hbs-case-writer

name

hbs-case-writer

description

Create Harvard Business School-style case studies on financial market events, products, or investment decisions. Use when the user wants to write a teaching case about a financial event, market phenomenon, investment product, or corporate decision. Triggers on requests like "write a case about X", "create an HBS case on Y", or "develop a teaching case for Z". Outputs immersive, decision-focused narratives with supporting exhibits and teaching notes.

HBS Case Writer

Write Harvard Business School-style teaching cases on financial market events and investment decisions.

Case Structure

Every case follows this architecture:

1. Title & Header — Case title, protagonist name/role, date, institution
2. Opening Hook — Start at the decision moment; protagonist faces a choice
3. Background — Industry context, market conditions, relevant history
4. The Situation — Specific events, data, and stakeholders involved
5. Analysis Framework — The apparatus the reader needs to analyze (not the analysis itself). See Analysis Framework below for the full specification; valuation/decision cases must carry a real framework, not just headline multiples.
6. The Decision — End with open questions; NO resolution

Angle Selection (Confirm Before You Research)

Every topic can become several different cases depending on the angle. Ask the user which angle they want before you research or draft (use AskUserQuestion), and do not default to valuation. The most common framing mistake is to collapse genuinely different case archetypes into valuation-flavored investor decisions. Offer a menu of distinct angles and let the user pick.

Present these standard angles (the Other option is supplied automatically, so do not add one):

Angle	What the case is about	Typical protagonist	Typical framework
Valuation	Price a company, asset, or deal; set a target price or buy/sell/hold/underwrite call	Sell-side analyst, growth investor, banker	DCF build, comps, precedents, a "what must the buyer believe" lever table, sensitivity, triangulation
Investment decision-making	How to act on a signal, position, or allocation, including forecasting and market-efficiency judgments	Portfolio manager, allocator, macro strategist	A decision-relevant apparatus (for example a calibration or efficiency lens, a risk-return tradeoff, portfolio impact), not necessarily a DCF
Financial product analysis	Analyze a novel instrument's mechanics, economics, risks, and users; decide whether to use, launch, underwrite, or regulate it	Product lead, risk officer, structurer, buy-side user	How-it-works mechanics, payoff and P&L decomposition, risk scenarios, user and market fit
Industry / market background	A primer on industry structure, competitive dynamics, or a market event as decision context	Strategist, analyst, new entrant	Structure and conduct, value chain, competitive forces, where value accrues
Corporate strategy & capital allocation	A company facing a strategic decision (raise capital, M&A, enter or exit a market, restructure, allocate across projects)	CEO, CFO, head of corp dev, board	Strategic options matrix, capital-allocation hierarchy, deal math, scenario analysis
Risk & crisis management	A firm under pressure making decisions in a run, blowup, or liquidity event	CEO, CRO, treasurer, risk officer	Exposure mapping, liquidity and run dynamics, stress scenarios, deciding under uncertainty

Why this matters (worked example). For Polymarket, the headline facts (a company repricing from about $1B to $15B on near-zero revenue) invite a valuation case. But the user may instead want an investment decision-making case on forecasting and market efficiency: whether a portfolio manager should act on prediction-market signals. These are different cases with different protagonists and different frameworks. Ask, do not assume.

The angle dictates the protagonist, the framework, and the exhibits (see Protagonist and Decision Depth and Analysis Framework). Confirm the angle at the very start; confirm output formats later (see Output Formats).

Writing Principles

Narrative Style

• Third-person limited (follow the protagonist)
• Present tense throughout
• Objective, journalistic tone — no judgments or conclusions
• Let facts and data tell the story
• Quote real sources when available

What Makes It a Case (Not an Article)

• Ends with a decision point, never a resolution
• Presents conflicting information or trade-offs
• Reader must analyze, not just absorb
• Avoid teaching explicitly — immerse the reader instead
• Include just enough context; let exhibits carry data

The Protagonist

• Name a specific person or team facing the decision
• Give them a role, organization, and stakes
• Show their thought process through action, not introspection
• They don't have to be heroic — real people with real constraints

Protagonist and Decision Depth

The best cases pair a decision that has genuine analytical structure with a protagonist whose role demands that analysis. Favor:

• A price target with a defensible framework, or an explicit buy / hold / sell / underwrite / allocate call — decisions the reader can recompute.
• Protagonists whose job is the analysis: a sell-side analyst setting a target price, a portfolio manager sizing a position, an allocator deciding inclusion, a banker pricing an offering, a risk officer sizing exposure. These roles force the case to carry a framework (see Analysis Framework).

A retail investor deciding sell-vs-hold after a pop is permissible but is the weakest decision shape: it rarely requires a framework and collapses to a behavioral call. If you use it, justify why (e.g., the case's purpose is market-microstructure or investor-behavior, not valuation) and still equip the reader with a framework to judge the hold. Avoid "should she sell or hold?" as the sole decision when a richer analytical decision is available on the same facts — that usually means the protagonist was chosen for color rather than for stakes.

When in doubt, ask: could a competent reader produce a different defensible answer from the same case? If the decision is so under-structured that any answer is as good as another, raise the decision's analytical stakes.

Data & Exhibits

Required Elements

• Financial data tables (performance, valuations, market data)
• Charts where visual data tells the story better
• Timeline of key events
• Competitive landscape or peer comparisons

Exhibit Standards

• Source all data (Bloomberg, SEC filings, company reports, etc.)
• Use clean, professional formatting
• Exhibit titles should be descriptive
• Number exhibits sequentially

Quantitative Discipline

Numbers are the load-bearing structure of a finance case. Hold them to these rules:

1. Separate offer-price from trading-price valuation. An IPO has (at least) two valuations: the offer price and the first-day/week trading price. State which multiple you are computing and against which share count. Never write "~100x revenue near $2T" without saying whether that is at offer or at market — the two can differ by a third. Show both when the gap is part of the story.
2. Reconcile share count before any market-cap claim. Float, shares outstanding, and fully-diluted differ. Pick one basis, name it, and use it consistently in every market-cap statement; state the basis in the exhibit.
3. Flag unverified intraday prints. A specific intraday high/low cited to no exchange feed or timestamp is a fabrication risk. Source it to the exchange or a data vendor with a timestamp, or soften it to a range and label it "unverified" in the text. Do not present a scraped price as fact.
4. Triangulate valuations for any valuation case. Show at least two independent estimates (fundamental, secondary-market, precedent, prediction-market) beside the offer/trading price. A single multiple in isolation is not a valuation.
5. Attribute peaks and records to the right day. "First-day high," "first-week close," and "all-time intraday" are different things; do not blur them. Cite the day.
6. Show your basis for growth/margin claims. "33% growth" must be traceable to two specific revenue figures with their fiscal years; a "net loss" must say GAAP and sit beside any adjusted/EBITDA figure so the reader sees both sides of the P&L.

Primary-Document Coverage

For any case about a specific company, offering, or transaction, locate and read the primary disclosure document — the S-1 / prospectus for an IPO or debt deal, the 10-K / 20-F for an operating company, the DEF 14A for governance and pay, the merger proxy for M&A, the 8-K for the triggering event. News and sell-side notes are secondary; they orient you but never replace the filing.

From the primary document, enumerate — explicitly, in your research notes — all of the following that exist, not just the ones that fit your first hypothesis:

1. Business segments — every reportable segment and every operating segment named in the business description, with each segment's revenue and (where given) operating income.
2. TAM / market-size breakdown — every addressable-market figure the issuer states, broken out by segment or geography, and the share each represents.
3. Use of proceeds — every named use, in the issuer's own words and order. Note what is not named.
4. Risk factors — the full set, grouped; flag the idiosyncratic ones (founder/key-person dependency, related-party, milestone-gated pay, regulatory, concentration).
5. Key-person / founder compensation and award structure — equity awards, tranche and milestone mechanics, audited grant-date values, and any "considered improbable" auditor notes.
6. Related-party transactions — anything between the issuer and founders, affiliates, or control persons.
7. Share-count basis — float, shares outstanding, and fully-diluted (options, RSUs, convertibles, tranche awards), stated separately.

Then run a disclosure-economics scan: skim the document for term-family frequencies and for what the issuer emphasizes versus what it buries. (Example: count AI-family vs. Mars-family terms in a SpaceX prospectus; note whether the use-of-proceeds names the headline theme or omits it.) The largest gap between the marketing narrative and the disclosure text is usually your sharpest case angle.

Coverage test before you outline: Can you name every material segment, the largest three risk factors, the CEO's pay mechanics, and the share-count basis? If you cannot, you have not read the document closely enough to write the case. A case that omits a material segment because the writer never enumerated segments is a defective case, not a stylistic choice — the reviewer's Pass 5 will flag it as HIGH.

Analysis Framework

The Analysis Framework section (Case Structure item 5) is the spine of the case. It gives the reader the tools to reach a view; it does not reach the view for them. The framework must match the angle the user chose (see Angle Selection): a valuation or buy/sell/hold case carries the build below, while an investment-decision, product-analysis, industry, strategy, or risk/crisis case carries the angle-appropriate apparatus instead (for example a calibration or efficiency lens, a payoff-and-risk decomposition, or an exposure map). Whatever the angle, the framework must let a competent reader recompute a defensible answer. For a valuation, capital-allocation, or buy/sell/hold case, the framework must include, as exhibits or clearly labeled tables, as many of the following as the decision demands:

• A valuation build, not just a headline multiple. A segment-level DCF (revenue path, margin path, terminal value, discount rate) or a defensible proxy (comps-derived range, precedent transactions) — with the inputs shown so the reader can flex them.
• Comps and precedents — peer multiples and precedent-transaction multiples, each sourced.
• A "what must the buyer believe" lever table — hold the framework at base case, solve each lever individually for the market price. This isolates which assumption the price is betting on (e.g., a discount-rate revision inside its evidence band vs. a revenue assumption many multiples beyond the TAM).
• Sensitivity / scenario structure — at minimum a base/bull/bear or a one-variable sensitivity on the load-bearing assumption(s).
• Real-options or contingent-claim framing when the value is dominated by optionalities — milestone-gated businesses, staged investments, abandonment options. Name the option, the gate, and the rough magnitude; cite the method (Myers, McDonald–Siegel, Longstaff–Schwartz) only if the case is methods-oriented.
• Triangulation across independent estimates — fundamental ranges, secondary-market / private-round prices, prediction-market prints, and the offer/trading price, shown together so the reader can see where the market sits relative to fundamental value.

Thin vs. adequate. A framework that quotes only a revenue multiple (e.g., "~100x revenue") and a price chart is thin and fails this requirement. An adequate framework lets a reader re-derive a defensible value range and identify the single assumption the price is most sensitive to. For a non-valuation case (e.g., a narrative risk event), the framework may be lighter, but it must still surface the decision-relevant structure (who is exposed, by how much, through what channel).

Figure Generation

Use the plotting script to create professional figures:

python scripts/case_plots.py

Available Plot Types

Function	Use Case	Example
plot_time_series	Trends over time	Stock prices, deposits, rates
plot_bar_chart	Comparisons	Revenue by segment, market share
plot_stacked_bar	Composition	Asset allocation, cost breakdown
plot_pie_chart	Proportions	Deposit composition, market share
plot_dual_axis	Two metrics	Price and volume, rates and deposits
plot_comparison_bars	Peer comparison	SVB vs competitors

Figure Style

All figures use professional HBS-style formatting:

• Clean white background with subtle grid
• Bold titles and axis labels
• Professional color palette
• High DPI (300) for print quality
• No top/right spines

Creating Figures

When writing a case, generate figures by:

1. Identify key data — What numbers tell the story?
2. Choose plot type — Time series for trends, bars for comparisons, pie for proportions
3. Run the script — Modify case_plots.py with your data
4. Save to case folder — Output to {case_folder}/figures/

Example: SVB Case Figures

# Figure 1: Deposit Growth
plot_time_series(
    dates=dates,
    values=deposits,
    title='SVB Total Deposits (2018-2022)',
    ylabel='Deposits ($ Billions)',
    filename='exhibit_deposits.png',
    show_values=True
)

# Figure 2: Uninsured Deposits
plot_pie_chart(
    labels=['Uninsured (>$250K)', 'Insured (<$250K)'],
    values=[94, 6],
    title='SVB Deposit Insurance Status',
    filename='exhibit_uninsured.png',
    colors=['#d62728', '#2ca02c']
)

Financial Data Collection

When building a case, gather:

1. Market Data — Prices, volumes, volatility, correlations (Yahoo Finance, FRED, Bloomberg)
2. Company Data — Financial statements, SEC filings, earnings calls, investor presentations
3. News & Analysis — WSJ, FT, Bloomberg, CNBC coverage of the event
4. Academic Context — Relevant research papers on the topic
5. Regulatory Context — SEC actions, Fed statements, regulatory changes

Use web search tools to find current data. Document all sources for exhibits.

Output Formats

The case is authored in Markdown (.md) as the single source of truth, then rendered into the deliverable formats the user requests. Confirm two things with the user, in this order: (1) the angle (Angle Selection, asked first), and (2) the output formats (asked before generating; do not assume).

1. Word Document (primary, default)

Convert the authored .md to .docx using the md-to-docx skill (or docx skill for richer layout). This is the standard HBS case deliverable.

• Keep the same case folder as the .md (e.g. Case bank/SVB Collapse/SVB_Case.docx)
• Exhibits/figures embed inline after their referencing text
• Optionally also export a .pdf via pandoc

2. Beamer Teaching Slides (optional)

On request, generate a companion LaTeX Beamer deck from the same material as the case. The deck is for leading the class discussion, not a re-telling — it surfaces the hook, key exhibits, and the open decision questions.

• Start from slide-template.tex (Metropolis style, compiles with xelatex)
• One slide per major case section; one slide per key exhibit
• Close with the decision questions and discussion prompts from the teaching note
• Save as {case_folder}/{Case_Name}_Slides.tex and compile to PDF
• If the user has a preferred Beamer style (e.g. the beamer-slides-teaching skill), match it instead

When to offer: After the Word document is drafted, ask: "Would you also like a Beamer teaching deck from this material?" Only build it if yes — do not auto-generate.

Output Decision Flow

Case drafted (.md) + figures
        │
        ├── Word (.docx) ─── default, always offer
        ├── PDF (.pdf)  ─── optional, via pandoc
        └── Beamer (.tex → .pdf) ─── optional, ask first

Teaching Note (Optional)

Create when the case is complete:

• Learning objectives (2-4 specific takeaways)
• Discussion questions (3-5 questions to guide class)
• Suggested teaching plan (how to structure the discussion)
• Key analysis points (what students should identify)

Workflow

When given a financial event/product:

0. Angle Selection (ask first): Before researching, ask the user which angle to pursue, using the taxonomy in Angle Selection. Do not default to valuation, and do not collapse the options into valuation-flavored investor decisions. The angle dictates the protagonist, framework, and exhibits. Confirm the output formats at the same time or immediately after (see Output Formats).

1. Research Phase

   1. Orient with news — Understand the narrative, the players, the key dates. (Secondary sources only; do not anchor here.)
   2. Anchor on the primary document — Read the S-1 / 10-K / proxy / 8-K and run the Primary-Document Coverage enumeration (segments, TAM, use of proceeds, risk factors, comp/award structure, related parties, share-count basis).
   3. Hunt for a non-obvious angle — Before settling on a protagonist and decision, search the primary document for a defensible angle that is not the headline narrative. Ask:
      • Disclosure-economics irony: where does the issuer's emphasis diverge from its actual capital allocation?
      • Repricing-of-risk framing: is the market debate really about cash flows, or about the price of risk / the discount rate?
      • Optionality: is most of the value in contingent, milestone-gated upside rather than the base business?
      • Flow vs. fundamental: is the price being set by mechanical flows (index inclusion, seasoning rules) rather than by views?
   4. Choose the protagonist and decision to fit the angle: Honor the angle the user confirmed in Phase 0. Pick a role whose job demands the analysis the angle requires (see Protagonist and Decision Depth). The angle should dictate the protagonist, not the other way around.
   5. Gather market data and academic context — Prices, comps, precedents, and any rigorous secondary analysis (working papers, valuation decks). A rigorous external analysis is a source to engage with, not a template to copy — use it to pressure-test your own framework and as a coverage checklist (what segments/frameworks does it surface that your case must at least acknowledge?).

2. Outline Phase

   • Map the narrative arc
   • Identify required exhibits
   • Determine the decision point

3. Writing Phase

   • Draft sections in order (hook → background → situation → decision)
   • Create exhibits with sourced data
   • Write teaching note if requested

4. Output Phase

   • Confirm with the user which formats they want (Word default; Beamer optional)
   • Render the .md to the chosen formats (see Output Formats)

5. Review & Revise Phase

   • Hand the drafted case to the case-reviewer agent (see Case Review & Revision below)
   • The reviewer critiques the case, verifies facts and references, and returns a scored report
   • Apply the revisions, then re-render the affected outputs
   • One review-revise round by default; loop again if the user requests or the score is below threshold

Case Review & Revision

Every drafted case goes through a review pass before it is considered final. This follows the project's worker-critic separation of powers: the case-reviewer agent only critiques and verifies — it never rewrites. The HBS Case Writer applies the fixes.

The Reviewer's Job

The case-reviewer agent (.claude/agents/case-reviewer.md) produces a scored report against the rubric in case-review-rubric.md. It checks:

• HBS style fidelity — narrative voice, unresolved decision, no hindsight, exhibits carry data
• Fact verification — spot-checks numbers, dates, names, and quotes against primary/credible sources via web search
• Reference & source integrity — every exhibit sourced, sources resolve, no fabricated citations
• Pedagogical soundness — clear decision point, conflicting information present, teaching note aligned
• Decision-relevant coverage — every material segment from the primary filing appears; valuation cases carry a real framework (build/comps/lever/sensitivity), not just headline multiples; use-of-proceeds, TAM, key-person pay, and share-count basis are faithfully represented

The Revise Loop

Drafted case (.md + exhibits)
        │
        ▼
  case-reviewer → scored report (0-100) + issue list with severity
        │
        ├── score >= 85 and no HIGH issues → done
        │
        └── otherwise → HBS Case Writer applies fixes
                │
                ▼
          (re-render outputs) → optional second review round

• Score starts at 100; deductions per the rubric
• HIGH severity issues (fabricated facts, missing sources, broken decision point) block delivery
• Max 2 review-revise rounds before surfacing remaining issues to the user
• The reviewer's findings are saved alongside the case as {Case_Name}_review.md

References

• hbs-patterns.md — Detailed patterns from real HBS cases
• financial-data-sources.md — Where to find financial data
• case-review-rubric.md — Scoring rubric used by the case-reviewer agent
• slide-template.tex — Beamer teaching deck template
• case-reviewer agent (.claude/agents/case-reviewer.md) — Reviews and verifies a drafted case