Two gaps replace-the-laptop-cell.md names and nothing tracks
Runtime on battery. The runbook's step 8 produces one number: how long prometheus runs with mains pulled, at one load, on one day. "There is no rule and no series that would notice it halving." The mains-cut path the whole exercise protects is measured once and never again.
Cell temperature. node_power_supply_temp_celsius for prometheus's BAT0 ran 32.7–39.2 °C over the 24 hours to 2026-09-17, and the runbook writes the band down "so a reading far above that band has something to be far above" — and then: "nothing alerts on it and nothing watches it." The failure mode #454 opens with is a swollen thirteen-year-old cell on a shelf, and it is the one failure mode here with no signal at all.
What to build
- A rule on
node_power_supply_temp_celsius for the two laptops, thresholded above the recorded band with margin, severity: critical — a hot cell is the one alert here that should wake someone. Unit-tested.
- For runtime: either a rule on the discharge rate while
HostOnBattery is firing (deriv() of charge_now gives minutes remaining without waiting for zero), or a written decision that the annual step-8 measurement is enough. The second is defensible; it should be written down in host.rules.yaml's header if taken.
- Both laptops.
oracle's cell measures worse and has the same series.
Done when
promtool test rules covers the temperature rule for both hosts, and the runtime question has either a rule or a sentence.
Refs #454, #519.
Two gaps
replace-the-laptop-cell.mdnames and nothing tracksRuntime on battery. The runbook's step 8 produces one number: how long
prometheusruns with mains pulled, at one load, on one day. "There is no rule and no series that would notice it halving." The mains-cut path the whole exercise protects is measured once and never again.Cell temperature.
node_power_supply_temp_celsiusforprometheus's BAT0 ran 32.7–39.2 °C over the 24 hours to 2026-09-17, and the runbook writes the band down "so a reading far above that band has something to be far above" — and then: "nothing alerts on it and nothing watches it." The failure mode #454 opens with is a swollen thirteen-year-old cell on a shelf, and it is the one failure mode here with no signal at all.What to build
node_power_supply_temp_celsiusfor the two laptops, thresholded above the recorded band with margin,severity: critical— a hot cell is the one alert here that should wake someone. Unit-tested.HostOnBatteryis firing (deriv()ofcharge_nowgives minutes remaining without waiting for zero), or a written decision that the annual step-8 measurement is enough. The second is defensible; it should be written down inhost.rules.yaml's header if taken.oracle's cell measures worse and has the same series.Done when
promtool test rulescovers the temperature rule for both hosts, and the runtime question has either a rule or a sentence.Refs #454, #519.