Execution & Logging Guide

Execution visibility is where users decide whether they trust a workflow platform. NeuronFlow is designed to show what happened, where it happened, and what to fix next.

Execution lifecycle

A workflow execution usually moves through these stages:

1. Queued
2. Running
3. Completed, Failed, Paused, or Waiting Approval

For agent and multi-branch workflows, several paths may run in parallel before the execution reaches a final state. Paused and Waiting Approval are resumable states, not finished outcomes.

Where to inspect a run

Use the Executions page to:

• filter by status
• inspect recent runs
• spot retries and failures
• review analytics and queue attention

Open the execution modal to inspect:

• node-by-node path
• step status
• input handed to the node
• output returned by the node
• retries and duration
• recovery events, when relevant

Reading statuses correctly

Completed

The execution reached a successful end state.

Failed

The workflow stopped or all meaningful paths failed before a successful finish.

Failed (continued)

A node failed, but Continue on error was enabled, so downstream steps were still allowed to run.

Waiting Approval

The workflow is paused at a human gate and needs an explicit approval action before it can continue. Reviewers can approve, reject, or request changes when that option is enabled.

Paused

The workflow saved its state and can be resumed later. This is different from Waiting Approval: a generic resume is for technical pauses, while approval waits must continue through an approval decision.

Queued

The run has been accepted and is waiting for a worker.

Timeline vs tree view

NeuronFlow uses two useful mental models:

Timeline

Best for linear workflows.

Tree

Best for workflows with branching or multiple parallel paths.

When a workflow execution contains multiple paths, tree view makes it much easier to understand which branch failed and which branch succeeded.

What the step details mean

Each execution step is designed to answer practical questions:

Input Data

What business data reached the node.

Result

What the node actually returned.

Metadata & Context

Helpful runtime details like:

• AI model used
• tool name
• action context
• safety details

Raw View

Use this only when you need a more technical debugging view. Internal-only metadata is filtered so the modal stays focused on user-relevant information.

Retries and idempotency

NeuronFlow can retry transient failures for nodes like:

• HTTP Request
• Tool
• AI Agent
• CMS actions

For write actions, retries are safest when you provide an Idempotency Key.

That prevents accidental duplicate actions such as:

• duplicate payment creation
• duplicate CRM record creation
• duplicate outbound messages

Queue and recovery

When infrastructure issues happen, NeuronFlow surfaces them instead of hiding them.

Admins can see:

• stuck jobs
• dead-letter failures
• orphaned executions
• recovery history

Users can see when an execution needs attention, and the system records recovery events so teams can understand what happened later.

Common debugging patterns

Tool node failed

Check:

• required fields
• selected connection
• expired credentials
• inline credential overrides

AI Agent output looks wrong

Check:

• model selection
• prompt clarity
• missing upstream input
• missing context memory

Switch or branch behaved unexpectedly

Check:

• actual input received by the switch
• string vs object mismatch
• condition formatting

Workflow stayed queued

Check:

• whether your admin reports healthy background processing
• the queue health page if you have access to operations tools
• whether other workflows are also waiting unusually long

Workflow is waiting for approval

Check:

• the approval step output for reviewer links and delivery status
• whether the reviewer should approve, reject, or request changes
• whether request-changes loops have a sensible revision limit

Operating tips

1. Review the Validation panel before you run.
2. Use Verify Setup for tool nodes before testing a live flow.
3. Watch retries and durations, not just final success/failure.
4. If a branch keeps failing, inspect the exact node input before changing prompts or connections.
5. Use version restore when a workflow suddenly becomes unstable after edits.

Trust-building rule

If a workflow platform cannot explain node-to-node behavior clearly, users stop trusting automation quickly. Always treat execution logs as part of the product, not just a debugging extra.