State-feedback vigilance
Correction effort that rises when observed error rises, and relaxes when it subsides — oversight as a thermostat, not a setting.
What it changes
Who can pull it
What it looks like institutionally
Fixed oversight is tuned for an average day; deployed systems have bad weeks. Vigilance escalation ties correction intensity to what monitoring actually observes: more sampling, more mandatory checks, more second reads when error or adoption signals climb — automatically, by rule, not by someone noticing and scheduling a meeting.
The controller framing matters: escalation and de-escalation are both specified in advance, so vigilance doesn't ratchet into permanent maximum-cost oversight or decay into none.
This is the pattern that turns monitoring from reporting into governance: a dashboard nobody is required to act on is journalism, not control.
Addresses: Static oversight vs dynamic risk · Slow institutional reaction. Test a version of this lever in the PAN Lab.