session 4.

Module 4 / Session 4: The Rebuild Exercise

Session outcome: Students can rebuild a system from scratch under explicit bedrock constraints and defend the design.

Session artifacts:

• Pair design worksheet
• Constraint list
• 3-minute pair pitch

Workshop Brief: Universal Financial Advice

Timing: Full 90-minute workshop

Challenge statement: Every person on Earth needs access to high-quality financial advice. Design the system from scratch.

Removed by default:

• Bank branches
• Financial advisors as the current profession
• Existing regulatory frameworks
• Existing financial products

Kept by default:

• Internet connectivity as generally available
• Human cognitive bias around risk, loss, status, and trust
• Need for trustworthy guidance

Workshop Sequence

0-10 min: Briefing

• Facilitator reviews the challenge, removed constraints, and preserved constraints.

10-20 min: Constraint extraction

• Pairs produce:
• 5 fundamental constraints
• 3 human behavior realities
• 2 likely failure modes

20-45 min: System design

• Pairs design the advice system:
• Intake
• Diagnosis
• Recommendation logic
• Trust/verification mechanism
• Incentive structure
• Delivery channel
• Feedback loop

45-60 min: Stress test

• Another pair attacks the design:
• Where can it fail?
• Where can it be gamed?
• What becomes too expensive?
• What assumption is still hidden?

60-80 min: Revision and pitch prep

• Pairs revise and prepare a 3-minute pitch.

80-90 min: Rapid pitches

• Selected pairs present live; others submit recorded or written pitches.

Facilitator moves:

• Ask "Why should anyone trust this system?"
• Ask "What incentive keeps the system honest?"
• Ask "How does this handle low-literacy, low-trust, or low-income users?"
• Stop pairs from unconsciously recreating banks under new language.

Student outputs:

• 5 bedrock constraints
• System diagram
• 3-minute pitch
• One-paragraph defense of why this beats the current system

Homework assigned: Workshop reflection

• 300-500 words: What hidden assumption did your pair initially carry into the exercise, and how did you remove it?

Module 5 / Session 5: Personal Application

Session outcome: Students can apply decomposition to their own education, career, goals, or life strategy and translate insight into action.

Session artifacts:

• Personal decomposition worksheet
• Rebuilt personal strategy statement
• One-week action plan

Lesson 5.1: Applying Decomposition to Your Own Life

Timing: 35 minutes

Activity sequence:

• 0-10 min: Facilitator frames the challenge
• Common hidden assumption: "Success looks like the inherited path."
• 10-20 min: Guided personal writing
• Students answer the four curriculum questions.
• 20-35 min: Partner coaching
• Each student explains their assumed path and receives challenge questions.

Facilitator moves:

• Protect psychological safety; students may discuss identity, family expectations, money, or prestige.
• Ask "What are you treating as inevitable?"
• Ask "What is the actual constraint: money, time, skill, credential, geography, fear, or social approval?"

Student outputs:

• One completed personal decomposition
• One explicit false constraint identified

Lesson 5.2: From Decomposition to Action

Timing: 35 minutes

Activity sequence:

• 0-15 min: Mini-lesson on minimum viable path
• Rebuilt solutions must connect back to reality through small executable moves.
• 15-30 min: Students create a one-week action plan and a 30-day experiment.
• 30-35 min: Peer review using credibility checks

Credibility checks:

• Is the goal tied to a real constraint?
• Is the next step executable this week?
• Does the step test the new model rather than just restate it?

Facilitator moves:

• Push students away from vague resolutions.
• Ask "What can you do in seven days that gives evidence?"
• Require observable actions and evidence of completion.

Student outputs:

• Rebuilt approach to a personal goal
• One-week action step
• 30-day experiment outline

Session 5 Close

Timing: 20 minutes

Final presentation prep:

• Students draft a 5-minute final presentation with this required structure:

1. Assumption identified
2. Current obvious solution
3. First-principles rebuild
4. Minimum viable path
5. Immediate next action

Assignment launched: Personal Decomposition final submission and presentation prep

Module 6 / Session 6: Final Presentations and Debrief

Session outcome: Students publicly demonstrate first-principles reasoning and receive peer and AI-supported evaluation.

Session artifacts:

• Final presentation
• Peer review form
• Final reflection

Final Presentations

Timing: 60 minutes

Default presentation structure per student:

• 5 minutes presentation
• 2 minutes peer/facilitator questions

Required presentation elements:

• Problem reframed as the current solution
• Assumptions identified
• Distinction between real and assumed constraints
• Rebuilt solution
• Plausibility rationale
• Next action

Facilitator moves:

• Enforce time limits.
• Ask one rigor question and one action question per presenter.
• Keep the distinction between "interesting" and "credible" visible.

Student outputs:

• Final presentation deck, one-pager, or structured memo

Peer Review Framework

Timing: 15 minutes

Peer review questions:

• Did the presenter identify real assumptions rather than surface frustrations?
• Is the rebuilt solution physically and economically plausible?
• Is there a credible path from current state to rebuilt state?
• What is strongest in the reasoning?
• What still feels inherited from the original system?

Closing Debrief: The Cost of Not Decomposing

Timing: 15 minutes

Discussion prompts:

• What problem were you carrying because you never questioned the assumption?
• Where in your life will this method change your decisions next?
• What daily habit will keep you from falling back into borrowed thinking?

Final student output:

• 150-250 word reflection on one assumption they will now notice differently

5. Assignments and Artifacts

Assignment 1: Assumption Log

• When assigned: End of Session 1
• Format: Individual notes
• Expected length: 10 observations minimum
• Purpose: Train assumption detection in ordinary contexts
• Graded: Completion only

Assignment 2: Mini Decomposition Memo

• When assigned: End of Session 2
• Format: 400-600 word memo
• Purpose: Demonstrate early use of the four-step protocol
• Suggested prompt: Choose one system you interact with. State the current solution, identify key assumptions, strip to fundamentals, and propose a rebuild.
• Graded: Low-stakes formative

Assignment 3: Industry Decomposition

• When assigned: Session 3
• Weight: 25%
• Format: Group one-pager, deck, or structured memo
• Required sections:
• Current system
• Assumptions
• Three layers
• Rebuild
• Minimum viable path
• Primary evidence of mastery: Ability to distinguish system function from inherited implementation

Assignment 4: Workshop Output - Universal Financial Advice Rebuild

• When assigned: Session 4
• Weight: 25%
• Format: Pair pitch plus worksheet
• Required artifacts:
• 5 constraints
• System design
• Trust/incentive logic
• Failure mode analysis
• 3-minute pitch
• Primary evidence of mastery: Ability to build from explicit bedrock constraints without recreating legacy institutions by default

Assignment 5: Personal Decomposition

• When assigned: Session 5
• Weight: 30%
• Format: Individual written submission
• Required sections:
• Inherited assumption
• Current obvious path
• Real constraints vs. assumed constraints
• Rebuilt path
• One-week action and 30-day experiment
• Primary evidence of mastery: Transfer of the method to the student's own life

Assignment 6: Final Presentation

• When assigned: Session 5, delivered Session 6
• Weight: 20%
• Format: 5-minute presentation
• Primary evidence of mastery: Coherent synthesis, plausibility, and actionability under questioning

6. AI/LLM Grading and Assessment Framework

Core LLM Evaluation Principle

The LLM is not grading for polish, ideology, or agreement with the student's conclusion. It is grading for evidence of disciplined first-principles reasoning.

Required evaluation dimensions across major assignments

1. Problem framing

• Did the student restate the issue as the current solution rather than the assumed object?

1. Assumption detection

• Did the student identify hidden assumptions explicitly and concretely?

1. Constraint discrimination

• Did the student separate physics/economics/human behavior from convention, regulation, preference, and habit?

1. Decomposition quality

• Did the student break the system into meaningful layers or components rather than provide vague critique?

1. Rebuild originality with plausibility

• Is the rebuilt proposal materially different from the inherited model while still feasible in principle?

1. Minimum viable path

• Did the student explain how one could move from present conditions toward the rebuilt approach?

1. Specificity and evidence of reasoning

• Are claims concrete, internally consistent, and tied to explicit assumptions?

Assignment-specific scoring heuristics

Industry Decomposition

• High score indicators:
• Names 8-10 real assumptions with low overlap
• Identifies at least one assumption most people in the field rarely question
• Cleanly separates system function from current delivery mechanism
• Rebuild changes structure, not just features
• Minimum viable path includes sequencing, stakeholders, and likely barriers
• Low score indicators:
• Complains about the system without naming assumptions
• Treats laws or current business models as immutable physics
• Rebuild is only a slightly improved version of the current system

Workshop Output

• High score indicators:
• Constraints reflect human trust, incentives, misuse, and access realities
• Design includes verification, recommendation logic, and anti-gaming logic
• Students visibly removed default finance-industry assumptions
• Low score indicators:
• Proposal is basically "an app with AI advice"
• No trust mechanism
• No failure mode analysis

Personal Decomposition

• High score indicators:
• Student names a real inherited script in career, education, or self-concept
• Distinguishes emotional/social constraints from physical/economic ones
• New path is concrete and paired with a real experiment
• Low score indicators:
• Submission stays abstract or performative
• "Rebuild" is wishful thinking without tradeoffs
• Action step is not testable within one week

Final Presentation

• High score indicators:
• Presentation structure is clear and complete
• Student withstands questions without collapsing into vagueness
• Audience can identify the old assumption, the rebuilt logic, and the next action
• Low score indicators:
• Presentation is inspirational but methodologically thin
• Constraints are not examined
• Next step is missing or generic

LLM scoring scale

• 4 - Exemplary: Clear, precise, layered reasoning; hidden assumptions surfaced; rebuild is plausible and meaningfully distinct; next steps are credible.
• 3 - Proficient: Good reasoning with moderate specificity; some assumptions and constraints identified; rebuild mostly plausible; action path present.
• 2 - Developing: Partial understanding; assumptions or constraints remain vague; rebuild underdeveloped or derivative; action path weak.
• 1 - Insufficient: Little evidence of first-principles reasoning; mostly opinion, complaint, or analogy; implausible or unsupported rebuild.

7. Rubrics, Scoring Criteria, and Evaluator Prompt Guidance

Master Rubric for Major Assignments

Score conversion guidance

• 90-100: Exemplary command of the method
• 75-89: Solid and credible application
• 60-74: Partial mastery; needs sharper decomposition
• Below 60: Insufficient evidence of first-principles reasoning

Evaluator prompt guidance for LLM grading

Use the following evaluator instructions when grading a major assignment:

You are evaluating student work for a course on first-principles problem decomposition.

Your job is to assess the quality of the student's reasoning, not whether you agree with their proposed solution.

Focus on:
1. Whether the student reframed the problem as the current solution.
2. Whether they identified concrete hidden assumptions.
3. Whether they distinguished bedrock constraints (physics, economics, human behavior) from convention or inherited design.
4. Whether their rebuild is materially different from the status quo and still plausible.
5. Whether they proposed a credible path from current reality to the rebuilt approach.

Do not reward vague boldness, buzzwords, or polished language.
Do not punish unconventional conclusions if the reasoning is coherent.
Penalize magical thinking, generic claims, or proposals that simply rename the current system.

Return:
- Criterion-by-criterion scores from 1 to 4
- A weighted total score out of 100
- 3 evidence-based strengths
- 3 evidence-based improvement points
- A short paragraph of feedback addressed to the student
```

### Optional structured output schema for LLM evaluators

```json
{
  "assignment_type": "industry_decomposition",
  "scores": {
    "problem_reframing": 0,
    "assumption_identification": 0,
    "constraint_analysis": 0,
    "rebuild_quality": 0,
    "plausibility_and_tradeoffs": 0,
    "actionability": 0,
    "communication_clarity": 0
  },
  "weighted_total": 0,
  "strengths": [
    "",
    "",
    ""
  ],
  "improvements": [
    "",
    "",
    ""
  ],
  "student_feedback": ""
}
```

## 8. Feedback Strategy: What Strong/Average/Weak Responses Look Like and How an LLM Should Respond

### Strong response profile

**What it looks like:**
- The student names the inherited system cleanly and does not confuse it with the underlying need.
- Assumptions are concrete and often surprising.
- Constraints are classified correctly.
- The rebuild meaningfully changes the structure of the solution.
- The student acknowledges tradeoffs and proposes a real next step.

**How the LLM should respond:**
- Validate the rigor specifically.
- Name the strongest reasoning move.
- Push the student one level deeper on second-order effects or implementation sequencing.

**Example LLM response pattern:**
"Your strongest move was separating the need for credentialing from the current school format. That shows real decomposition rather than dissatisfaction with school. Your rebuild is plausible because you addressed cost, trust, and transition. To improve further, specify which stakeholder you would test this with first and what evidence would prove the model works."

### Average response profile

**What it looks like:**
- The student understands the assignment and identifies some assumptions.
- The rebuild has potential but remains partly trapped inside inherited structures.
- Constraint analysis is uneven.
- The action path exists but lacks specificity.

**How the LLM should respond:**
- Acknowledge the valid reasoning already present.
- Point out exactly where convention is still being treated as necessary.
- Ask for one revision that sharpens constraints and one revision that sharpens execution.

**Example LLM response pattern:**
"You correctly identified that the current system bundles several different functions together. However, your rebuild still assumes the same institutional structure without explaining why it must remain. Revise by naming which part is truly constrained by economics or human behavior and which part is simply inherited. Then tighten your next step into a seven-day experiment."

### Weak response profile

**What it looks like:**
- The student stays at the level of complaint, slogan, or fantasy.
- Assumptions are generic or absent.
- The rebuild is vague, impossible, or just a rebranding of the current system.
- No clear distinction is made between physical constraints and habits.

**How the LLM should respond:**
- Be direct and diagnostic, not punitive.
- Explain what is missing in the method.
- Give a narrow revision path rather than overwhelming the student.

**Example LLM response pattern:**
"This response does not yet show first-principles reasoning. You describe what you dislike about the current system, but you do not identify the assumptions holding it in place. Start over with two steps only: first, write 'The current solution is...' in one sentence; second, list five assumptions that must be true for that solution to make sense. Then decide which of those are physics, economics, human behavior, or convention."

### LLM feedback tone requirements

- Be precise, not flattering.
- Reference evidence from the submission.
- Separate reasoning quality from writing polish.
- Prefer one concrete revision step over broad advice.
- When weak, diagnose the missing reasoning move explicitly.

### LLM guardrails

- Do not invent evidence not present in the student's work.
- Do not over-credit futuristic ideas that ignore cost, trust, incentives, time, or maintenance.
- Do not under-credit simple solutions if they are correctly reasoned from bedrock constraints.
- When uncertain, score down for missing evidence rather than infer competence.

## Implementation Notes for Course Builders

- Build worksheets around the repeated sequence: current solution, assumptions, constraints, rebuild, path from here.
- Keep naming consistent across slides, LMS pages, and downloadable templates:
  - "The Assumption Audit"
  - "The Three Layers"
  - "The Tearing Down Protocol"
  - "Before / After"
  - "What are you assuming?"
- Use the website prompt's visual language for presentation assets:
  - Warm off-white background
  - Near-black text
  - Acid yellow-green for insight moments
  - Red for inherited assumptions
  - Steel blue for structural/physics concepts
- Preserve ample discussion time in Sessions 3, 4, and 6. Those sessions carry the deepest learning and strongest assessment evidence.

## Recommended Facilitator Prep Checklist

- Prepare slides with at least one completed model decomposition.
- Create worksheets for assumption audit, three-layer map, protocol steps, and personal decomposition.
- Decide in advance how groups will submit artifacts.
- Prepare two or three backup case examples in case students choose trivial or overly familiar systems.
- Align any AI grading tool to the rubric before the course begins.