Recorder · last 15 min
Primary channels
Chamber & micro-climate
CO₂ flux
—
Skin temp
—
Chamber RH
—
Dew point
—
Skin indices · relative 0–100
MiCS-6814 equivalents · 8 channels
Single-gas equivalents from the datasheet curves — what each channel
would read if that gas were the only one present. Skin emits many at once, so read these
as eight correlated features, not eight measurements.
Raw resistance · the real ML features
MiCS reducing
—
MiCS oxidising
—
MiCS ammonia
—
BME690 gas
—
Capture R0 in fresh outdoor air, away from skin. Every ppm figure is
Rs/R0 against that capture — a bad R0 poisons the whole run.
History
—
Window statistics
Recording
—
Card storage
Used
—
Free
—
Days left
—
Sessions by day
Card setup
Format
FAT32 only. The ESP32 SD library does not
read exFAT, which is what most 64 GB+ cards ship with — reformat before first use.
Layout
/skingas/
YYYY-MM-DD.csv
YYYY-MM-DD.csv
Bus
SD_MMC 4-bit on GPIO 14/15/16/17/18/21.
Already routed on the board.
Row size
~180 bytes × 31 columns. At 5 s that is
about 3 MB per 24 h.
Claude analyst
idle
Nothing analysed yet.
Pick a mode above, then run it. The reference table below is sent along with your
readings so the interpretation is anchored to published skin-gas work rather than
to the model's general impressions.
Not a medical device and not diagnostic. MOX sensors cannot identify a
specific compound — the literature below says what a real instrument would find, which is
exactly the context needed to judge whether a channel movement is plausible.
Literature reference · sent with every analysis
Appearance
Text size
Typeface
Palette
Preview
NH\u2083
ammonia
1.55
Small print looks like this \u2014 notes,
captions and the literature text all use it.
Saved on the device, so it follows you to any phone that opens
the dashboard.
Connection
Router network (SSID)
Password
The device joins your router and picks up an IP by DHCP. Its address
appears on the device SYSTEM page after it reconnects.
Analyst key
Anthropic API key
Stored in the device's NVS, never in the firmware source and never in
this page. It travels over plain HTTP on your LAN, so set it while you are on a network
you trust — or in AP mode, where nothing else is listening.
Run label
Written to every CSV row as the supervised-learning target. Set it
before the run, not after.
Display & logging
Backlight 80%
Log interval (seconds)
Calibration
Let the array burn in 24–48 h before the first capture — new MiCS and
BME690 elements drift heavily and anything logged before then is not comparable to what
comes after.
Device
Battery
—
Runtime left
—
Uptime
—
Free heap
—
WiFi
—
Address
—