deciqAI Industry Learning Sprint
Overview
A structured 3-step process (financial reports → expert dialogue → unique view) for building a working industry mental model in approximately one week. The sequence is strict: financials before experts, experts before view formation. Financial reports reveal how an industry actually works stripped of marketing narrative; gross margin, capex pattern, and disclosed risk factors encode economic reality.
Neighbors: probabilistic-thinking (assign confidence intervals before expert conversations) · first-principles (stress-test the view after Step 3) · confirmation-bias (audit Step 3) · non-consensus-thinking (evaluate if the view is truly non-consensus) · narrow-gate-strategy (identify the leverage point for focused entry).
When to Use
Trigger conditions: Entering an industry for the first time (investor, founder, executive, advisor) · Evaluating an acquisition or partnership in an unfamiliar sector · Preparing for a high-stakes expert conversation with limited prep time · Producing an investment thesis or market entry recommendation under time pressure.
When NOT to use: Deep domain expertise already exists · Timeline under 48 hours (mark output as preliminary) · Industry is primarily informal/unregistered (financial reports will be unrepresentative).
Coaching Novices (Adaptive Front Door)
- Engine mode: user has a concrete industry target → run The Process directly.
- Coach mode: user unfamiliar with financial analysis → guide step by step.
In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.
- Reframe the goal: "You need one falsifiable hypothesis that would be contested by an insider — not comprehensive understanding."
- Check fit: confirm this is a new industry entry scenario, not a domain they already know deeply.
- Elicit their real case: "Which industry, and what decision are you trying to make at the end of this sprint?"
[WAIT — do not advance until user responds]
- Run The Process one step at a time with their input: start with financial structure mapping using their named industry.
[WAIT — do not advance until user responds]
- Close by naming the insight: "Your non-consensus view is [X] — here's why it would be contested by an insider."
[WAIT — do not advance until user responds]
The Process
Step 1 — Financial Structure Mapping (Day 1–2). Pull 3–5 years of annual reports for 2–3 leading companies. Do not read analyst commentary first. Extract: Revenue model · Gross margin (>60% = platform economics; <20% = commodity) · **Capex vs. opex split** (determines moat and entry barrier) · **Customer concentration** (>30% from one customer = disclosed systemic risk) · Disclosed risk factors (the most honest document a company publishes — read as a map of industry failure modes). Output: one-page financial structure map.
Step 2 — Expert Dialogue (Day 3–5). Conduct 3–5 conversations using the financial structure as your hypothesis base. Target four categories: operators, investors, ex-employees, regulators. Design each conversation as a hypothesis stress-test: "I noticed [X] in the financials — is that because [Y] or [Z]?" Ask: "What does the financial structure not capture?" Output: 3–5 corrections or confirmations + 2–3 structural insights the financials did not reveal.
Step 3 — Unique View Formation (Day 6–7). Synthesize into one non-consensus hypothesis: specific (name the mechanism), falsifiable (state what would prove it wrong), contested (a domain expert would disagree). Stop-rule: if you cannot state a contested view, you have summarized, not analyzed. Return to expert corrections: "What do experts believe that I saw evidence against in the financials?"
Output Template
| Section | Contents |
|---|---|
| Financial Structure Map | Revenue model, gross margin, capex/opex, customer concentration, top 3 risk factors — each with source |
| Expert Dialogue Corrections | Hypothesis confirmed / corrected, source (name, role, date); plus 3 structural insights not in financials |
| Unique View | Specific falsifiable hypothesis · evidence base (financial finding + expert correction + tension) · falsifier · confidence |
| Known Gaps | What was not covered and what would change the view |
→ Method in Action: Graham's Analysis of Northern Pipeline (1926)
Domain Packs
Pharma / Biotech: Diagnostic: R&D-to-revenue ratio (>25% = pipeline-dependent), gross margin by product line, patent expiry schedule. Best experts: clinical scientists, formulary managers, ex-FDA reviewers. Reject: "Strong pipeline = strong future."
Logistics / Freight: Diagnostic: operating ratio (<85% = healthy), fuel cost sensitivity, top-10 shipper concentration. Best experts: freight brokers, dispatch supervisors, shippers' logistics managers.
Contribution invitation: submit domain packs via the deciqAI repository.
Applying It Well
- Sequence strictly — financials before experts; experts before view formation.
- Read primary documents — annual reports and earnings transcripts, not analyst summaries.
- Design expert conversations as hypothesis tests — specific financial questions yield 10x more signal than "What should I know?"
- Target four expert categories — operators, investors, ex-employees, regulators each have a structurally different view.
- Apply the stop-rule — "Would a domain expert be surprised by this?" If no, revise.
- Document known gaps — prevents overconfidence.
→ Primary sources: references/sources.md
Common Rationalizations
[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.
| Fake move | Reality |
|---|---|
| [D] "I've read five analyst reports." | Five consensus documents give higher-confidence consensus, not independent analysis. |
| [D] "I talked to insiders for hours." | Without a financial hypothesis base, expert talk produces orientation, not stress-testing. |
| [D] "My view is that this is a great industry." | That is the marketing narrative. A view names the structural mechanism most people are wrong about. |
| [D] "I don't know how to read financials." | The sprint requires only four numbers: revenue model, gross margin, capex/opex, customer concentration. |
| [D] "All the experts agree, so the view is right." | Expert consensus is what the sprint is designed to think against. |
| [D] "I need much more research before forming a view." | More research without a view target produces information, not insight. Commit at Day 7. |
| [D] "My unique view might be wrong." | Specify what would falsify it. Being wrong about a falsifiable view beats vaguely right about consensus. |
| → Add [O] entries here after each real use — paste the actual failure pattern | What went wrong and why |
Red Flags
- No view that would be contested by a domain expert — produced a summary, not an analysis.
- Expert conversations conducted before reading financial reports — sequence reversed.
- Financial structure map missing gross margin — the most diagnostic number omitted.
- Sprint took more than two weeks — time pressure is a design feature, not a bug.
- Unique view is not falsifiable — it is an opinion, not a hypothesis.
- Only one expert category consulted — one-dimensional model.
- Sprint conducted entirely from secondary sources — primary documents and expert dialogue skipped.
Verification
- Annual reports for 2–3 leading companies (3+ years) read as primary sources.
- Financial structure map contains all four dimensions: revenue model, gross margin, capex/opex, customer concentration.
- Expert conversations designed as hypothesis stress-tests with specific financial hypotheses as agenda.
- At least 3 distinct expert categories consulted (operators, investors, ex-employees, regulators).
- Unique view is specific, falsifiable, and would be contested by at least one domain expert.
- Stop-rule applied: "Would an insider be surprised by this?" — if no, view was revised.
- Known gaps explicitly documented. Sprint completed within approximately one week.
Part of deciqAI Knowledge Skills — 237 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. See it run → https://www.deciqai.com/s/industry-learning-sprint · Built by deciqAI · github.com/deciqAI · Contributions welcome.
Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/industry-learning-sprint.json