Watercress pilot (monitoring only)¶

A small two-zone trial to collect Pomona's first real sensor data: one hydroponic and one soil zone of watercress (Nasturtium officinale), sown at the same time. The aim is a trustworthy data set, not automation. Pomona drives no pumps, dosing or relays in this pilot.

Zones and IDs¶

	Hydro	Soil
`farm_id`	`home-pilot`	`home-pilot`
`zone_id`	`cress-hydro-a`	`cress-soil-a`
`device_id`	`cress-hydro-node`	`cress-soil-node`
`crop`	`watercress`	`watercress`
`system_type`	`hydroponic_dwc` (or `hydroponic_passive`)	`soil_pot`
Logged by node	air temp, RH, water temp, pH (ADS1115)	air temp, RH, substrate temp, soil moisture
Logged by hand	EC, pH check, water temp check	—

Hydro method: watercress naturally grows in cool, moving and oxygenated water, so DWC with the air pump and air stone is the recommended setup. Kratky/passive is an acceptable fallback if the pump or fittings aren't ready. Record which one you used as system_type.

Tools¶

Firmware: devices/esp32-cress-logger writes CSV to SD. It has not been bench-tested yet.
Importer: scripts/import_sd_csv.py sends SD and manual CSV rows to Core. It reports rejected rows, and re-importing the same file is safe.
3D-printed mounts: hardware/a1mini-sensor-station holds the probes on a rail (soil tray, SHT31). For the hydro zone, print its probe lid for a 1–2 L round container: pot seat plus pH, DS18B20, airline and float-switch ports that put the probes in the water. Print the fit ring first; not print-tested yet.
Templates: examples/pilot/manual_readings.csv (meter readings) and examples/pilot/watercress_plant_log.csv (plant observations).

# Before the pilot: in .env set RETENTION_DAYS=120. The default of 7 counts
# from import time, so older pilot data would otherwise be pruned.
./scripts/up.sh
python3 scripts/import_sd_csv.py /Volumes/SD/pomona_cress-hydro-node.csv --dry-run
python3 scripts/import_sd_csv.py /Volumes/SD/pomona_cress-hydro-node.csv
python3 scripts/import_sd_csv.py examples/pilot/manual_readings.csv --farm-id home-pilot

Always keep the original SD CSV files (for example under private/pilot-data/). They are the raw record. The Core database is a working copy.

Record probe calibrations in Core so later readings have provenance (raw is the probe voltage in volts; the dashboard's pH probe health panel then tracks sensitivity from one calibration to the next):

curl -X POST localhost:8080/v1/sensors/calibrations -H 'Content-Type: application/json' -d '{
  "device_id":"cress-hydro-node","farm_id":"home-pilot","zone_id":"cress-hydro-a",
  "sensor_id":"ph-probe-1","measurement":"ph","method":"two_point_buffer",
  "points":[{"reference":7.0,"raw":1.50},{"reference":4.0,"raw":2.03}],
  "performed_at":"2026-10-04T18:00:00+04:00","notes":"example values - use your readings"}'

Schedule¶

Dry run (before sowing, 24–48 h, no plants). Both nodes log with water and soil in place. Pull the power once to check that a new boot_id appears and the rows continue. Import, then check both zones in the dashboard's Modular observations history and in /v1/sensors/export.csv?kind=observations&farm_id=home-pilot. Fix wiring or pins before sowing.
Day 0: sowing. Record:
seed source and lot
sowing date and time
seeds per zone and sowing density
medium (soil type; hydro net pot and medium)
starting water pH, EC and temperature
a photo of each zone
a calibration record for each probe
Daily. Add a row to the plant log: germinated count, mean height, leaf colour 1–5, wilting or yellowing, water top-up, photo. Take one handheld pH/EC/water-temperature reading in the hydro zone and add it to manual_readings.csv.
Every 2–3 days. Copy the SD files to the Mac and import them. Check the import summary for rejected rows.
Weekly. Compare the probe pH with the handheld pH to spot drift, and recalibrate if they differ by more than about 0.3 pH.

Reference ranges (advisory, not wired into rules)¶

The ranges below are commonly cited values for watercress. They have not been checked against a reviewed source in this repo, so treat them as a note to verify:

water temperature roughly 10–20 °C (it dislikes warm water)
pH roughly 6.5–7.5
low to moderate EC

Imported readings are modular observations. They are stored, shown and exported, but they do not feed the reasoner pipeline yet. That is intended for a monitoring-only pilot. Pomona currently has no watercress reasoner. The tomato-risk endpoint returns source: not_applicable for other crops. The generic nutrient pH/EC reasoner still applies its general bands, such as high_ph at 7.2 and above. For watercress those labels are review prompts, not crop advice.

What this pilot does not claim¶

It does not validate the sensors. The dry run and handheld cross-checks are the evidence.
It does not use a watercress model or threshold rules.
It has no actuator path. Approving a suggestion only records a decision.

After 2–3 weeks of clean data, the next steps are:

replaying the digital twin against persistence on a real single-zone log
a watercress simulator scenario (GitHub issue #3)
possibly an anonymised dataset release, which needs separate approval