Webhook automation: master pushes force a git pull then reload_all;
core restart only for configuration.yaml / packages platform files.
Add-on auto_restart disabled separately on the host (race source).
Detail: https://projects.knownelement.com/issues/734
Founder ruling: proprietary nvidia driver only (needed for CUDA, version
choice his) — no nouveau. Anchor pre-flipped; sensor stays unknown until
the driver lands, then populates with zero further changes.
NVMe 150F, DIMM 140F, PCH 160F triggers on the new storage/mem/chipset
sensors (104F general threshold retained for air/CPU); UPS runtime-low
alert (<10 min while not Online) guards the #455 shutdown window.
https://projects.knownelement.com/issues/344
HA sidebar has no dashboard folders — nesting = one dashboard with tab
views (same pattern as Server Room). Art Room/Garage views merged into
stlp.yaml verbatim; old single-purpose dashboards removed.
https://projects.knownelement.com/issues/344
notify.send_message rejects extra data['data'] (old mobile-app push
syntax) — UPS battery low, UPS comm lost, and high-temperature alerts
all errored on every fire. Blocks removed; notifications send plain.
https://projects.knownelement.com/issues/344
8 new two-layer sensors from data already in lm-sensors extends: NVMe
composite (tsys3/5), 4x jc42 DIMM temps anchored by i2c address (tsys4),
PCH (tsys3/9). GPU/SATA/DRAC-DIMM temps need host-side work — ticketed.
https://projects.knownelement.com/issues/344
Raw entity_id comes from the name slug (pfv_garage_pdu_output_current_raw),
not the unique_id (pfv_garage_pdu_amps_raw) — only this sensor's name and
unique_id diverge, so only its semantic was reading a nonexistent entity.
https://projects.knownelement.com/issues/344#note-3916
HA 2026.8 rejects ANY non-numeric native_value on a unit-bearing snmp
sensor at entity ADD (host down at boot renders ''/'unknown'/'None' as
strings) — unit-bearing adds crash and the entity vanishes until manual
reload. Raw layer now carries no unit (validation off); semantic template
layer carries unit + device_class under the original unique_ids so
entity_ids, dashboards, and automations are unchanged.
https://projects.knownelement.com/issues/344#note-3916
HA 2026.8 raises ValueError(int('')) at entity ADD when a device_classed
snmp sensor's first render is empty (host down at boot/reload), so the
entity silently vanishes (artroom/garage/subopi3). Units alone keep the
°F/W/A display; automations are numeric_state, dashboards don't filter
on device_class.
https://projects.knownelement.com/issues/344#note-3916
Second top-level sensor: key (PDU energy) made YAML last-one-wins drop
the whole first block at config load — every SNMP temp/power entity
stopped being created after the restart that pulled this file.
https://projects.knownelement.com/issues/344#note-3916
mrcharles