430,022 readings from three probes on one charge each.
Three Setpoint units were charged, placed in a bedroom, a refrigerator and a freezer, and left logging in always-on mode — roughly one reading per second — until each stopped on its own. Here is what that first round of data shows.
across 3 probes · 421,945 unique timestamps (1.9% dup)
Best single-charge run
160,916
bedroom unit · one continuous 46.2 h session, zero gaps
Continuous runtime
46.2 h
≈1.9 days always-on at ~1 reading/sec (bedroom)
Effective sample rate
0.97 /s
bedroom · fridge 0.71/s · freezer 0.87/s
Read this as a data-capture result, not a battery spec. These units ran in the most power-hungry always-on ~1 reading/second mode. Battery voltage was not logged, so "one charge" is the test setup, not a measured run-to-empty. The slow-reporting / deep-sleep mode behind the "weeks per charge" figure was not part of this round.
Readings captured per unit
Every unit filled its log with six figures of data on a single charge. The bedroom unit ran uninterrupted; the two cold units were re-powered later to check their on-board buffer, which shows as gaps.
Two days, three environments
Mean temperature per unit through the logging window; the shaded band is each unit's min–max spread. Each trace ends where that unit stopped logging.
FreezerRefrigeratorBedroom0 °C (freezing)
What the temperatures revealed
Descriptive of these three appliances over ~2–3 days — not a product accuracy spec.
Freezer
Average−21.9 °C
Coldest−27.3 °C
Time below 0 °C99.8 %
Readings139,449
Rock-steady deep-freeze with clean, repeating defrost/compressor cycles captured at full resolution.
Refrigerator
Average+0.5 °C
Coldest−4.4 °C
Time below 0 °C45 %
Readings129,657
Runs cold. Nearly half the readings were below freezing, dipping to −4.4 °C — the kind of hidden swing that freezes produce and is exactly what a logger is for.
Bedroom
Average21.2 °C
Range19.8–24.9 °C
Std. dev.±0.5 °C
Readings160,916
A calm reference environment — small, slow daily drift, no surprises. This is the clean uninterrupted run.
What this round does and doesn't prove
Every number above is measured. Here is what it supports, and — just as important — what it does not.
✓Data throughput on one charge.A single unit captured 160,916 readings over 46 continuous hours at ~1/second — six figures of data with zero dropouts.
✓High-resolution capture.1-second sampling resolved every defrost cycle and the fridge's sub-freezing dips — detail a periodic logger would miss.
✓On-board buffering works.The freezer still held readings in its buffer after the run, recoverable later — data survives the hub or Wi-Fi being away.
✕Not a measured battery life.No voltage/charge was logged, so "ran to empty" isn't proven. Add a battery-voltage column and re-test to claim true run-to-empty.
✕Not the deep-sleep number.This was always-on ~1/s — the worst case. The "weeks per charge" slow-reporting mode was not exercised and remains pending.
✕Cold units weren't perfectly continuous — since fixed.Fridge and freezer showed ~4.6% and ~1.5% duplicate timestamps, from the buffer re-flushing over weak in-appliance Wi-Fi. This test is what found it. The store now enforces a unique index per probe and timestamp and ignores a re-sent reading, so a repeated flush can no longer duplicate a row; rows that arrive without a clock are stamped a millisecond apart so a whole chunk isn't collapsed into one. Left on this page rather than quietly removed, because a test that only ever confirms what you hoped isn't worth publishing.
Setpoint field test · 3 × ESP32-C3 + DS18B20 · always-on ~1 Hz · figures independently recomputed from the raw CSVs.n = 3 units, one run each.