Course 03 Implementation Spec: The "No Playbook" Case Studies

1. Title and Source Files Used

Course title: The "No Playbook" Case Studies

Owned output file: 03-no-playbook-case-studies/03_no_playbook_case_studies_course_content.md

Source files used as base truth:

• 03-no-playbook-case-studies/curriculum.md
• 03-no-playbook-case-studies/website-prompt.md

Purpose of this file: Translate the course curriculum into a delivery-ready course-content specification that a facilitator, LMS builder, evaluator, or AI-assisted teaching workflow can execute without inventing missing instructional structure.

Source-of-truth commitments for this spec:

• Preserve the curriculum's 3-week, 6-session, 90-minute live format.
• Preserve the central thesis: the advantage is not following a playbook but operating well when none exists.
• Preserve the named core cases from the curriculum:
• XPRIZE origin
• SpaceX early years and Falcon 1 period
• OpenAI formation and restructuring
• Cross-case synthesis including Airbnb
• Preserve the curriculum's four assessed outputs:
• case study analysis
• cross-case matrix plus personal map
• ambiguity simulation participation
• final reflection
• Preserve the website prompt's tone and visual direction as implementation guidance for slides, handouts, and LMS framing, without changing the curriculum itself.

2. Design Decisions at the Top

Instructional design decisions

1. This course teaches decision quality under ambiguity, not founder trivia.

Students should leave with a reusable way of thinking about unclear, high-stakes situations, not a scrapbook of startup stories.

1. Case studies are structured around decision points.

Every major case is taught through moments when the actors had to choose with incomplete information, contradictory signals, and visible downside.

1. Reasoning matters more than outcome worship.

Students should learn to analyze whether a decision was rigorous given the information available then, not whether the eventual outcome became famous.

1. The emotional load of ambiguity is part of the curriculum.

Students need language for fear, doubt, social pressure, and team morale, because no-playbook situations are not purely analytical.

1. The course escalates from recognition to reenactment.

The sequence intentionally moves from spotting ambiguity, to dissecting cases, to extracting a framework, to applying it to the student's own life, to live simulation.

1. Students must repeatedly separate knowns, unknowns, and unknowables.

This distinction is a core operating habit for ambiguous environments and should appear in discussion, worksheets, and grading.

1. Outputs are designed to be LLM-evaluable.

Every major deliverable uses explicit sections and evidence expectations so an AI evaluator can assess depth of reasoning instead of rewarding style alone.

1. The website prompt informs presentation style.

Slides, worksheets, and LMS modules should echo the documentary framing:

• episode-like case structure
• high-contrast decision moments
• recurring language such as "The Four Constants," "decision reenactment," and "when the map didn't exist"

1. Supplemental mini-cases support comparison but do not replace the core cases.

Airbnb is the required cross-case comparator from the curriculum. Stripe and Square may appear as short examples only if used to sharpen pattern recognition.

Facilitation design decisions

1. Facilitators should act like disciplined case moderators.

The job is to keep students close to evidence, timing, incentives, constraints, and decision logic.

1. Hero worship and cynicism both need correction.

If students romanticize founders, ask about costs and tradeoffs. If students dismiss them, ask what alternative they would have chosen with the same information.

1. Discussion should privilege specific claims.

The facilitator should frequently ask:

• What did they actually know?
• What did they think they knew?
• What was still unknowable?
• What options were live at that moment?

1. Personal transfer is required, not optional.

The course only lands if students can identify a real no-playbook situation in school, work, family, community, or creative life.

3. Delivery Model Assumptions

Cohort and format assumptions

• Duration: 3 weeks
• Live sessions: 6 sessions
• Session length: 90 minutes each
• Recommended cohort size: 12 to 28 students
• Recommended staffing: 1 lead facilitator; add 1 teaching assistant for cohorts above 20
• Delivery mode: in-person, live online, or hybrid
• Default assumption for this spec: live synchronous discussion/workshop plus short asynchronous assignments
• Student age range: 15 to 25
• Prior background: mixed; no prerequisite technical or startup knowledge required

Instructional environment assumptions

• Students have access to a slide deck, shared docs or LMS, and either breakout rooms or small-group tables.
• Facilitator can project or distribute short case packets, quotes, timelines, and decision-point handouts.
• Students can submit written work digitally.
• Students may use AI tools for brainstorming or first-pass outlining, but all final submissions must show the student's own judgment, synthesis, and evidence selection.

Time and workload assumptions

• Live time: 9 total hours
• Async workload: approximately 5 to 7 hours over 3 weeks
• Reading load should stay light and targeted:
• brief excerpts
• short interview clips or transcripts
• decision-point summaries
• The course should feel demanding in reasoning, not burdensome in reading volume.

Materials assumptions

Required materials:

• facilitator slide deck
• case packets for XPRIZE, SpaceX, OpenAI, and Airbnb mini-case comparison
• decision memo worksheet
• Four Constants worksheet
• cross-case matrix template
• personal no-playbook map template
• scenario room packet for Session 6
• grading rubrics and LLM evaluator prompts

Recommended instructional routines:

• quickwrite
• pair share
• small-group decision lab
• cold-call evidence defense
• structured debrief
• exit ticket

Accessibility and inclusion assumptions

• Examples should be translated into terms students from non-startup backgrounds can engage with.
• "No-playbook" situations may come from school leadership, family responsibilities, community organizing, freelancing, sports, religious groups, or creative projects.
• Students may anonymize personal situations in written work.

Assessment assumptions

• The LLM acts as a first-pass grader and feedback assistant.
• A human reviews:
• borderline submissions
• emotionally sensitive reflections
• suspected fabricated case evidence
• unusually high or low scores

4. Detailed Course Content

Course-Level Throughline

Students repeatedly practice one operating sequence:

1. Detect that a situation is genuinely no-playbook.
2. Name the underlying problem instead of hiding inside inherited language.
3. Separate what is known, unknown, and unknowable.
4. Clarify stakes, constraints, and downside tolerance.
5. Generate options without pretending certainty exists.
6. Choose a move and defend why it is appropriate now.
7. Treat failure, surprise, and feedback as information for the next move.

Course-Level Recurring Framework: The Four Constants

The curriculum names four constants in no-playbook situations. They should recur across sessions, worksheets, and grading:

1. The Physics Test

What is actually true about constraints, timing, cost, capabilities, and reality?

1. The Stakes Clarity

What are we actually willing to lose, spend, delay, or risk?

1. The Information Asymmetry Map

What do we know, need to know, and accept that we cannot know in time?

1. The First-Mover Calculus

Is the advantage in being correct, or in moving before certainty exists?

Module 1 / Session 1: Recognizing No-Playbook Moments

Session goal: Students can distinguish routine, complicated, complex, and chaotic situations and identify what makes a situation "no-playbook."

Session outcomes:

• Students can define a no-playbook situation in their own words.
• Students can classify ambiguous scenarios with justification.
• Students can analyze the XPRIZE origin as a case of creating a structure where no template existed.

Session artifacts:

• scenario classification sheet
• XPRIZE decision note
• exit ticket definition

Session 1 run of show

Session 1 facilitator guidance

• Do not let students call everything "chaotic." Precision matters.
• When students confuse difficulty with ambiguity, ask whether a competent expert already knows the path.
• In the XPRIZE discussion, redirect from "genius idea" language toward incentive design, legitimacy, and timing.

Session 1 assignment launch

Homework: Ambiguity observation log

• Students collect 3 examples from real life or media where a person or organization faced unclear precedent.
• For each example, students record:
• What made it no-playbook?
• What was known?
• What was unknown?
• What was the main perceived risk?

Module 2 / Session 2: SpaceX Deep Dive I - Early Formation and Failure

Session goal: Students analyze how SpaceX operated when the private-space playbook did not yet exist and evaluate early decisions under uncertainty.

Session outcomes:

• Students can identify the specific assumptions SpaceX broke.
• Students can analyze early SpaceX decisions using knowns, unknowns, and alternatives.
• Students can produce a defensible judgment about whether persistence after failure was rational.

Session artifacts:

• SpaceX timeline sheet
• failure analysis worksheet
• decision memo draft

Session 2 run of show

Session 2 facilitator guidance

• Keep the case anchored to timing. Students must answer as if they are in 2002-2006, not with hindsight.
• If students say "they should just raise more money," ask whether credibility, timing, and investor appetite made that realistic then.
• Push the distinction between mission commitment and irrational stubbornness.

Session 2 homework

Homework: SpaceX decision memo

• 500-700 words
• Prompt: "After one failed launch, what should SpaceX do next, and why?"
• Required structure:
• situation summary
• what is known
• what is unknown
• options considered
• recommended move
• biggest downside of your recommendation

Module 2 / Session 3: SpaceX Deep Dive II - Persistence, Reuse, and Framework Extraction

Session goal: Students analyze later SpaceX decision points and extract general principles for operating under uncertainty.

Session outcomes:

• Students can evaluate persistence after repeated failure.
• Students can analyze vertical integration and reuse as no-playbook strategic choices.
• Students can extract transferable decision principles from the SpaceX case.

Session artifacts:

• post-failure option matrix
• principle extraction sheet
• case analysis outline

Session 3 run of show

Required principle categories for Session 3

Each group must state one principle in each category:

• how to treat failure
• how to treat expertise
• how to think about cost
• how to decide when persistence becomes irrational
• how to protect morale without lying

Session 3 facilitator guidance

• Reject vague principles such as "never give up" or "believe in yourself."
• Ask, "What evidence from the case supports that rule?"
• If students ignore downside, ask what would make the same behavior look irresponsible.

Session 3 assignment launch

Major assignment launched: Case Study Analysis

• Students choose SpaceX or OpenAI.
• Due after Session 4.
• See Section 5 for full requirements.

Module 3 / Session 4: OpenAI's Formation and Structural Ambiguity

Session goal: Students examine a different type of no-playbook problem: not rockets, but organization design under emerging technology, mission tension, and capital constraints.

Session outcomes:

• Students can explain why OpenAI's original structure was unusual relative to academic labs and corporate AI labs.
• Students can analyze the tension between safety, openness, talent competition, and capital needs.
• Students can compare structural ambiguity to technical ambiguity.

Session artifacts:

• OpenAI structure map
• tradeoff analysis chart
• case analysis planning notes

Session 4 run of show

Session 4 facilitator guidance

• Keep the class away from tribal debates about whether a given organization is "good" or "bad."
• The teaching question is: what structure would you design when your aims conflict and no stable institutional template exists?
• If students produce simplistic answers like "just stay nonprofit," ask how talent, compute, and capital would work in practice.

Optional supplemental mini-case notes

If time allows, the facilitator may introduce 5-minute comparison snapshots of Airbnb, Stripe, or Square to illustrate different forms of no-playbook uncertainty:

• Airbnb: market legitimacy and trust creation
• Stripe: infrastructure simplification in a painful domain
• Square: productizing behavior change for small merchants

These are support examples only and should not displace the core OpenAI session content.

Module 4 / Session 5: Cross-Case Synthesis and Personal Mapping

Session goal: Students synthesize patterns across cases and apply the Four Constants to a personal no-playbook situation.

Session outcomes:

• Students can compare technical, organizational, and market ambiguity across cases.
• Students can use the Four Constants as a practical framework.
• Students can locate one current or upcoming no-playbook situation in their own life.

Session artifacts:

• cross-case matrix
• personal no-playbook map
• draft final reflection notes

Session 5 run of show

Required structure for the personal no-playbook map

Each student's map must include:

• one-sentence situation statement
• why this is genuinely no-playbook
• 3 knowns
• 3 unknowns
• 2 unknowables
• main downside if they move too early
• main downside if they wait too long
• one next action
• one evidence signal that would cause them to update the plan

Session 5 facilitator guidance

• Some students will choose situations that are merely stressful, not actually ambiguous. Ask whether a clear playbook exists and they are simply avoiding it.
• Some students will choose cases too large and abstract. Push them toward a decision they may realistically face in the next 3 to 12 months.
• Reward specificity over drama.

Session 5 assignment launch

Assignment: Cross-Case Matrix + Personal Map

• Due before Session 6
• See Section 5 for details

Module 5 / Session 6: Operating in Ambiguity - The Scenario Room

Session goal: Students demonstrate live reasoning in a simulated no-playbook environment and reflect on how they behave under ambiguity.

Session outcomes:

• Students can apply the Four Constants under time pressure.
• Students can explain and defend a recommendation without pretending certainty.
• Students can reflect on their instincts under ambiguity and identify what to improve.

Session artifacts:

• scenario analysis worksheet
• group reasoning presentation
• final reflection launch notes

Session 6 run of show

Scenario Room required deliverable structure

Each group must submit:

1. Problem framing
2. Four Constants analysis
3. Three options considered
4. Recommended move now
5. Biggest unresolved uncertainty
6. Trigger that would cause a course correction
7. One sentence on what failure would teach them next

Session 6 facilitator guidance

• Teams often jump to solutions too fast. Slow them down by asking what reality they are assuming.
• Do not reward polished confidence that skips evidence.
• If teams fixate on only one scenario detail, ask what second-order consequences they are missing.
• The debrief should include feeling-state analysis. Students need to notice whether they freeze, bluff, over-control, or over-index on consensus.

5. Assignments and Artifacts

A. Case Study Analysis

Weight: 30%

Student choice: SpaceX or OpenAI

Deliverable length: 1,400 to 1,800 words

Purpose: Analyze a real no-playbook decision environment, reconstruct the options available at the time, and judge the quality of the decision logic.

Required sections:

1. case context and timeline
2. the no-playbook moment
3. knowns, unknowns, and unknowables
4. options available at the time
5. decision taken and why it may have seemed rational
6. strongest alternative path
7. what this case teaches about operating without a playbook

Minimum evidence requirements:

• at least 4 concrete case details
• at least 2 clearly stated tradeoffs
• at least 1 paragraph that resists hindsight bias
• explicit use of at least 2 of the Four Constants

Common failure modes:

• summary without analysis
• hindsight certainty
• founder worship or moral grandstanding
• vague claims like "they believed in the mission"

B. Cross-Case Matrix + Personal No-Playbook Map

Weight: 20%

Deliverable format: one visual matrix plus a 350-500 word personal explanation

Purpose: Compare ambiguity types across cases and transfer the framework to the student's own situation.

Required components:

• matrix with SpaceX, OpenAI, Airbnb, and one student-selected comparator
• labels for both axes
• 2-3 sentence justification for each placement
• personal no-playbook map using the required structure from Session 5
• short explanation of what type of ambiguity the student's own situation most resembles

Minimum evidence requirements:

• at least one comparison sentence across cases
• at least one sentence explaining why the student's situation is not merely "difficult" but actually no-playbook

C. Scenario Room Participation and Group Reasoning Artifact

Weight: 20%

Deliverable format: group worksheet plus live participation score

Purpose: Evaluate reasoning quality under time pressure and ambiguous conditions.

Scored elements:

• evidence of structured analysis
• contribution to team reasoning
• quality of recommendation
• quality of uncertainty acknowledgment
• quality of verbal defense during presentation or discussion

Required group artifact:

• the 7-part Scenario Room structure listed in Session 6

Participation note: This component scores the quality of the student's reasoning moves, not whether the team picked the same option the facilitator would have chosen.

D. Final Reflection: "My No-Playbook Moment"

Weight: 30%

Deliverable length: 1,000 to 1,200 words

Purpose: Apply the course to a real ambiguity in the student's life and articulate a practical framework for action.

Required sections:

1. the situation
2. why no ready-made playbook exists
3. stakes and downside
4. what is known, unknown, and unknowable
5. how the Four Constants apply
6. what the student is currently tempted to do
7. what the student now believes is the best next move
8. what would cause them to update their view

Minimum evidence requirements:

• one clearly stated tension or fear
• one rejected option and why it was rejected
• one update trigger
• one specific next action within the next 14 days

Formative Artifacts Across the Course

These may be ungraded or low-stakes completion items, but they should be collected because they support final performance and can be used by LLM tutors:

• ambiguity observation log
• Session 2 SpaceX decision memo
• Session 3 principle extraction sheet
• Session 4 structure design proposal
• Session 5 personal map draft
• Session 6 scenario worksheet

6. AI/LLM Grading and Assessment Framework

Assessment model

The LLM should function as a disciplined first-pass evaluator. Its job is to:

• check completion
• score visible evidence against rubrics
• identify weak reasoning patterns
• generate revision guidance

The LLM is not the sole authority on final grades. Human review is required for emotional nuance, suspected fabrication, and edge cases.

Core grading principles

1. Grade reasoning under uncertainty, not rhetorical confidence.

Confident tone should not raise scores unless the submission shows disciplined analysis.

1. Reward accurate reconstruction of the decision environment.

High scores require evidence that the student understands what the actors knew then, not just what we know now.

1. Penalize hindsight bias.

If the student judges a decision mainly because the eventual outcome was good or bad, scores should drop.

1. Reward explicit uncertainty handling.

Strong work names unknowns, unknowables, tradeoffs, and update triggers.

1. Reward transfer, not just case recall.

The best submissions move from case facts to reusable principles or personal application.

1. Do not over-penalize writing polish.

Minor grammar issues matter only if meaning becomes unclear.

Artifact-by-artifact LLM evidence checks

Case Study Analysis

The LLM should check for:

• clear case timeline
• a real decision point rather than broad narrative summary
• differentiation between knowns, unknowns, and unknowables
• at least one alternative path evaluated seriously
• visible resistance to hindsight bias
• use of the Four Constants

Automatic downgrades:

• if the essay mostly retells history with little judgment
• if the student treats outcome as proof of decision quality
• if no downside or tradeoff is named

Cross-Case Matrix + Personal Map

The LLM should check for:

• all required cases present
• justification for each placement
• consistent axis logic
• evidence that the student's own situation is genuinely ambiguous
• one plausible next move and one update trigger

Automatic downgrades:

• if the matrix is visually present but unjustified
• if the student's personal example is just a routine task with discomfort attached
• if "unknowns" are really just missing effort rather than ambiguity

Scenario Room Participation and Group Artifact

The LLM should check for:

• structured use of the Four Constants
• clear option generation
• recommendation with uncertainty acknowledged
• course-correction trigger
• evidence of reasoning rather than bluffing

Automatic downgrades:

• if the group jumps from problem to answer without mapping constraints
• if the presentation claims certainty where the scenario does not allow it
• if no rejected option is discussed

Final Reflection

The LLM should check for:

• a real situation with consequences
• clear explanation of why no ready-made playbook exists
• honest articulation of tension, risk, or fear
• use of all Four Constants
• a next action and an update rule

Automatic downgrades:

• if the reflection reads like generic ambition advice
• if the student avoids real stakes
• if no rejected option appears
• if the "next move" is too vague to execute

Concrete assessment heuristics for LLM evaluation

High-signal indicators

The submission is likely strong if it includes several of these:

• the student names what would have had to be true for a decision to work
• the student distinguishes missing information from unknowable information
• the student identifies a non-obvious tradeoff
• the student names an update trigger or falsification condition
• the student explains why a plausible alternative was rejected
• the student uses precise situational language rather than motivational slogans

Low-signal indicators

The submission is likely weak if it includes several of these:

• "They just believed" or "they took a risk" with no mechanism
• "No one knew" when some things were in fact knowable
• "The lesson is never give up" without boundary logic
• generic praise or condemnation of a founder
• personal reflection with no specific next action
• repeated buzzwords such as innovation, disruption, mindset, vision, or courage without attached evidence

Heuristic caps for scoring

• If the student never separates knowns from unknowns, the work cannot score above competent on reasoning quality.
• If the student never names a real tradeoff, the work cannot score above competent on analysis quality.
• If the student evaluates a decision entirely through eventual outcome, the work cannot score above partial on decision judgment.
• If the personal reflection contains no concrete next step within 14 days, it cannot score above partial on actionability.
• If the Scenario Room artifact contains no update trigger, it cannot score above partial on uncertainty handling.

Automatic human-review flags

• suspected fabricated case details or invented quotations
• emotional distress, coercion, or high-risk personal situations in the final reflection
• clear evidence of plagiarism or AI-generated filler
• major contradictions within the submission
• too little content to verify completion

Recommended LLM output structure

For each assessed artifact, the LLM should return:

1. completion check
2. rubric scores by criterion
3. 2 to 4 evidence-based observations
4. 2 concrete revision moves
5. overall level: Strong / Competent / Weak / Incomplete
6. human-review flag: Yes / No, with reason if yes

Suggested scoring mechanics

• Score each rubric criterion on a 1 to 4 scale.
• Use artifact-specific weights from Section 7.
• Convert rubric totals into the assignment's weighted course percentage.

Score meaning

• 4 - Strong: Specific, evidence-backed, and structurally sound reasoning
• 3 - Competent: Mostly complete and credible, with some thin areas
• 2 - Partial: Some correct elements, but important reasoning gaps remain
• 1 - Weak: Missing, generic, confused, or unsupported

7. Rubrics, Scoring Criteria, and Evaluator Prompt Guidance

A. Case Study Analysis Rubric

B. Cross-Case Matrix + Personal Map Rubric

C. Scenario Room Participation and Group Artifact Rubric

D. Final Reflection Rubric

Course score conversion guidance

• 90-100: Strong command of ambiguous decision-making
• 75-89: Credible and mostly complete reasoning
• 60-74: Partial mastery; needs stronger handling of uncertainty and tradeoffs
• Below 60: Insufficient evidence of the course method

Suggested evaluator prompt for the LLM

Use this prompt with the relevant submission pasted after it:

You are evaluating student work for Course 03, "The 'No Playbook' Case Studies."

Your job is to assess the student's reasoning under ambiguity based only on visible evidence in the submission.

Do not reward confidence, polish, or admiration for famous founders unless the student shows disciplined analysis.
Reward:
- clear decision-point framing
- distinction between knowns, unknowns, and unknowables
- explicit tradeoffs
- resistance to hindsight bias
- use of the Four Constants
- concrete next actions or update triggers where relevant

Penalize:
- generic storytelling
- founder worship or reflexive moral condemnation
- judging decisions only by outcomes
- vague claims like "they took a big risk" without mechanism
- motivational language that replaces analysis

Return your evaluation in this order:
1. Completion check against the required components
2. Criterion-by-criterion scores from 1 to 4
3. Short evidence-based justification for each score
4. Two strengths
5. Two revision moves
6. Overall level: Strong, Competent, Weak, or Incomplete
7. Human-review flag: Yes or No, with reason if Yes

Important constraints:
- Do not invent missing evidence.
- If the student does not separate knowns from unknowns, cap reasoning quality at 3.
- If the student evaluates a decision mainly by outcome, cap decision judgment at 2.
- If the personal artifact lacks a concrete next action, cap actionability at 2.
- Minor writing issues should not heavily affect the score unless they block meaning.
```

## Optional structured output schema for LLM evaluators

```json
{
  "assignment_type": "",
  "completion_check": {
    "complete": true,
    "missing_components": []
  },
  "scores": {
    "criterion_1": 0,
    "criterion_2": 0,
    "criterion_3": 0,
    "criterion_4": 0,
    "criterion_5": 0
  },
  "weighted_total": 0,
  "strengths": [
    "",
    ""
  ],
  "revision_moves": [
    "",
    ""
  ],
  "overall_level": "",
  "human_review_flag": {
    "flagged": false,
    "reason": ""
  }
}
```

## Additional evaluator heuristics

- If the student never names what would cause them to change course, treat the work as weak on adaptive reasoning.
- If the student treats uncertainty as an excuse for hand-waving, score down on rigor.
- If the student includes a rejected option with a clear reason, score up on judgment.
- If the student recognizes that some information cannot be obtained in time, score up on realism.
- If the student confuses disagreement with ambiguity, score down on conceptual precision.

---

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

## Global feedback posture

The LLM should be direct, sober, and developmental. It should not flatter students with vague praise. It should identify:
- what evidence is working
- what reasoning move is missing
- what revision would materially improve the submission

## Strong responses usually look like this

- They reconstruct the decision environment rather than retell the whole story.
- They distinguish knowns, unknowns, and unknowables with precision.
- They analyze at least one real alternative rather than pretending only one path existed.
- They acknowledge fear, downside, or institutional tension.
- They make a recommendation or personal next move without pretending certainty.

**How the LLM should respond to strong work:**
- name the exact reasoning move that made the work strong
- reinforce discipline, not just insight
- push for one layer deeper, usually second-order effects or update logic

**Example LLM response pattern:**
"Your strongest move was separating what the actors knew from what later observers know. That kept your analysis from collapsing into hindsight. Your tradeoff section is also strong because you treated capital, morale, and credibility as competing constraints rather than a single 'risk' bucket. To improve further, add one explicit signal that would have told the team their current path was no longer rational."

## Average responses usually look like this

- They include the right sections but some are thin.
- They summarize the case better than they analyze it.
- They identify uncertainty but do not classify it carefully.
- They have a personal application, but the next move is under-specified.

**How the LLM should respond to average work:**
- identify the strongest existing element
- name the specific missing layer of reasoning
- ask for one concrete revision that upgrades the work

**Example LLM response pattern:**
"You clearly identified that this was a no-playbook situation and you included the main case facts. The weakest part is the treatment of uncertainty: several items listed as 'unknown' were actually unknowable in the moment, and that distinction matters. Revise by separating those categories and adding one rejected option with a clear reason it was not chosen."

## Weak responses usually look like this

- They rely on slogans such as "take bold risks" or "trust the vision."
- They judge famous decisions mainly by whether they worked.
- They never identify a serious alternative.
- They confuse stress with ambiguity.
- Their personal reflection is abstract, sanitized, or non-actionable.

**How the LLM should respond to weak work:**
- state plainly that the work is too generic to score highly
- identify the exact missing evidence
- give a narrow rescue plan with 2 or 3 concrete fixes

**Example LLM response pattern:**
"This submission does not yet show strong reasoning under ambiguity. It names a famous case and a broad lesson, but it does not identify a specific decision point, real tradeoffs, or what was unknown at the time. Start by choosing one exact moment, listing three knowns and three unknowns, and explaining one alternative the actors could have taken."

## Artifact-specific feedback guidance

### Case Study Analysis

**Strong:** Specific decision point, disciplined uncertainty map, clear tradeoffs, low hindsight bias.

**Average:** Good summary and some analysis, but tradeoffs or alternatives are thin.

**Weak:** Mostly biography, opinion, or outcome-based judgment.

**LLM response move:** Ask, "What made this decision hard in real time?" and "What plausible alternative did you take seriously?"

### Cross-Case Matrix + Personal Map

**Strong:** Placements are justified, comparisons are meaningful, and the personal case is genuinely ambiguous.

**Average:** Matrix exists but some placements feel intuitive rather than argued.

**Weak:** Cases are dropped on the chart with little logic; personal case is not truly no-playbook.

**LLM response move:** Ask the student to defend one matrix placement and one personal unknown more precisely.

### Scenario Room Artifact

**Strong:** The team shows structure, options, uncertainty, and update logic.

**Average:** Reasonable recommendation, but the analysis skipped one or two framework steps.

**Weak:** Fast opinion masquerading as strategy.

**LLM response move:** Ask what they ruled out, what they still do not know, and what would force a pivot.

### Final Reflection

**Strong:** Honest, concrete, and actionable; the student shows tension and judgment.

**Average:** Relevant but somewhat abstract; the framework is present but not fully operational.

**Weak:** Generic self-help essay with little real decision logic.

**LLM response move:** Ask for one actual next move inside 14 days and one condition that would change that move.

## Final note for evaluators

The course succeeds when students become more disciplined in situations where precedent is weak. The highest-scoring work should show calm reasoning without false certainty, ambition without mythmaking, and action without pretending the unknown has disappeared.