Observed effects

For each run step, the runtime records what the script process did. It does not change the script. The run record keeps observed_effects for each step:

Field	What the process did
`network`	Hosts that it connected to.
`reads`	Files that it read.
`writes`	Files that it wrote.
`env`	Environment variables that it read.
`runs`	Programs that it started.

Paths inside the run’s own temporary and output folders are left out. Python scripts are observed with an audit hook, and Node scripts with a preload. Other runtimes are not observed: the record says runtime_not_observed. Observation never fails a step; a problem gives unavailable with the reason.

On the dashboard

Each script step on the run page shows what the script did. When the step has a script card, the page compares the two and shows what the script did beyond its card: for example a host, a file, or a variable that code analysis did not find.