# Experiments

URL: /systems/experiments · Id: `system:experiments`

> Experiments tests Hypotheses by varying copy, Price or behavior through Flags and decides each winner from primary and guardrail Metrics.

[Experiments](/business/nouns/experiment) tests a [Hypothesis](/business/nouns/hypothesis) by splitting traffic across variants of copy, [Offers](/business/nouns/offer) or behavior through [Flags](/business/nouns/flag). Primary and guardrail [Metrics](/business/nouns/metric) pick the winner, and the winner becomes the definition.

## Summary
| Part | This System |
| --- | --- |
| Group | Data and insights |
| Status | Beta: it works, and its Verbs may still change |
| Shelf | Front: offered from a Startup’s first days |

You can ask:

- “Test a $39 price against $29”
- “Try a new headline on half the visitors”
- “Which version won?”
- “Stop the test”

## What you get
| Part | This System |
| --- | --- |
| Starts | On for every Startup at creation, with nothing to install. |
| Built from | [Event streams](/code/platform/event-streams), [Functions](/code/platform/capabilities/functions), [Database](/code/platform/capabilities/database), [Analytics](/systems/data-and-insights/analytics) |

## Nouns it owns
| Noun | Definition |
| --- | --- |
| [Experiment](/business/nouns/experiment) | An Experiment is a test of a Hypothesis that varies copy or behavior through Flags and judges the variants by a primary Metric. |
| [Hypothesis](/business/nouns/hypothesis) | A Hypothesis is a belief about a Startup stated so it can be proved wrong, with the Metric and threshold that settle it. |
| [Variant](/business/nouns/variant) | A Variant is one value an Experiment can assign, such as a revision, a Price on a sibling Offer, or a prompt, with its lineage. |
| [Exposure](/business/nouns/exposure) | An Exposure is the moment one unit meets the Variant its Flag assigned, logged with every option and its propensity. |
| [Readout](/business/nouns/readout) | A Readout is an Experiment’s result per Variant on its Metric, with the sample size, the interval, and how far the number can be trusted. |

## Verbs
| Verb | Object | Approval | Event |
| --- | --- | --- | --- |
| [Experiment.start](/business/nouns/experiment/start) | [Experiment](/business/nouns/experiment) | None | [`Experiment.started`](/business/events/Experiment.started) |
| [Experiment.conclude](/business/nouns/experiment/conclude) | [Experiment](/business/nouns/experiment) | None | [`Experiment.concluded`](/business/events/Experiment.concluded) |
| [Hypothesis.test](/business/nouns/hypothesis/test) | [Hypothesis](/business/nouns/hypothesis) | None | [`Hypothesis.tested`](/business/events/Hypothesis.tested) |
| [Hypothesis.drop](/business/nouns/hypothesis/drop) | [Hypothesis](/business/nouns/hypothesis) | None | [`Hypothesis.dropped`](/business/events/Hypothesis.dropped) |

## Events
| Direction | Event | How |
| --- | --- | --- |
| Emits | [`Experiment.started`](/business/events/Experiment.started) | By [Experiment.start](/business/nouns/experiment/start) |
| Emits | [`Experiment.concluded`](/business/events/Experiment.concluded) | By [Experiment.conclude](/business/nouns/experiment/conclude) |
| Emits | [`Hypothesis.tested`](/business/events/Hypothesis.tested) | By [Hypothesis.test](/business/nouns/hypothesis/test) |
| Emits | [`Hypothesis.dropped`](/business/events/Hypothesis.dropped) | By [Hypothesis.drop](/business/nouns/hypothesis/drop) |

## States
The lifecycle of an [Experiment](/business/nouns/experiment):

| State | Meaning | Moves to |
| --- | --- | --- |
| `draft` (first) | Its variants and Metrics are set, and no traffic is split yet. | `running` by [Experiment.start](/business/nouns/experiment/start) |
| `running` | Traffic is split across its variants while its Metrics are measured. | `concluded` by [Experiment.conclude](/business/nouns/experiment/conclude) |
| `concluded` | It ended with a Decision to ship, kill, or iterate. | None: it ends here. |

Its other Nouns with a lifecycle: [Hypothesis](/business/nouns/hypothesis).

## Who operates it
| Part | This System |
| --- | --- |
| Accountable | [Analyst](/agents/marketing/analyst) |
| Team | [Marketing Team](/agents/marketing) |
| Seat | A role agent holds it until a Human is seated. |
| Autonomy | `supervised` |
| Also calls its Verbs | [Strategist](/agents/founding/strategist) |

## Approvals and limits
None.

## Processes that use it
| Process | How it uses it |
| --- | --- |
| [Pricing](/workflows/processes/pricing) | A step calls [Experiment.start](/business/nouns/experiment/start) |
| [Experimentation](/workflows/processes/experimentation) | A step calls [Experiment.conclude](/business/nouns/experiment/conclude) |
| [Product definition](/workflows/processes/product-definition) | A step calls [Experiment.start](/business/nouns/experiment/start) |

## Rail underneath
None.

## Views and dashboards
| View | Shows | Metrics |
| --- | --- | --- |
| The [Analyst](/agents/marketing/analyst)’s view | The Metrics its accountable Role answers for. | [Metric freshness](/agents/marketing#metric-freshness) |

## Access
What each Access level may call here.

| Access level | May call |
| --- | --- |
| `owner` | Every Verb (23) |
| `admin` | Every Verb (23) |
| `member` | Every Verb (23) |
| `viewer` | Read only (10 of 23) |

## Import and export
None.

## Fees and pricing
None.

## Availability and limits
| Part | This System |
| --- | --- |
| Status | Beta: it works, and its Verbs may still change |

## Compliance notes
Controls it carries:

- [Log completeness](/systems/trust-and-governance/controls/log-completeness)
- [Append-only records](/systems/trust-and-governance/controls/append-only-records)
- [Point-in-time recovery](/systems/trust-and-governance/controls/point-in-time-recovery)
- [Processing definitions](/systems/trust-and-governance/controls/processing-definitions)

## Related Systems
- [Flags](/systems/flags)
- [Product analytics](/systems/product-analytics)

## Example
A [Directory](/products/templates/directory) of AI tools tests two headlines on the [Page](/business/nouns/page) where vendors buy listings. After 1,900 [Visits](/business/nouns/visit), the second headline lifts purchases 22% while refunds, the guardrail, hold flat. The [Analyst](/agents/marketing/analyst) [Agent](/business/nouns/agent) concludes the Experiment, and the losing variant stays on record with its [Evidence](/business/nouns/evidence).