One year of office weather through a matchbox-sized sensor
– days observed
– sensor rows
– → –
Central Time (America/Chicago)
–°F median
Sensor-frame room temperature
–rows / day
Effective sample rate after cleaning
–°F
Median day swing
–%
Median humidity
–lux
Median peak daylight (1 pm)
–days
Longest continuous uptime
–
Storm-front signatures
–%
Time sensor saw darkness
Section 01 · Seasonal climate
A dry office, a warmer summer, and a winter that drains the humidity
Monthly medians with 10–90th percentile ribbons. The CLUE's SHT sensor reads a few degrees warm from self-heating, so treat the level as sensor-frame; the shape of the curve is the real signal.
TEMPERATURE · °F
HUMIDITY · % RH
–
Section 02 · Daily rhythm
The room has a heartbeat, and it's 1 pm
Median light level for every hour × weekday cell across the full year. Darker = dimmer, warmer = brighter. Sound turned out to be noisy-in-all-hours — ambient HVAC and traffic dominate — so daylight is the honest occupancy proxy.
LIGHT · weekday vs weekend · median lux by hour
SOUND · loud-event rate · % of samples > 100
–
Section 03 · Thermal memory & HVAC
The room doesn't cycle — it drifts
No crisp furnace/AC oscillation shows up in the autocorrelation. Instead, the office has a slow thermal memory: any disturbance to temperature decays on a ~40-minute timescale. The day-to-day swing is the stronger HVAC fingerprint.
DAILY TEMP RANGE · °F (max − min per day)
AUTOCORRELATION · temperature vs. its past
SAMPLE DAY · 2025-05-12 (Monday) · raw 30-second temperature trace
–
Section 04 · What the CLUE sees
38% black, 15% red-brown, and a story of day vs night
The APDS-9960 color sensor reads whatever surface sits in front of it. Over a year, the dominant tones are deep reds/oranges and black — consistent with a wood-toned desk, a dark surface after hours, and daylight glancing off warmer surfaces.
OVERALL SHARE · binned hue
DAY (7a–6p) vs NIGHT · hue share
TOP 12 DISTINCT COLORS · excluding black/white/gray
PERCEIVED BRIGHTNESS · median by hour
–
Section 05 · The office barometer
A $40 dev board caught 87 weather fronts
The LPS22 pressure sensor is accurate enough that sub-hourly pressure changes track weather systems. A drop of > 2.5 hPa in six hours flags an oncoming front — 87 candidate events over the year.
DAILY PRESSURE · min / median / max · hPa
BIGGEST PRESSURE SWINGS · 24-hour Δ · hPa
STORM-FRONT SIGNATURES · Δ6h < −2.5 hPa
–
Section 06 · Device health
21 reboots, one 53-day run, and a strong preference for rebooting at 10 a.m.
The timestamp_monotonic field resets whenever the CLUE reboots. Those resets are themselves a diagnostic — they cluster at hours consistent with human intervention, not random crashes.
REBOOT EVENTS · uptime before restart · days
REBOOTS BY HOUR OF DAY · Central time
–
Section 07 · Ground truth · Open-Meteo cross-check
The $40 dev board caught –% of real storms
Joined the CLUE's hourly pressure to Open-Meteo's ERA5 reanalysis for the exact same coordinates (Goodlettsville, elevation 148 m). The elevation-corrected offset lines up within a fraction of a hPa with physics, and two-thirds of the flagged storm fronts coincide with actual precipitation or storm weather codes.
–
Pearson r · raw hourly pressure
–hPa
Elevation offset · exp. −17.8
–%
Storm signatures verified
–°F
Shoulder-season indoor-outdoor gap
PRESSURE · CLUE vs Open-Meteo · hourly · hPa
INDOOR vs OUTDOOR · monthly median · °F
VERIFIED CATCHES · CLUE front → Open-Meteo rain
DRY FRONTS · CLUE flagged, no Open-Meteo rain
–
Anomalies & quirks
Three rows the dataset didn't want you to see
DROPPED DURING CLEANING
corrupt2025-07-17 14:18:06 UTC — a hex color (#1C1211) appears in the timestamp_monotonic field. Classic column-shift from a missing value upstream of the gateway.
glitch2025-07-17 13:58:54 UTC — light reading of 5,061,000 lux, roughly 40× the solar maximum. Same afternoon as the corrupt row; the sensor had a bad 20 minutes.
impossible2025-09-27 12:47:29 UTC — pressure of 72.0 hPa, a value that would require sitting at 18 km altitude. Single-frame glitch.
ceiling246 samples at 65,535 lux — direct sun hit the 16-bit sensor; kept in the dataset but capped.
gapDec 14 2025 → Feb 17 2026 — the device was dark for most of the winter (two separate outages of 30 and 50 days).