How we validate detection
A stove-safety device is only as good as its detection. This page is our public, evidence-first scorecard — the exact metrics we measure, kept honest. Most are not yet measured: Stovyn is pre-launch and hasn't run its kitchen pilot. We publish that plainly rather than imply proof we don't have.
The safety invariant
AI can raise an alert’s severity, but can never lower it below the deterministic thermal safety floor. If the AI says "safe", times out, returns a malformed response, or is unavailable, a thermal alert still fires.
Enforced in firmware and backend, not just in copy — and covered by automated tests for the “AI says safe / times out / errors / unavailable” cases.
Every alert is explainable
We keep the deterministic rule, the measured evidence, and the AI's contribution distinguishable — never a bare “ALERT” or an AI confidence score dressed up as certainty. An alert reads like:
The numbers we publish
Each metric is tagged with the kind of evidence behind it. We never blend a lab result with a customer-reported label.
Performance by stovetop
Gas, electric coil, glass-top, and induction produce different thermal signatures, so we report each separately rather than as one blended average.
Two weekly tests
Automatic self-check
Runs on the device weekly: battery, thermal-sensor plausibility, memory, buzzer, connectivity, firmware, and notification credentials.
Optional end-to-end drill
A synthetic alert travels device → cloud → push/SMS → your confirmation. It proves the alert chain delivers — it does NOT prove the thermal sensor can detect heat. We report exactly what was and was not tested.
How to read the evidence tags
Controlled tests with instrumented heat sources.
End-to-end alert-chain test (device → cloud → push/SMS). Does NOT test the thermal sensor.
In-app outcome taps after an alert. Labels, not verified ground truth.
Reviewed or reproduced by a qualified third party.
Known limitations
- No field detection data yet — the device is pre-launch and has not run a kitchen pilot.
- Battery, thermal, latency, and connectivity figures are design targets, not measured results.
- Customer-reported alert outcomes (once collected) are labels, not independently verified ground truth, and are reported separately from lab and reviewed results.
- A synthetic alert-chain drill proves delivery, not that the thermal sensor can detect heat — the two are reported separately.
- Hand-wave dismiss, TV/voice integrations, and advanced AI features are planned and excluded from these figures until validated.
Test protocol v0 (draft) · last updated June 14, 2026. Stovyn is a monitoring aid, not a certified safety device, and does not guarantee fire prevention. Always keep working smoke and CO alarms.
We use cookies to improve your experience
We use essential cookies to make our site work. With your consent, we may also use non-essential cookies to improve user experience. Learn more