> ## Documentation Index
> Fetch the complete documentation index at: https://docs.keloa.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Publish & test a flow

> Safely ship a flow — dry-run, publish, monitor, roll back.

A flow can sit as a **draft** forever — it won't affect a real conversation until you publish. Always test before you publish.

## Dry-run

In the flow editor, click **Test run**. You get a simulation panel:

1. Pick a **trigger event** (e.g. `message_received` on WhatsApp).
2. Fill in simulated message text, tags, contact fields.
3. Click **Run**.

The canvas lights up each node as it executes. You see:

* which conditions matched
* what message would have been sent (not actually sent)
* which assignee would have been chosen
* any tags / actions that would have fired

Iterate until it matches your expectation.

## Validation

When you click **Publish**, Keloa validates the flow and blocks publishing if:

* A node is missing a required config (e.g. Message with empty template).
* A Condition has an unreachable branch.
* A WhatsApp Message is outside the 24-hour window and isn't a template.
* The flow would loop indefinitely.

Fix the red nodes it highlights, then publish.

## Publishing

Click **Publish**. The flow:

* Gets a version stamp.
* Becomes live immediately — next matching trigger runs it.
* Previous version is archived (not deleted).

## Monitor

Open the flow → **Runs** tab for a live log:

* Every trigger firing (timestamp, conversation link)
* Which branch was taken at each condition
* What nodes ran
* Any error messages

You can filter by status (success / partial / failed) and date.

## Rollback

Go to **Versions** → pick a previous published version → **Restore**. It becomes live immediately. The broken version stays archived in case you want to compare or patch.

## Pausing

Toggle the flow's status to **paused** to stop all runs without deleting. Useful during incidents or while redesigning.

## Best practices

* **Ship small, iterate.** One flow, one purpose. Don't build a mega-flow with 30 nodes.
* **Tag from the flow**, so you can measure its impact later in analytics.
* **Dry-run three scenarios**: the happy path, an edge case, and a "nothing matches" case.
* **Review runs weekly** for the first month. Most surprises show up in the first 100 runs.