The sensor tells you about now. The historian remembers everything — and remembering is where the money is.
A monitoring system without a historian is a smoke detector: it tells you about the present moment and forgets it the instant it passes. A monitoring system with one is your operation's memory — every reading kept, stamped with when and where, outliving the sensor that took it, the app that showed it, and the vendor that sold it. The historian is the part of the system where "Data is King" stops being a slogan and starts earning money, because the most powerful question in CEA is not what's wrong right now. It's "where did this crop diverge from the best one I ever grew?" — and only a historian can answer it.
What to log, and how often
The rule is blunt: if you didn't log it, it didn't happen — and you can't ask a question later about a number you never wrote down.
- Every input, at one-minute intervals at minimum; five-second intervals for the fast, critical ones (temperature, VPD, dissolved oxygen).
- Every setpoint change — who changed it, when, the old value, the new value. Most "the system did something weird" mysteries are a setpoint someone moved and forgot.
- Every alarm — when it fired, when it was acknowledged, when it resolved, and the corrective action that was logged with it.
- Every recipe change and growth-stage transition, so the record knows which targets were in force for every data point.
- Calibration events — date, slope, offset — so a probe's slow death is visible in the trend before it fails.
- Equipment runtime — pumps, chillers, fans, dehumidifiers — which turns into predictive maintenance and an early read on efficiency loss.
The site's own line says it best: AI can't tell you a thing about a crop you never measured. The historian is where "measured" becomes "kept."
You can't keep five-second data forever — and you don't need to
Storing every five-second reading for years would bankrupt the system, so it doesn't. It downsamples instead: keep the fine-grained detail for a week, five-minute averages for a few months, hourly averages for a couple of years, and a daily figure forever. You never lose the daily picture, and you never drown in the per-second one. That single design choice is what makes a real historian affordable at the scale of a working farm.
The payoff: comparing this cycle to your best one
Here is what the kept record unlocks, and it's the capability nothing else on the system can match.
Cross-cycle analysis. Lay this crop's environmental data over your best-performing past cycle for the same cultivar, and see exactly where the two diverge. Post-harvest review stops being "what went wrong?" — a question nobody can really answer — and becomes "where, precisely, did conditions leave the pattern that produced our best result?" That's a question with an address.
Correlation with outcomes. With enough cycles logged, you can associate each input's history with the things that actually pay: yield per square metre, quality (potency, Brix, shelf life, visual grade), cycle length, when disease showed up, and input cost. The ten-input framework is the structure for that correlation — each input's trend tested against each outcome — and it tells you which inputs have the most leverage on the results you care about most. That is how a grower stops guessing which knob matters.
Derived numbers are data too
The most useful metrics aren't ones a sensor reads directly — they're calculated: VPD, the daily light total, the oxygen margin, the Pythium-risk index, the estimated calcium-delivery rate. A serious historian stores these as first-class data, not as lines on a chart that vanish when you close the screen. Because they're stored, they can be trended over months and alarmed on like any raw reading — the Pythium index can fire a critical alarm; the calcium-delivery estimate can warn you before a deficiency shows. (How those numbers are computed and stored is the subject of the architecture.)
The sensor reads the present. The historian builds the institutional knowledge no single person can carry — the record that's still yours, and still valuable, when the sensor, the app, and the vendor are all long gone.