Test demand
Route each Hypothesis to the cheapest test that can settle it, and decide in advance what result counts.
Before a Startup builds, it tests whether anyone pays. Each Hypothesis goes to the cheapest test that can settle it, with pass and kill lines fixed before the first Visitor arrives.
The test ladder
Tests form a ladder from free to costly. Each rung has a cost, a time to verdict, and the strongest Evidence grade it can yield. A Hypothesis goes to the lowest rung that can yield the grade its threshold needs.
| Rung | What it is | Spend | Days to verdict | Strongest grade |
|---|---|---|---|---|
| Portfolio prior | what sibling Startups at the same template and state learned | $0 | 0 | behavior, from siblings |
| Replay | the candidate scored against the kept record | $0 | 1 | opinion |
| Problem interview | the last time the Problem happened, and what it cost | $0 | 3 to 7 | behavior |
| Smoke page | a Page for the Offer, counting who tries to accept it | $0 to $100 | 3 to 7 | behavior |
| Fake door | an Offer that is not open, counting Waitlist joins | $0 | 7 | behavior |
| Pre-sale | a reserve page taking refundable deposits | $0 | 14 | commitment |
| Concierge | the outcome delivered by hand, openly, for money | hours | 14 to 30 | behavior |
| Manual back end | an automated front with people or agents doing the work behind it | hours | 14 to 30 | behavior |
| Paid probe | $100 to $1,000 of Ads on one message | $100 to $1,000 | 3 to 7 | behavior |
| Single capability | the one feature that carries the promise, built and sold | build time | 14 or more | behavior |
| Flag split | variants of the built thing, randomized per Visitor | traffic | 14 to 60 | behavior |
The ladder settles most early questions in days. The customer and Problem questions go to interviews. Whether people switch goes to a pre-sale against today's workaround, and which promise wins goes to a smoke page or a paid probe.
Pass and kill lines
Every test fixes five numbers before it starts: the primary Metric, a pass line, a kill line, a minimum sample, and a window. Fixing them first is what makes the result Evidence instead of a story.
| Test | Primary Metric | Pass | Kill | Minimum sample | Window |
|---|---|---|---|---|---|
| Pre-sale | outside deposits or paid pre-orders | 3 or more | 0 | 100 qualified Visitors | 14 days |
| Concierge | outside Customers served by hand and paying | 3, with 1 repeat | no repeat | 3 Customers | 30 days |
| Paid probe | cost per commitment | at or below the first payment's margin | above 3 times that margin | 30 clicks per variant | 7 days |
| Fake door | Waitlist joins per 100 Visitors | 5 or more | under 1 | 200 Visitors | 14 days |
| Problem interviews | interviews that describe a recent, costly instance | 5 of 8 | 2 of 8 or fewer | 8 interviews | 7 days |
These are starting defaults. Each is a Setting that the Studio, a Portfolio, or the Startup can tighten and never loosen.
A result between the lines reads iterate: the learn step drafts the next variant, and the test runs again. A test that misses its minimum sample reads underpowered and pools forward into the next run instead of counting as a failure.
Exploring without an audience
Free exploration works where no audience is needed: search, answer engines, directory and registry listings, community threads, and cold outbound. It also works on borrowed audiences, such as the Studio's shared social accounts, house Ads across sibling Startups, and partners.
A Startup with no audience buys learning speed instead. A $100 to $1,000 paid probe is the fastest exploration instrument, and it counts as exploration, not growth spend. Spend scales only on variants that already won.
A new Startup's default demand budget is $0, so its first tests run on free rungs and borrowed audiences. A founder who funds a budget records it as a Decision with a target and a stop rule. The Smoke test, Reserve drop, and Creator probe Plays package the common rungs.
Message first
The message is the unit of a demand test. Fast Plays such as paid search and cold email return a verdict in days, so they test the message. Only a winning message earns slow Plays, such as SEO and answer engine optimization, which take months to read.
A bookkeeping Agent for independent dental practices tests three promises on $150 of search Ads: hours saved, missed deductions, and audit risk. Missed deductions wins at $11 per Waitlist join, against $34 and $52. That promise becomes the Page headline, and the editorial Plays start from it.
Reading the result
Each test ends in Evidence on its Hypothesis and a Decision. For one change, the choices are ship, kill, or iterate. For the Startup as a whole, they are persevere or pivot.
A pivot forks the Startup. The new definition runs as a sibling under the same Idea, new demand splits between the two, and existing Customers stay where they are. The loser goes Dormant, kept and revivable.
A b2a Dataset API of commercial lease comparables runs a pre-sale to buying Agents and one to analysts. Agents leave 6 deposits in 14 days; analysts leave 1. The Startup pivots to Agents, and the analyst version goes Dormant with its Evidence.
Next
- Problem interview: the first rung that reaches real people.
- Wants: unmet demand, and reserve-first.
- Experimentation: the loop every test runs inside.
- Office hours: where each week's result becomes a Decision.