The Seven Questions Test

SKILL.md

---
name: seven-questions-test
description: "Activate when: user says 'seven questions,' 'Zero to One test,' 'Thiel checklist,' 'should I start this business,' 'is this startup idea any good,' or 'stress-test my startup idea'; a business idea, pivot, product line, or funding decision is about to proceed on enthusiasm rather than evidence; a pitch or plan is strong on one dimension (usually technology or market size) and silent on the others; the user wants a structured go/no-go gate before committing capital, time, or reputation to a venture. Do NOT activate when: the venture is already running and the question is a single-dimension diagnosis (pricing, one channel, one hire) — use the specific skill for that dimension instead; the decision is a small reversible experiment where a full seven-question gate is heavier than the bet itself; the user asked only to polish pitch wording, not to test whether the underlying business should exist. More: deciqai.com/c/seven-questions-test"
---

# The Seven Questions Test

## Overview

Most business failures are not bad luck — they are unexamined answers. In *Zero to One* (2014, ch. 13), Peter Thiel and Blake Masters argue that every business must answer seven questions: **Engineering** (breakthrough technology, not incremental improvement), **Timing** (why now), **Monopoly** (big share of a small market), **People** (the right team), **Distribution** (a way to deliver, not just create), **Durability** (defensible in 10–20 years), and **Secret** (a unique opportunity others don't see). Thiel's calibration is blunt: answer all seven and you can build a monopoly; answer five or six and it might work; fewer, and failure will arrive looking like bad luck — but it wasn't luck, it was skipped questions.

The test's power is not the list — anyone can recite seven words. The power is the **gate**: each question demands a *written, evidence-backed answer scored 1–5*, and a question you cannot answer in writing with evidence scores a 1, not a hopeful 3. Vibes do not pass the gate. The book's own natural experiment makes the case: cleantech companies of the ~2005–2012 bubble routinely skipped four or more questions and died (Solyndra's 2011 bankruptcy is the emblem); Tesla, starting from the same sector in the same years, answered all seven.

This is the capstone skill of the Zero to One set — it operationalizes the others. Compose with monopoly-vs-competition (Q3 in depth) · zero-to-one-secrets (Q7 in depth) · last-mover-advantage (Q6 in depth) · definite-optimism (Q2's worldview) · contrarian-question (the interview form of Q7) · bullseye-traction-channels (Q5 in depth) · checklist (why written gates beat memory).

## When to Use

**Use when:** deciding whether to start, fund, join, or seriously pivot a business; a plan is compelling on one axis (usually the tech or the TAM slide) and vague on the rest; preparing or evaluating a pitch and you want the holes found before an investor finds them; a previous venture failed "because of bad luck" and you want to know which questions were actually skipped; running an AI startup idea through a structured gate before building; user says "seven questions," "Thiel test," "is this idea any good," "stress-test my startup."

**Skip when:** the venture exists and the question is one dimension deep (a pricing change, one channel, one hire) — go straight to the dimension-specific skill; the bet is small and reversible (a weekend prototype, a landing-page test) where the gate costs more than the experiment; the user wants pitch polish, not a verdict on whether the business should exist.

## Coaching Novices (Adaptive Front Door)

Before running the test, read the user. Two delivery modes — pick one, don't default to dumping a finished scorecard.

- **Engine mode (do-it-for-me):** the user brought a concrete business idea, pitch, or plan and wants it tested → run the full Seven-Question Gate directly and concisely.
- **Coach mode (teach-me):** the user gave no concrete venture, or signals unfamiliarity with the framework → guide, don't score at them.

When unsure which they want, ask one line first: *"Want me to score a specific business idea against all seven questions, or walk you through the test one question at a time?"*

In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output that step's question and nothing more.

In coach mode:

1. **One-line what-it-is.** Say what the test buys them, in plain words (≤2 sentences, no jargon): most startups fail not from bad luck but from never answering basic questions — about the tech, the timing, the market, the team, the way to reach customers, the moat, and the insight. This test forces a written, evidence-backed answer to each before you commit.
2. **Check fit.** Match their situation against *Use when* / *Skip when*. If it's a single-dimension question or a tiny reversible bet, say so and point to the narrower skill.
3. **Elicit their real venture.** If they have no concrete case, ask for one: the business idea in one sentence, and what stage it's at. Never run the gate on a hypothetical when a real decision is available.

> **[WAIT — do not advance until user responds]**

4. **One question at a time.** Walk the seven questions one per turn: state the question, ask for *their* written answer with evidence, then score it together 1–5 — wait for their input before advancing to the next question. Do not batch questions; the discipline of answering each one separately is where skipped questions get caught.

> **[WAIT — do not advance until user responds]**

5. **Close by naming the verdict.** End with the tally against Thiel's calibration, the weakest one or two questions by name, and what evidence would raise each score — so they remember which question their venture cannot yet answer, not just the number.

> **[WAIT — do not advance until user responds]**

Then enter the Process below at the depth the chosen mode calls for.

## Process

Run the **Seven-Question Gate** (7 scored questions → one verdict artifact). One question at a time, in order. The unbreakable rule: **a question without a defensible written answer scores 1.** No proceeding on vibes.

1. **State the venture** in one sentence: what it sells, to whom. *Gate: if this sentence can't be written, stop — there is nothing to test yet.*
2. **Q1 Engineering.** Can you create breakthrough technology — an order-of-magnitude (≥10x) improvement on the best existing substitute on a dimension the customer cares about — instead of an incremental one? Write the answer with the comparison baseline and the measured or estimated multiple. Score 1–5. *Gate: written, evidence-backed answer exists → else score 1 and move on.*
3. **Q2 Timing.** Is now the right time to start this particular business? Write what changed recently (technology cost curve, regulation, platform shift, behavior) that makes the idea viable today when it wasn't three years ago — and why it isn't already too late. Score 1–5. *Gate: same rule.*
4. **Q3 Monopoly.** Are you starting with a big share of a small market? Name the small market, its size, and your plausible share of it — not the trillion-dollar TAM sliced to a "conservative 1%." Score 1–5. See monopoly-vs-competition. *Gate: same rule.*
5. **Q4 People.** Do you have the right team? Write why *these specific people* are suited to *this specific problem* — domain history, complementary skills, how the founders met and why they'd endure a decade together. Score 1–5. *Gate: same rule.*
6. **Q5 Distribution.** Do you have a way not just to create your product but to *deliver* it? Name the channel, the unit economics of acquiring one customer through it, and evidence the channel works at your price point. "It will spread by word of mouth" is not an answer. Score 1–5. See bullseye-traction-channels. *Gate: same rule.*
7. **Q6 Durability.** Will your market position be defensible 10 and 20 years from now? Write what stops a well-funded imitator, and what the position looks like after the current technology wave commoditizes. Score 1–5. See last-mover-advantage. *Gate: same rule.*
8. **Q7 Secret.** Have you identified a unique opportunity that others don't see? Write the specific true-but-unpopular belief the business is built on, and why the people who could act on it haven't. Score 1–5. See zero-to-one-secrets and contrarian-question. *Gate: same rule.*
9. **Tally against Thiel's calibration.** Count questions scoring ≥4. **7/7:** monopoly-grade — proceed with conviction. **5–6/7:** might work — proceed only with a named plan and date to fix each weak question. **≤4/7:** do not proceed as designed; the failure ahead will be described as bad luck, but it is visible right now, in writing. *Gate: the verdict must cite the per-question scores — no overall gut number.*
10. **Stop-rule:** if scores are shifting because the *venture definition* keeps changing mid-test, stop, rewrite step 1, and rerun — a moving target passes no gate. If the honest tally is ≤4 and the user wants to proceed anyway, record which questions were skipped and why; the artifact is the future post-mortem, pre-written.

Founders should not be the only scorers of their own venture. Where possible, have someone with no stake score the written answers independently and reconcile the gaps — the size of the founder-vs-skeptic gap on each question is itself diagnostic.

### Scoring anchors

Use the same anchor scale for every question. The written answer determines the score — never the reverse.

| Score | Meaning |
|---|---|
| **1** | No written answer, or an answer that is pure assertion ("we'll figure it out," "obviously"). This is the mandatory score for any unanswered question. |
| **2** | A written answer exists but the evidence contradicts it or the claimed advantage is marginal (e.g., 2x on Q1, "hot sector" on Q2, TAM-slicing on Q3). |
| **3** | A plausible written answer with partial evidence — one number or comparison, untested assumptions still load-bearing. |
| **4** | A specific written answer backed by verifiable numbers, named comparisons, or demonstrated results; one identified weakness with a plan. |
| **5** | The answer is demonstrated, not projected: measured 10x (Q1), a closing window you're already through (Q2), dominant share held today (Q3), channel with proven unit economics (Q5), moat already compounding (Q6). |

**Output template:** `Venture (1 sentence) / Q1–Q7: written answer + evidence + score / Tally vs calibration / Verdict / Weakest questions + what evidence would raise them`

### Method in Action: Cleantech vs Tesla (2005–2013)

The cleantech bubble is the book's controlled experiment. Between roughly 2005 and 2012, venture investors poured billions into solar, biofuel, and battery startups; an MIT Energy Initiative analysis found VCs lost more than half of the ~$25B invested in cleantech between 2006 and 2011 (Gaddy, Sivaram & O'Sullivan, 2016). **Solyndra** — $535M federal loan guarantee in 2009, Chapter 11 bankruptcy filed 31 Aug 2011 — is the emblem. **Tesla**, over the same window (Roadster 2008, DOE loan 2010, Model S 2012), answered all seven; it repaid its $465M DOE ATVM loan in full in May 2013, nine years early — the first US auto company to fully repay one. Same sector, same years, opposite tally, opposite outcome.

| Question | Typical bubble cleantech (Solyndra as emblem) | Tesla (2008–2013) |
|---|---|---|
| Q1 Engineering | Incremental panel design; cylindrical form's cost edge erased when polysilicon prices collapsed — arguably worse than flat panels. **1–2** | Integrated EV powertrain an order of magnitude beyond existing electric offerings. **5** |
| Q2 Timing | Bet on solar economics that were moving *against* them; assumed the subsidy era was permanent. **1–2** | Saw a one-time window of government appetite; secured the $465M DOE loan in 2010 before the window shut. **5** |
| Q3 Monopoly | Small slice of a vast, undifferentiated global panel market against subsidized Chinese manufacturers. **1** | Dominant share of the small high-end EV niche first; expand outward from a base it owned. **5** |
| Q4 People | Suits-and-slideware energy generalists; thin hardware operating history. **2** | Engineering-and-manufacturing team with real hardware pedigree, led by a founder-operator. **5** |
| Q5 Distribution | No channel edge; sold a commodity through the same channels as everyone else. **1–2** | Company-owned stores and direct sales instead of dealer networks — a distribution asset competitors couldn't copy quickly. **5** |
| Q6 Durability | Nothing stops a lower-cost imitator; position gone the moment subsidies or silicon prices move. **1** | Compounding brand-plus-technology lead; each generation extends the gap. **4–5** |
| Q7 Secret | Consensus belief ("clean energy is the future") shared by every competitor and every government. **1** | People would buy an electric car because it was *cool*, not because it was green. **5** |

Solyndra-pattern companies skipped four or more questions and their failures were narrated afterward as bad luck and bad subsidies; the skipped questions were visible, in writing, years earlier — for anyone who had written them down.

### 2023–2026 lens: running an AI startup through the seven

Run a typical AI-era idea ("GPT-wrapper for X") through the gate and the failure pattern inverts cleantech's: **Engineering usually passes** — frontier models genuinely deliver 10x on many tasks — but the 10x belongs to the model provider, not the startup. The questions that fail are **Distribution** (Q5: incumbents like Microsoft, Google, and Salesforce ship the same capability inside software the customer already pays for, at near-zero acquisition cost) and **Durability** (Q6: the next base-model release absorbs the feature; a thin prompt-and-UI layer is defensible for months, not decades). An AI idea that passes the gate needs a written answer for proprietary data or workflow lock-in (Q6), an owned channel (Q5), and a secret that is *not* "LLMs are powerful" — which, since 2023, everyone sees.

The AI-era pattern, question by question:

- **Q1 Engineering:** often a genuine pass — but verify the 10x belongs to *you*, not to the base model every competitor also rents.
- **Q2 Timing:** "why now" is easy (capability just arrived); "why not too late" is the hard half — hundreds of teams saw the same capability the same week.
- **Q3 Monopoly:** "AI for [industry]" is a category, not a small market. Name the specific niche and wedge.
- **Q4 People:** domain depth in the *customer's* problem now matters more than ML credentials — the model layer is bought, not built.
- **Q5 Distribution:** the usual killer. Incumbents bundle the same capability into software the customer already pays for, at near-zero acquisition cost.
- **Q6 Durability:** the second killer. Ask: "what survives the next base-model release?" If the answer is only prompts and UI, score it honestly.
- **Q7 Secret:** must be a non-consensus truth about the *customer or workflow*, not about the technology — everyone already believes in the technology.

## Applying It Well

1. **Answer in order, one at a time.** Batch-answering lets the strong questions lend unearned confidence to the weak ones. The discipline of stopping at each question is where skipped questions get caught.
2. **Write before you score.** The score is a compression of the written answer, never a substitute for it. If you catch yourself picking a number first and drafting a justification second, restart the question.
3. **Compare against the best substitute, not the worst.** Q1's 10x is measured against the strongest thing the customer could do instead — including doing nothing, which is often the true incumbent.
4. **Treat "might work" (5–6/7) as a work order, not a pass.** The output of a 5–6 tally is a named plan and date for each weak question — see checklist for making that plan survive contact with a busy quarter.
5. **Rerun the gate at each major commitment.** Answers decay: a Timing answer from 18 months ago, a Durability answer from before the last platform release. The test gates the *next* irreversible commitment, not just the first one.
6. **Keep the artifact.** A dated scorecard is the cheapest post-mortem insurance available — when the venture struggles, you can distinguish "the world changed" from "the question was skipped."

## Common Rationalizations

**Note — [D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**

| Fake move | Reality |
|---|---|
| [D] "Our tech is 2x better, which is basically 10x" | It is basically 2x. Thiel's threshold is order-of-magnitude precisely because 2–3x gets eaten by switching costs, incumbency, and marketing noise. Write the actual multiple against the *best existing substitute*, not the worst. If it's 2x, Q1 scores 2, and the plan must win elsewhere. |
| [D] "Distribution will follow a great product" | The book's core warning: if you can make it but not deliver it, you have a bad business no matter how good the product. "Build it and they will come" is Q5 scored 1 while claiming 5. Name the channel and its unit economics or take the 1. |
| [D] Scoring a question 3 "to be safe" when there's no written answer | The rule exists to kill exactly this. No defensible written answer = 1. A wall of 3s is not a moderate venture; it's an untested one wearing a moderate costume. |
| [D] "Our TAM is $800B, so even 1% is huge" | That is the opposite of Q3. Thiel's question is a *big share of a small market* — a 1% sliver of a giant market means brutal competition and no pricing power. Name the small market you can dominate first. |
| [D] "We're smart and hardworking, so People is a 5" | Q4 asks for problem-founder fit, not IQ. Smart generalists chasing a domain they learned last quarter is how cleantech burned money. Write the specific history that makes this team unusually right for this problem. |
| [D] "Everyone's doing AI now, so Timing is obviously a 5" | Everyone doing it is evidence *against* you on Q3 (crowded) and Q7 (no secret), and neutral on Q2. "Why now" must explain why *you* have a window, not why the sector is hot. Bubbles are when the most questions get skipped — that is the cleantech lesson verbatim. |
| [D] "We answered all seven in the pitch deck" | A slide bullet is not an evidence-backed answer. The gate requires written answers with baselines, numbers, and sources — the deck asserts; the gate verifies. Re-run each answer through "what's the evidence?" before scoring. |
| [D] "We'll fix Distribution/Durability after product-market fit" | Deferral is a score of 1 with a promissory note attached. Tally it as 1 now; if the tally still clears 5/7, fine — but don't let a deferred question masquerade as an answered one. |
| [D] "We ran the test and scored ourselves 6/7" | Self-scored gates inflate. Have a no-stake skeptic score the same written answers independently; where founder and skeptic diverge by ≥2 points, the founder's evidence — not confidence — settles it. |
| [D] "The test is for investors; we're bootstrapping" | The gate protects whoever pays — and bootstrappers pay with years of their own life instead of someone else's fund. Fewer external checkpoints means the written gate matters *more*, not less. |
| *To add [O] entries: paste a real failure instance here after each production use* | *Description of what happened* |

## Red Flags

- No written artifact — the seven questions were "discussed" in conversation and a verdict announced from memory
- Every score is 4 or 5 — either a once-in-a-decade company or (almost always) the scorer graded their own enthusiasm
- A question scored ≥3 whose written answer contains no number, no source, and no named comparison
- The Engineering answer compares against the *worst* incumbent or against "doing nothing" instead of the best existing substitute
- The Monopoly answer leads with total market size instead of the small market's name and your share of it
- The Secret answer is something widely believed ("AI will change everything," "people want convenience") — a consensus view is not a secret
- The tally is ≤4 but the recommendation is "proceed" with no record of which questions were knowingly skipped
- The venture description mutated between Q1 and Q7 so that each question was answered about a slightly different company
- Scores changed after the tally was computed — the verdict is being reverse-engineered to justify a decision already made
- The gate was run once at founding and never rerun — Timing and Durability answers are being treated as permanent when they decay fastest
- For an AI venture: the Engineering answer describes the base model's capability, and Distribution and Durability were the two shortest answers on the page

## Verification

- [ ] The venture is stated in one sentence before any question was scored
- [ ] All seven questions have a *written* answer — none answered verbally or skipped
- [ ] Every score ≥3 is backed by at least one number, source, or named comparison in its written answer
- [ ] Every question lacking a defensible written answer was scored 1 — no charitable 3s
- [ ] Questions were answered one at a time, in order, against the same unchanged venture definition
- [ ] The tally is computed and explicitly matched against Thiel's calibration (7 → monopoly; 5–6 → might work; ≤4 → don't proceed as designed)
- [ ] The verdict names the weakest questions and the specific evidence that would raise each score
- [ ] Q1's comparison baseline is the best existing substitute (including "do nothing"), not the weakest incumbent
- [ ] The scorecard is dated, so a future rerun can show which answers decayed
- [ ] If proceeding despite a low tally, the skipped questions are recorded in the artifact as the pre-written post-mortem

## Sources

- Thiel, Peter, with Blake Masters. *Zero to One: Notes on Startups, or How to Build the Future.* Crown Business, 2014 — ch. 13, "Seeing Green" (the seven questions, the cleantech post-mortem, the Tesla scorecard); ch. 3 (monopoly), ch. 8 (secrets), ch. 11 (distribution) for the underlying questions.
- Gaddy, Benjamin; Sivaram, Varun; O'Sullivan, Francis. "Venture Capital and Cleantech: The Wrong Model for Clean Energy Innovation." MIT Energy Initiative Working Paper (2016) — documents VC losses of over half of ~$25B invested in cleantech 2006–2011, corroborating the bubble's scale and failure rate.
- U.S. Department of Energy, Loan Programs Office — Tesla received a $465M ATVM loan (2010) and repaid it in full with interest in May 2013, nine years ahead of schedule; Solyndra received a $535M loan guarantee (2009) and filed for Chapter 11 bankruptcy on 31 Aug 2011.

---

*Part of **deciqAI Knowledge Skills** — 233 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/c/seven-questions-test** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*

*Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/seven-questions-test.json*