Safety, Energy, and System algorithm
Emergency Protocols
One-tap stand-down modes for too-cold, someone-upset, and suspicion situations.
These algorithms keep the product honest: real Home Assistant commands, real errors, real weather or usage data, and safety-first fallbacks whenever comfort or equipment protection matters.
The short version
One-tap stand-down modes for too-cold, someone-upset, and suspicion situations.
What it watches
The chosen protocol and its remaining window.
How it decides
Too cold (30 min) pauses the defender and turns the thermostat off. Someone upset (45 min) and Suspicion quiet (90 min) keep reading the thermostat 24/7 but send no corrective commands until the window ends. Emergency actions bypass the website debounce.
What it changes
Suppresses corrective commands for the protocol window.
Safety boundaries
- Uses the real inputs listed above. It does not invent thermostat, weather, usage, or sensor state.
- Changes only the output listed above. Thermostat-affecting work goes through Home Assistant or returns a real error.
- The global AC Defender rules still apply: the website target remains the floor for cooling commands, the worker keeps refreshing real Home Assistant state 24/7, and comfort/safety rules are not bypassed by decorative timing.
BROTHER MAD, someone-upset, and suspicion stand-downs control thermostat corrections only. They do not detect yelling, contact emergency services, treat pain or heat illness, or make an unsafe interpersonal situation safe.- If there is immediate danger or a medical emergency, leave if possible and call 9-1-1; do not wait for a protocol timer.
Settings
(run from the Controls page)
Where to see it
- Defense page: live card with state, verdict, evidence, and metrics.
- Guide page: generated from the same guard catalog entry.
- Source:
Guards/GuardCatalog.csdescribes this page; the implementation is coordinated byServices/DefenderStateStore.csandServices/AcDefenderService.cs.
Failure modes
If Emergency Protocols cannot obtain one of its required real inputs, it reports a blocked, held, or unavailable result and leaves the background worker’s Home Assistant refresh running. It never fills a missing room reading, audit event, weather sample, usage value, or device state with a simulator value. If a real Home Assistant command is rejected, the user sees the service’s actual error and the article’s surface remains available for recovery.
Security considerations
This feature consumes only the configured Home Assistant entity data, local settings, and the audit context named above. Tokens and credentials stay in the server environment; the static documentation site does not collect analytics, transmit search text, or embed third-party assets. Logs and exports should be reviewed before sharing because real entity names and timestamps can identify a household.
Verification
Verify the shipped behavior at the feature’s live page or endpoint, then run the repository’s documented build and test commands. Confirm the real-input and real-error paths, keyboard access, reduced-motion behavior, and a 390 px viewport without horizontal overflow. Record the exact commit and workflow result when publishing a release; a static screenshot alone is not proof of a live Home Assistant command.
Suggested articles
- Feature briefs — find every documented surface and guard.
- Defender Logic — follow the complete decision cycle and its bypass rules.
- Settings — inspect persisted configuration, language modes, and safety limits.
- Yelling Survival Guide — practical before, during, and after steps.
- Heat, Pain & Survival Facts — sourced hearing, heat, and personal-safety facts.