Good catch, and you’re right that it’s not real. On May 5, 2023 your Tempest had an off day: for that one date its readings were degraded. A lot of values came through missing or stuck, and the temperature channel reported a string of impossible numbers (down to −39 °F and up to 138 °F in bursts) before returning to normal. The real temperature that day was in the high teens °C. The days before and after look completely normal and the battery was healthy throughout, so it appears to be a one-day device hiccup rather than an ongoing or power problem.
Why the Tempest app doesn’t show it: its History view runs the data through a cleaning pass that hides momentary glitches. Our Charts page currently plots the raw, unfiltered feed. Great for fidelity, but it means a hiccup like this can show up as a false “record.”
Two things we’re doing about it. First, a filter to screen out physically-impossible readings so glitches won’t surface as records or distort the charts. Second — the part your report nudged us toward — rather than just hiding them, we’ll surface flagged anomalies in a data-health view, so you can see exactly when your station misreported. That makes it easy to tell a real record from a glitch, and if a unit starts misbehaving you’ll have the dates and values to report it to Tempest/WeatherFlow support.
I’ve pulled the full set of anomalies for that day (attached), and can send the raw minute-by-minute readings too. Given it looks like a one-day hardware hiccup, it may be worth flagging to WeatherFlow support. Happy to help you put that together. Thanks for flagging it. Exactly the kind of thing that helps us harden the data handling.
Why the Tempest app doesn’t show it: its History view runs the data through a cleaning pass that hides momentary glitches. Our Charts page currently plots the raw, unfiltered station feed. Great for fidelity, but it means a one-off hardware hiccup can show up as a false “record.”
Two things we’re doing about it. First, we’re adding a filter to screen out physically-impossible readings so these glitches won’t surface as records or distort the charts. Second, and this is the part your report nudged us toward, rather than just hiding them, we’ll surface the flagged anomalies in a data-health view, so you can see exactly when and how your station misreported. That makes it easy to tell a real record from a sensor fault, and if a sensor starts misbehaving you’ll have the dates and values on hand to report it to Tempest/WeatherFlow support.
On that note, I’ve already pulled the full set of anomalies for that day for you (attached). Thanks for flagging it. It’s exactly the kind of thing that helps us harden the data handling.
Sensor anomaly report — Eldred Acres (station 36121)
Device: Tempest 181828 · Date: 2023-05-05 (local, America/Los_Angeles)
Source: WeatherFlow device-observations API (raw, unfiltered)
Summary
On 2023-05-05 this Tempest malfunctioned for the day. Two problems, both confined to that one date:
- Degraded / incomplete reporting. Most sensor readings were missing or stuck. Of the 1,241 records logged that day: temperature present in only 88, pressure in 20 (a healthy day logs ~1,440), wind in 428, light/UV/solar in 37. Relative humidity was pinned at exactly 15% in 763 records (61%) — a stuck value, not a real reading.
- Physically-impossible temperatures. 23 readings outside any real-world range — ramping down to −39.5 °C (−39 °F) and spiking to +58.8 °C (+138 °F) — across 5 bursts. The genuine temperature that day was roughly 17–20 °C.
How May 5 compares to other days
A fully healthy day for this device is ~1,440 complete readings (one per minute, all sensors populated). May 5 stands out as corrupt; note May 12 a week later shows a different fault (an outage):
| Date |
Rows |
Temp present |
Impossible temps |
Pressure present |
Humidity stuck at 15% |
State |
| 2023-05-05 (this day) |
1,241 |
88 |
23 |
20 |
763 |
online but malfunctioning |
| 2023-05-12 |
381 |
381 |
0 |
381 |
0 |
offline ~17 h, then clean ¹ |
| 2023-08-14 |
1,439 |
1,439 |
0 |
1,439 |
0 |
healthy |
| 2023-02-01 |
1,434 |
1,434 |
0 |
1,434 |
0 |
healthy |
¹ On May 12 the device sent nothing from ~00:01 to 17:31 local (~17.5 h dropout), then reported cleanly every minute through midnight — a connectivity/power outage, not corruption.
Interpretation: May 5 was a one-day device malfunction — degraded reporting combined with impossible temperature output. Battery was healthy throughout (2.54–2.60 V), so it isn’t a low-battery brownout. Combined with the ~17-hour outage on May 12, this unit had a rough stretch in May 2023, which is worth reporting to WeatherFlow/Tempest support for a hardware check.
Temperature fault bursts (the impossible readings)
00:18–00:26 local — 9 readings, −15.6 °C → −39.1 °C (4 °F → −38 °F)
- 00:18 −15.6 °C (3.9 °F)
- 00:19 −20.9 °C (−5.6 °F)
- 00:20 −25.3 °C (−13.5 °F)
- 00:21 −28.9 °C (−20.0 °F)
- 00:22 −31.8 °C (−25.2 °F)
- 00:23 −34.2 °C (−29.6 °F)
- 00:24 −36.2 °C (−33.2 °F)
- 00:25 −37.8 °C (−36.0 °F)
- 00:26 −39.1 °C (−38.4 °F)
12:09–12:15 local — 6 readings, −30.1 °C → −39.5 °C (−22 °F → −39 °F)
- 12:09 −30.1 °C (−22.2 °F)
- 12:11 −32.8 °C (−27.0 °F)
- 12:12 −35.0 °C (−31.0 °F)
- 12:13 −36.9 °C (−34.4 °F)
- 12:14 −38.3 °C (−36.9 °F)
- 12:15 −39.5 °C (−39.1 °F)
13:36–13:38 local — 2 readings, +58.8 °C / +49.9 °C (138 °F / 122 °F)
- 13:36 +58.8 °C (137.8 °F) ← the “138 °F hottest day”
- 13:38 +49.9 °C (121.8 °F)
13:51–13:56 local — 5 readings, −16 °C → −27.4 °C (3 °F → −17 °F)
- 13:51 −16.0 °C (3.2 °F)
- 13:53 −19.3 °C (−2.7 °F)
- 13:54 −23.4 °C (−10.1 °F)
- 13:55 −27.4 °C (−17.3 °F)
- 13:56 −25.4 °C (−13.7 °F)
22:20 local — 1 reading
- 22:20 −38.6 °C (−37.5 °F)
Note
These spikes are why Halcyon’s Records showed 138 °F / −39 °F for this day. The official Tempest app filters momentary sensor glitches out of its History view, which is why you don’t see them there.