╔══════════════════════════════════════════╗ ║ ║ ║ S T A N D A R D F I E L D ║ ║ I N S T R U M E N T S ║ ║ ║ ╚══════════════════════════════════════════╝
The measurement gap
The cost of intelligence is falling toward zero. Large language models and modern inference infrastructure are making it cheaper every year to process, analyze, and reason about information. But intelligence is only as useful as the information it operates on.
The sciences that study the physical world (ecology, environmental science, geology, hydrology) are bottlenecked by observation. The models are sophisticated. The satellites improve every year. But the ground truth layer, the actual measurements taken from actual systems, remains sparse. Most of the planet's forests, watersheds, soils, and subsurface geology are measured at a handful of points, a handful of times.
Two theses:
We will soon be more constrained by our ability to observe the world than by our ability to reason about it. This applies to both the capture of information and the delivery of information, from raw measurement to reproducible, published result.
Today, scaling observation requires real tradeoffs. You can zoom out and cover more ground, but at increasing coarseness. You can zoom in for detail, but only at a few points. These tradeoffs are artifacts of the current economics of instrumentation. They can be obsoleted with better tools.
Why the instruments are expensive
The sensing components inside most field instruments are widely available and inexpensive. The cost comes from the market structure. Scientific instrumentation is a small, captive market where manufacturers vertically integrate the sensor, the logger, the software, and the calibration into a single proprietary bundle. There is no interoperability between vendors. No standard interfaces. Each new sensor purchase locks you deeper into one ecosystem. Volume stays low. Prices stay high.
The gap between what modern commodity electronics can deliver and what the scientific sensing market actually provides is wide and growing. This is a lagging-edge opportunity. The manufacturing and standardization practices needed to close it already exist in adjacent industries.
Standardization is the unlock
The instruments get cheaper when they can be manufactured at volume. They can be manufactured at volume when designs converge on shared interfaces. Shared interfaces emerge from standards.
When every instrument in an ecosystem shares a common connector, a common voltage range, and a common communication protocol, several things happen. Components become interchangeable and individually replaceable. Manufacturers can specialize and compete on the parts they do well rather than building entire vertically integrated systems. The ecosystem grows horizontally: a new sensor type enters the network without requiring a new logger, a new cable, or a new software stack. And scientists gain the ability to build and modify the tools to suit their specific research needs, rather than working within the constraints of a vendor's product roadmap.
Standard Field Instruments is an effort to define those shared interfaces for field and laboratory sensing. The work is early. The principles below are a working draft, revised as we build and deploy real hardware.
Principles
A sensor reports voltages, resistances, and counts. Derived quantities like sap flux or soil moisture content are computed in separate, versioned, reproducible software. Algorithms improve. Calibration models get revised. Raw data can always be reprocessed. Derived values are frozen to the assumptions that produced them.
A voltage reading without its excitation voltage, gain setting, timestamp, sensor identity, and calibration coefficients is just a number. Provenance and metadata must be automatic and structural, embedded in the device and its output stream. The metadata crisis in the natural sciences loses more data than noisy ADCs do.
Define the mechanical, electrical, and data contracts. A standard connector. A standard voltage range. A standard data format with a standard metadata schema. Everything behind that interface is free to vary: the sensing element, the analog front end, the microcontroller, the enclosure. Standardization at the boundary enables diversity and competition behind it.
A sensor measures. A logger logs. A radio transmits. A power supply provides power. When these functions are combined into a single sealed unit, the result is vendor lock-in and reduced modularity. When they are separated by standard interfaces, each component can be developed, tested, replaced, and scaled independently.
Automated PCB assembly on modern surface-mount lines produces more consistent, more reliable hardware than hand-soldered boutique builds. Standardization enables volume. Volume enables automated manufacturing. Automated manufacturing simultaneously reduces cost and improves quality. The same forces that made consumer electronics reliable and affordable have not yet reached scientific field instruments.
The firmware ecosystem for field instruments has historically suffered from poor dependency management, platform lock-in, and limited code reuse. Modern embedded Rust, with cargo's package management, a rich trait and generics system, and async runtimes like Embassy, provides a foundation for genuine composability across hardware platforms. A sensor that implements a shared trait can be understood and integrated by any compatible tool in the ecosystem. The barrier to writing systems-level Rust is also falling as LLM-assisted development matures. The compiler's strictness becomes an asset in this context: it catches the mistakes that both humans and language models make.
The full pipeline
Scaling observation requires solving both capture and delivery. Standard Field Instruments addresses the capture side: making the hardware cheaper, more open, and more interoperable so that measurement density can increase by orders of magnitude.
The delivery side, getting raw data through reproducible transformations into trustworthy published results, is an equally important bottleneck. OzzyDB is a companion project addressing this problem. OzzyDB treats data as functions: it stores raw inputs and versioned transformation code, executes transforms in deterministic sandboxed environments, and identifies outputs by cryptographic hashes of their full lineage. Any derived dataset can be independently reproduced from its raw inputs.
Together, these projects aim to make the full chain from transducer to publication auditable. The sensor reports exactly what it measured and how. The data pipeline records exactly what was done to that data and why. Every step is versioned, every output is reproducible.
Current status
This is early-stage work. A prototype sap flux sensor interface is in development. The principles above are a working draft. No hardware is in production. The standard will be revised as real deployments reveal what works and what was wrong.