> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ingestly.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Insights
> See how a workflow performs over time, what it costs per run and per node, and where it fails.
Open **Workflow insights** from the **More actions** menu in the editor, or open the command palette and
type "insights". The page reads the active environment and shows the last 7, 30 or 90 days. When the workflow is open over another one, as a called sub-workflow, Insights opens in a new tab so
neither workflow loses unsaved changes.
## What counts as a run
Insights counts **full runs**, started from the workflow's trigger, and the **calls this workflow received**
when another workflow's Call Workflow step ran it. The child runs a Split or Loop creates, restarts from a
node, and builder test runs are not counted as runs. They still contribute to the node rows and to cost,
because their steps ran on this workflow's behalf.
Dry runs are hidden by default. Turn on **Include dry runs** to add them to the run, node and review
numbers. Cost always includes dry runs, because a dry run bills the same as a real one.
Insights are written when a run finishes and are kept when the run is deleted. Deleting a run removes it
from the runs list and from the **Top errors** list, and nothing else on this page changes; **Failures by
kind** is kept. Runs that finished before this feature shipped are not counted; their credits still are.
Days are grouped in UTC. A run that starts late in your evening may land on the next day's bar.
## Tiles
| Tile | Meaning |
| --------------------- | ---------------------------------------------------------------------------------------------------------------------- |
| **Runs** | Full runs and received calls that started in the window and have finished, with how many are still in flight right now |
| **Success rate** | Completed runs over completed, failed and cancelled runs |
| **Run duration** | The median (p50) and the 95th percentile (p95) of completed runs, from start to finish |
| **Credits** | Credits billed in the window, and the average per run that billed anything |
| **Review turnaround** | The median time from a review task being created to it being approved or rejected |
## Charts
**Run outcomes** stacks completed, failed and cancelled runs per day. **Credits per day** shows what the
workflow billed each day.
## Nodes
The nodes table is the reason this page exists. Every canvas node that ran or billed in the window gets a
row with its executions, failures and failure rate, retries, average and p95 duration, credits and its
share of the workflow's bill. The table is sorted by credits, so the most expensive node is first.
A node's name is recorded when it runs, so a node you have since renamed keeps one row under its old name
until new runs replace it. A node you have deleted shows its id with a **Removed** badge, because its
history is still part of the bill.
A node with a high share of cost and a high failure rate is paying twice: once for the work and once for
the retries. Start there.
## Top errors
The five most common failure messages across failed runs in the window, with how many runs hit each.
## Failures by kind
Every failed or cancelled run is given a category when it ends, so the page can still say why runs
failed after the runs and their error messages are deleted. Runs that failed before this feature shipped
have no category and appear only in the failed count.
Each row shows how many runs ended that way and its share of the failed and cancelled runs in the window.
| Kind | Meaning |
| ------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Timed out** | A node ran past its configured timeout |
| **Provider error** | An AI, OCR, email or storage provider failed and retries did not recover |
| **Connector error** | A connector's authentication or token endpoint rejected the call, or an HTTP action failed |
| **Configuration** | The workflow cannot run as configured: a reference to a missing node, a pruned graph revision, an unrouted Classify or Split group, a bad HTTP action |
| **Document rejected** | The document could not be processed (size, page or image limits) |
| **Out of credits** | The credit hold was denied |
| **Organization inactive** | The organization was suspended |
| **Could not start** | The run could not be started |
| **Child run failed** | A Split, Loop or called run failed and the parent failed with it |
| **Review could not resume** | A review was decided but the run could not resume |
| **Step budget exhausted** | The run hit its step limit, usually a loop that never ends |
| **Run deleted** | The run was deleted while a step was executing |
| **Step failed** | A node failed for a reason outside the categories above |
| **Cancelled by user** | Someone cancelled the run |
| **Cancelled or timed out** | The run was cancelled by the platform or exceeded its overall time limit |
| **Parent cancelled** and **Parent failed** | A child run ended because its parent did |