When the test harness is the bug
Two harness fixes. Hung steps now get force-stopped, and the gap between calls is measured from where the audio really ended.
The overnight loop kept getting stuck, and some playtests reported spacing problems that weren’t really in the game.
Two fixes:
- Force-stop hung steps. A browser step that never returned would stall the whole night. Every step now has a hard timeout and gets killed if it runs over.
- Measure gaps from the real end of a cut call. When a station is shot mid-call, its audio stops early. The harness was measuring the gap to the next call from where the call would have ended, so it flagged silences that didn’t exist. It now measures from where the audio actually stopped.
The lesson: when an automated judge is grading your work, check the judge too.