# Research and define

URL: /workflows/lifecycle/research-and-define · Id: `page:workflows/lifecycle/research-and-define`

> How a Startup writes its founding Hypothesis, grades the Evidence behind each slot, and earns its way from Idea to Building.

Research settles the uncertainty that would change the next decision; definition writes the answer where every [Agent](/business/nouns/agent) reads it. Together they carry a [Startup](/business/nouns/startup) from its first [Idea](/business/nouns/idea) to the [Building](/business/lifecycle/states/building) state.

## The founding Hypothesis
A Startup begins as one record: its founding [Hypothesis](/business/nouns/hypothesis). It has five slots: the customer, their [Problem](/business/nouns/problem), the approach, why they would switch, and how it differs from what they do today.

Take a Startup whose Agent chases overdue invoices for freelance designers. Its customer is a solo designer billing $5,000 to $20,000 a month. The Problem: one invoice in five is paid late, and chasing it costs an afternoon and some goodwill.

The approach is an Agent that sends reminders in the designer's own voice and escalates on a schedule. The reason to switch is money days sooner with no awkward emails. Today, designers chase by hand or let it slide.

The lean canvas, the brand story and the jobs-to-be-done fields are views of this one record, never separate documents. Beside it sits one [Goal](/business/nouns/goal): the least result worth running for, such as $4,000 of outside revenue a month within 90 days.

## Grade evidence by behavior
[Evidence](/business/nouns/evidence) has three grades. Behavior is what a person did: exported an aging report, paid a collections service. Commitment is what they gave up: a deposit, a pre-order, an hour on a call with their real invoices.

Opinion is what a person says they would do, and it weighs zero. A designer who calls the idea great has moved nothing; one who forwards three overdue invoices has.

Each slot shows the strongest grade of outside Evidence behind it. Every entry check and weekly decision reads the grades, never the prose.

## Test the riskiest slot first
Each week the Startup ranks its slots by how weak their Evidence is and how much they matter, then tests the top one. The riskiest slot is usually demand, not feasibility. An Agent that writes a polite reminder takes an afternoon to prove; five designers paying $29 this week takes selling.

The test is the cheapest one that can earn the grade the slot needs.

| Test | Earns | Takes |
| --- | --- | --- |
| [Interview](/business/nouns/interview) about the last time it happened | Behavior, as reported | Hours |
| Reserve page with a refundable deposit | Commitment | A day |
| Fake door: a priced button that records the click | Behavior | A day of traffic |
| Concierge: do the work by hand for three [Customers](/business/nouns/customer) | Behavior and revenue | A week |
| Paid probe | Behavior at a known cost | $100 |

[Reserve drop](/workflows/plays/reserve-drop) and [Smoke test](/workflows/plays/smoke-test) run the middle rows as [Plays](/business/nouns/play). A test that cannot earn the grade the slot needs is skipped, however cheap it is.

## Talk to buyers about the past
In the Idea state, the number that matters is [Problem interviews](/agents/founding#problem-interviews): conversations with distinct outside members of the [ICP](/business/nouns/icp). The [Researcher](/agents/product/researcher) books them from a named list and asks about the last late invoice, what the designer did, and what it cost.

[Questions](/business/nouns/question) about the future collect opinion. Asking whether a designer would use it earns a yes from nearly everyone; asking what they did about the last late invoice earns a fact.

Each [Interview](/business/nouns/interview) ends with an ask: a deposit, a pre-order, or an introduction. An Agent that interviews says it is an Agent, and one that writes for a human founder does so under that founder's [Mandate](/business/nouns/mandate).

## Write the answer into the definition
A result that stays in a transcript changes nothing. The answer becomes a [Change](/business/nouns/change) to the founding Hypothesis, and every [Page](/business/nouns/page), [Offer](/business/nouns/offer) and [Interface](/business/nouns/interface) reads it at the next [Version](/business/nouns/version).

Suppose the interviews show designers lose more to unbilled hours than to late invoices. The Startup keeps its customer and changes its Problem. That is a [pivot](/business/lifecycle/pivots), recorded as a [Decision](/business/nouns/decision) with the options, the Evidence and the choice.

## The checks that admit Building
A Startup holds the [Idea state](/business/lifecycle/states/idea) once its Idea has a one-line statement and names who pays for what work. It [advances](/business/nouns/startup/advance) to Building when four more checks pass. It extends a [Startup Template](/business/nouns/startup-template), states its Problem with today's workarounds, names its [Brand](/business/nouns/brand), and declares each [Product](/business/nouns/product) kind.

In Building, the number becomes [Committed ICP members](/agents/founding#committed-icp-members): outside ICP members whose Evidence is commitment-grade or better. Selling starts here, before anything ships.

## Next
[Build and publish](/workflows/lifecycle/build-and-publish) turns the definition into Products and the Interfaces that sell them.