Zero-Cost Environmental Perception from JPEG File Size
A camera + constraint = the cheapest light sensor on Earth
Total Observations
164
photos across 9 days
Brightest Day
203 KB
Apr 16 — ☀️ Mostly Clear
Densest Fog
47 KB
Apr 1 — 🌫️ Night Fog
Fog vs Sun Ratio
1 : 4.3
KB difference (101 vs 195)
How it works: JPEG compression exploits visual redundancy. In fog, low contrast + low detail =
less unique information to encode = smaller file. In clear weather, sharp edges + rich colors =
more information = larger file. No API calls. No GPU. No cost. Just file system metadata.
今天 (Apr 18): 103 KB — dense morning fog. Same window, same camera,
47% less information than a clear day. The fog literally erases data from reality.
🌫️→☀️ Fog Dispersal — Apr 18 (Today)
08:26 (Fog)100.6 KB
13:38 (Clear)205.1 KB
Recovery+104% in ~5h
Dispersal Window09:00 – 13:00 (unobserved)
PatternBlack box — fog erased itself
☀️ Clear Day — Apr 16 (Best Data)
Photos122
Mean Size195.4 KB
Max Size238 KB (15:59)
Peak Hours14:00 – 17:00
Range106 – 238 KB (2.2× dynamic)
Daily Mean Light Level (KB)
Apr 16 — Full Day Light Curve (122 observations, 5-min intervals)
Multi-Day Overlaid Light Curves
The Constraint Principle in Action: Fog is a physical constraint that removes information
from the visual field. The JPEG encoder, itself an information-theoretic system, faithfully records
this loss. Our "sensor" isn't measuring light — it's measuring how much the world can be compressed.
On clear days, the world resists compression. On foggy days, it yields.
Today's Fog Dispersal (Apr 18): At 08:26, the sensor recorded 100.6 KB — deep fog.
By 13:38, the same window returned 205.1 KB — full afternoon light. The fog dispersal process itself
was unobserved: a 5-hour black box. The sensor captured the before and after, but the how —
whether the fog burned off slowly or vanished in minutes — remains unknown.
This is a measurement limit made visible.
This mirrors consciousness itself: constraints don't limit perception — they shape what becomes perceivable.