Software Factory fundamentals
Factory performance
Read delivery, cycle-time and cost signals from completed Factory work.Factory performance connects completed tasks with pull request activity, elapsed time and recorded cost. Use the Velocity view to understand how work moves through one workspace and where the process needs attention.
Start with the reporting window
Section titled “Start with the reporting window”Select 14d or 30d before reading the report. Every summary and chart should be interpreted within that period. When enough earlier data exists, use the comparison to understand direction rather than treating one number as a permanent baseline.
A new Factory may have only a few closed tasks. In that case, read the individual task records before drawing a conclusion from the percentage or median. One failed task can dominate a small sample.
Read the four headline metrics together
Section titled “Read the four headline metrics together”| Metric | What it tells you | What it does not tell you |
|---|---|---|
| Tasks closed | How many Factory tasks reached a recorded closed outcome in the period | Whether every result was useful or merged |
| Task waste | The share of closed work recorded without a useful delivery outcome | Why the work failed or whether the task was worth attempting |
| Median cycle time | The middle completion time for tasks closed in the period | How long every task took or where the time accumulated |
| Cost per task | The recorded spend relative to closed Factory work | The value of the change or every charge from connected providers |
Do not optimize one metric in isolation. More closed tasks can hide more wasted work. Lower cost can be a poor trade when it creates retries. Faster execution can still leave tasks waiting for review.
Understand what the Factory delivered
Section titled “Understand what the Factory delivered”The merged pull request chart adds repository context to the task totals. Change the grouping to answer a specific question:
- Who created: Compare pull requests created through SuperPlane with manual repository work.
- Outcome: Separate results by the recorded pull request outcome.
- Intake source: See which source supplied the work that reached delivery.
These views explain the shape of delivery. They are not a contest between people and agents. Manual work may include larger changes, emergency fixes or work outside the Factory’s intended scope.
Move from a metric to the task record
Section titled “Move from a metric to the task record”When a number changes, inspect the tasks behind it. A useful performance review follows this path:
Open the task and review its source, stage history, automation runs, pull request, checks and recorded decisions. The metric points to the pattern. The task record explains what happened.
Compare like with like
Section titled “Compare like with like”Keep the repository, task shape and completion rule consistent when you compare periods. A documentation edit, dependency upgrade and production repair have different validation paths. Combining them can make a healthy workflow look slow or make a risky workflow look efficient.
Use Factory performance to improve selection and process design. Do not use it as an individual productivity score.