Skip to main content
Case study · live deployment

A California operation.

4,000 contracted acres of citrus and pistachios across our paying customers (October 2026).

See how a California orchard operation brings field readings, trained model estimates and adviser-reviewed water orders into one weekly workflow. We never name customers. The numbers below come from this deployment's sensor and meter records. Probe reading summaries are signed and can be checked with our public key. Well-meter readings from the telemetry feed are sealed into a separate signed chain as they arrive.

Deployment figures as of Jul 8, 2026. The network includes long-term training sites beyond the commercial deployments: 70 instrumented sensor sites as of Aug 30, 2026.

A crop adviser walks an orchard row at sunrise with a tablet and a soil probe
4,000contracted acres across our paying customers (Oct 2026)Citrus and pistachios, block by block (as of Jul 8, 2026)
30+live soil-sensor sitesMulti-depth probes, 10 to 100 cm, as of Jul 8, 2026; reporting gaps are flagged and labeled
Meteredwell flow telemetryLive pump and flow telemetry where a meter is installed; unmetered wells are labeled estimates
Signedreading summariesSigned at attestation; certificates checkable with the public key
Tested

The Monday list, tested.

In a pre-registered replay on a citrus estate, at the same number of stem water potential (SWP) readings, a fixed rotation on a shorter round would have left about 20 of 102 deep-stress block-weeks unread in their week and the Monday list 3. A careful crew sorting last week’s readings by hand would have left about as many as the list. We also asked the same question of Claude and Gemini.

Smarter rotations that ignore the readings closed most of the gap. Counts this small do not separate these orders. On the shorter round the list reached about 11% more stressed blocks per reading than a fixed rotation and somewhat more than the strongest rotation that ignores the readings, counting a block as reached when it is read in the week it is stressed or early the next. It was level with a careful hand sort. The list does that kind of sorting every Monday, for every block. Over a full week a plain rotation catches at least as many.

Replay on our longest-running orchard’s past seasons, 2023 to 2025, the orchard the list was developed on, on a shorter round. The crew read every one of these blocks in its week, nearly all by midweek; the replay asks which a shorter round would reach. Half of the 102 sit on three blocks. These figures are the replay’s. The list in service sets the stress line another way, so fewer block-weeks count as stressed and too few as severely stressed to compare orders. Scored its own way, it again reached more stressed blocks in their week than a fixed rotation and was not separated from a careful hand sort or from the strongest rotation that ignores the readings. Scored on the block-weeks the replay counts as stressed, its lead over a fixed rotation was smaller. Deficit-trial rows excluded. Development result, not a field trial. The adviser makes every call. Methods, plans and figures are available in diligence. This result comes from one citrus estate, the one the list was developed on. It has not yet been shown on another farm.

The starting point

Permanent crops under an allocation cap.

Permanent crops under a SGMA allocation cap mean every acre-foot has to be placed where the trees actually need it.

  • Cold soil can slow root uptake while moisture reads fine.
  • Salt builds where roots feed.
  • A pistachio year is decided by stress timing around hull split.

The operation needed block-level evidence, and records that show what was recorded and whether it changed after signing.

The deployment

Four steps, all running today.

  1. 01

    Instrument

    Multi-depth soil probes across 30+ sites (as of Jul 8, 2026) and flow meters on the wells that have them, feeding one platform.

  2. 02

    Verify

    Probe readings and telemetry well-meter readings sealed into a signed chain that is tamper-evident from signing forward. A change after signing can be detected.

  3. 03

    Advise

    Three trained models run every week: two estimate crop stress, trained on our own stem water potential (SWP) readings matched to soil probes and weather, and one estimates the hours each block’s pump ran, from probes, weather and pump records. The stress models have not beaten the simple rules a grower already has, such as the block’s latest SWP reading. The Thursday draft water order comes from rules mined from our agronomist’s past orders for each block; it does not read that week’s weather, sensor data or any model. On Platform, your own Certified Crop Adviser or adviser makes the call. On Managed, our Certified Crop Adviser does. No trained model’s score enters the order on its own. The operation gets a Monday written briefing and a Thursday water order, plus a live dashboard.

  4. 04

    Account

    SGMA allocation tracked per entity in real time, with signed reading summaries behind the figures for any reporting or verification ask.

What the operation gets

Each block's available readings and clearly labeled estimates.

A live operations dashboard

Each block card shows soil moisture by depth, pump activity and water used against the allocation, on the web and in the app.

Labeled reads on every ranch card

Soil moisture by depth and soil temperature, the water-year burn rate and, on pistachio blocks, the hull split window projection in field beta, not yet scored. Each read is labeled by source: a measurement, a calculation, a model estimate or a planning estimate. The cold-soil and salt calls are off: neither held up when checked.

Twice-weekly written guidance

Model outputs depend on the site's data and promotion status; planning outputs remain labeled as estimates. The operation receives written irrigation guidance twice a week, reviewed by our agronomist, plus alerts when something needs attention sooner.

SGMA allocation tracking per entity

Live usage against allocation for every entity, with over-allocation exposure estimates. A monthly meter record with dated totals and missing readings flagged, in the form a GSA reviewer asks for. Your GSA decides what it accepts.

Tamper-evident verification

Reading summaries are signed at attestation and chained per site, tamper-evident from signing forward. Signed probe reading summaries are checked against our public key when their verification page opens. Well-meter readings from the telemetry feed are sealed into a separate signed chain as they arrive. Issued water orders are kept with the readings behind them and are not signed. Tamper-evident records preserve attested readings and reports; they do not by themselves establish measurement accuracy, water savings or regulatory compliance.

Sensor and equipment health watch

The platform monitors its own instruments: probes that stop reporting or return error values get flagged so the data feeding decisions stays trustworthy.

The platform behind it

No trained model’s score enters the order on its own.

We report each model's test method, evaluation data and limits separately.

A trained model projects each probe’s soil moisture three days ahead if no irrigation or rain occurs. On a second farm never used in fitting, for probes with their own history, its three-day projection in periods when the probe saw no wetting was closer than every plain rule we tested, in the test plan and after it: its miss was about seven tenths of the best plain rule’s. For a brand-new probe with no history (tested on established probes shown none of their history) a plain rule does as well or better.

Only periods when the probe saw no wetting, on one farm and one probe make, and not a fresh blind test: that ground had been scored before. With that farm’s locations weighted equally, the model’s lead over the best plain rule is not clear. With irrigated periods removed the lead held; most remaining periods could not be checked, and on checked dry ones it is unclear. On the checked dry periods the lead is in the top 60 centimeters; deeper down, a plain rule does as well. Its printed range can be too narrow on ground it was not fitted on. The first forward score is November 23, 2026. It is a projection, not a call to irrigate.

Planning outputs remain labeled as estimates.

On five mature orchards (113 acres), median pump hours fell 17.7% after sensors went in, adjusted for rainfall. That was observed while the agronomist and the sensors ran the irrigation, with no model in the loop; across all 15 sites the result was not statistically significant (p = 0.24).

What we do not claim: water savings attributed to the AI. We label each figure by what it represents: a measurement, an estimate or a benchmark.

Probe readings are sealed into a signed, tamper-evident chain once each logger day closes, on the sites where daily sealing runs today.

The groundwater program has no independently validated result.

See the same product with sample data.

The sample dashboard is the same product with fictional data.

Sample pistachio orchard Sample data
Water used to date
96 AF Estimated

of a 672 AF allocation. Sample planning figure, not measured use.

Sample position as of Oct 1, 2026

Estimated Projected 130 AF by Dec 31, inside the allocation

Soil probes
3 of 3 live

Block 06 · probe NR-1
Above the block’s lineMoisture
1.42Salt sensor index
72°FSoil temp
12.5 bars, FriLast SWP reading