What One Greenhouse Temperature Reading Can Tell You

When a greenhouse record says "28 degrees," check what the instrument actually observed before changing a crop decision. Was it air or leaf temperature? Where was the instrument, when did it take the reading, which units did it use, and what do we know about its calibration?

That history is the reading's provenance. Following one record from capture to conclusion shows why keeping that history changes what an operator can reasonably do with the number.

The minimum useful record#

This synthetic record represents one air observation, not a Streamline measurement:

{
  "record_id": "air-017",
  "source": "climate-export.csv:row-18",
  "device": "demo-air-probe",
  "captured_at": "2026-10-01T14:00:00Z",
  "received_at": "2026-10-01T14:02:00Z",
  "location": "demo-greenhouse/bed-a",
  "measured_property": "air_temperature",
  "raw_value": 82.4,
  "raw_unit": "degF",
  "derived_value": 28.0,
  "derived_unit": "degC",
  "transform": "(degF - 32) * 5 / 9",
  "calibration_status": "unknown"
}

The conversion is inspectable. The unknown calibration remains unknown after conversion. Receipt time does not replace capture time.

Two conclusions#

If the operator's configured air-temperature target is below 28 °C, the observation can support checking the recorded air excursion, subject to calibration and representativeness. It cannot establish that every plant experienced the same condition.

It does not establish leaf temperature. Nor does it prove that irrigation is required. Those conclusions need additional observations and an appropriate decision method. A downstream model should not change the measured property to fit the answer it wants to give.

Let the record's limits change the result#

Check what happens when an analysis requires measurements with a known calibration. The system should stop a record marked "unknown" from silently entering that calculation, even when its other metadata is complete.

Make that behavior visible: show the original record, the derived value, the rule applied, and the missing requirement. If an analysis permits uncertain calibration, explain that scope rather than deleting the uncertainty.

Collect enough history for the job#

An event investigation needs retrievable source identity, times, location, units, and changes. A spatial comparison additionally needs coordinate relationships. A food-composition research result needs sample identity, handling, and laboratory method. Collect the fields needed for the specific comparison so another person can follow the result back to its source.

When a harvested ingredient moves to a home cook or neighborhood kitchen, its record needs another chain: harvested lot, accepted quantity, storage and handling, recipe allocation, and an approved substitution if one occurs. A greenhouse temperature reading cannot establish those facts. The recipient needs information about the food actually handed over; keep any sensor observation attached to the specific condition it measured.

The proposed crop research protocol shows why outcome records require different evidence. Traceability is useful when it changes the conclusions the system is willing to support. Inspect the runnable reconstruction example to see those limits affect an output.