# When to Stop an AGD Experiment: Budget and Evidence Rules

Canonical: https://theorganiclub.com/en/read/lab/stop-rules/

Published: 2026-09-08T12:00:00Z
Updated: 2026-09-09T07:41:56.167Z
Language: en
Evidence: hypothesis

A stop rule specifies when to end, pause or reconsider an experiment.

![A rusted railway buffer and heavy bolts beside a moss-covered wooden sleeper.](https://theorganiclub.com/images/editorial/stop-rules.jpg)

Dietmar Rabich. Wikimedia Commons photograph. Resized and JPEG-compressed from the source photograph. Original source SHA1: b4d1327c14b2b9518d0812daf4f9dc941fcdf629.

## The takeaway

A stop rule specifies when to end, pause or reconsider an experiment. It should account for expected information and operational constraints.

## The argument

Set a duration limit and incidents requiring a pause: missing data, wrong assets, changed scope. Stopping as soon as a number looks attractive can select favorable noise. Continuing indefinitely until success appears is no better method.

A pause for an incorrect file should preserve valid observations and identify affected posts. Do not automatically merge results before and after correction. Depending on the change, continue, restart part of the test or accept a limited conclusion. The stop rule protects interpretation as well as budget. Record who made the decision and why so a later reviewer can understand whether the final dataset still answers the original question.

## Do not let the dashboard negotiate the budget

Without a stopping rule, a test can end at the first flattering spike or continue indefinitely in search of one. Both habits distort learning. Set a bounded duration or number of units, a maximum spend and any operational conditions that require an immediate pause. A safety or access failure is different from an ordinary weak result and should have its own response. Make the rule specific enough that an operator can apply it without asking whether the founder still feels optimistic about the concept that morning.

Choose the evaluation method to match the way you inspect the data. Repeatedly checking results and stopping when they look convincing can invalidate methods that assume a fixed analysis point. If you lack a suitable sequential design, use interim checks for operational problems and reserve the outcome judgement for the planned review. At the end, record whether the decision is to stop, revise or run a new experiment with a new question. Continuing under a different creative or budget is a new phase, not a retroactive rescue of the original test.

### Three reasons to stop, three meanings

- **Operational:** Access, rights or publication integrity has failed.

- **Budget:** The authorised exploration limit has been reached.

- **Evaluation:** The planned evidence window is complete.

Editorial model: a way to reason about the process, not measured platform results.

## Story: The idea in three screens.

### Knowing when to stop is part of the test.
A stop rule specifies when to end, pause or reconsider an experiment.

### Stop rules
A stop rule specifies when to end, pause or reconsider an experiment. It should account for expected information and operational constraints.

### What can mislead.
An interrupted experiment need not have a verdict; it may simply be incomplete.

## Story: From concept to practice.

### Define deadline and pause criteria.
Adapt this to your audience, budget and test scope.

### Separate operational stopping from statistical conclusions.
Adapt this to your audience, budget and test scope.

## Put it to work

1. Define deadline and pause criteria.

2. Separate operational stopping from statistical conclusions.

Limit: An interrupted experiment need not have a verdict; it may simply be incomplete.

## Sources

This publication is initiated by TokPortal. AI-assisted editorial production and corrections policy: https://theorganiclub.com/en/about/