Equipment
OEM integration: the interface between a decision and a machine
Definition
OEM integration is the defined software interface through which an agricultural equipment manufacturer consumes a pest-management decision: how a treatment zone is expressed geometrically, which coordinate conventions apply, what runs on the machine versus in the cloud, which exports are produced, and what operating limits travel with the decision. Swathmark builds the decision layer, not treatment machinery.
This page describes the interface being designed and states plainly what has not been demonstrated: no equipment platform has been integrated or tested.
- Published
- Last updated
What does OEM integration mean for Swathmark?
It means treating the boundary between decision and machine as a software contract rather than a bespoke project. The interface specifies what a decision contains, what it means, and what limits it carries — so an equipment team can build against it.
Equipment capable of targeted application — spot spraying rather than a uniform broadcast pass — has to be told where to act, and that instruction is only as good as the decision behind it. The reason to define the contract first is that the alternative does not scale and does not stay honest. A one-off integration hides its assumptions inside a project, and the next one repeats the work while quietly making different assumptions about geometry, timing, and what the machine is permitted to do with an ambiguous result.
Does Swathmark make treatment equipment?
No. Swathmark is the decision layer and is intended to work with compatible third-party equipment. Dedicated treatment machinery is outside the current platform scope.
This is a deliberate sequence rather than a stage in a roadmap towards building machines. A treatment machine built before the decision engine is validated is a machine acting on unvalidated decisions, and the hard problem — knowing whether to act — is not made easier by owning the actuator.
What runs at the edge and what runs in the cloud?
The split follows latency and evidence. Work that has to keep up with a machine belongs on the machine; work that needs full lineage, review, and versioned policy belongs where that record lives.
| Concern | Where it belongs | Why |
|---|---|---|
| Image capture and quality screening | Edge | A capture defect is worth catching before the pass ends, not after. |
| Detection inference | Edge or cloud | Depends on the compute budget available and on whether latency binds. |
| Threshold policy evaluation | Cloud, or an edge copy pinned to a policy version | The policy version must be identifiable either way. |
| Decision record retention | Cloud | Lineage, review history, and replay need durable storage. |
| Human adjudication and review | Cloud | Ambiguous cases need a reviewer and a record, not a field decision. |
| Treatment-zone export | Either | The export is a representation; where it is produced does not change its meaning. |
Edge inference constrains model size and compute, and it constrains what evidence can be retained. Both consequences are part of the interface, not implementation detail to be discovered later.
What does a machine-readable treatment zone contain?
Georeferenced geometry for the area a treat decision covers, plus the references and limits that make it interpretable: the decision it came from, the policy version behind it, and the window in which it is valid.
- Geometry for the area the decision applies to, with an explicit coordinate reference system — an unstated CRS is the classic way a zone lands in the wrong place.
- A reference to the decision record that produced it, so the zone can be traced back rather than taken on trust.
- The threshold policy version and model version behind the decision.
- A validity window, because pest development and conditions make a stale zone unsafe.
- The operating limits the decision carries, including the conditions it was produced under.
- Areas of abstention marked as such — distinct from no-treat, because they mean something different.
A treatment zone is a representation of a decision, not an instruction. Whether an implement can act at that resolution, at that speed, with that product, under that label, is an equipment and operator question that the zone does not answer and must not appear to.
Which equipment is compatible?
None has been integrated or tested, and no compatibility is claimed. Equipment compatibility requires a named platform, a tested interface version, and an approved evidence record; none of those exists yet.
That is a direct answer to a question equipment teams are right to ask first. The interface is being designed around conventions that site-specific application workflows already use — georeferenced vector geometry with attributes, an explicit coordinate reference system, and per-integration format agreement — but designing against a convention is not conformance to it, and this site will not describe it as such before it has been verified.
No equipment manufacturer, dealer, or platform is named anywhere on this site as a partner, customer, or integration. Any such statement would require written authorization.
What does the interface not do?
It does not control machinery, select or meter product, or authorize an application. It delivers a decision and its limits; acting on it remains the equipment's and the operator's responsibility.
- No machine control, actuation, boom, or nozzle command.
- No product selection, rate calculation, or mixing instruction.
- No pesticide-label interpretation or regulatory determination.
- No override of an operator's judgment or of a machine's own safety interlocks.
- No implicit permission to act on a stale decision or on an abstention.
How would an equipment team engage?
Through the field program, as an integration conversation rather than a purchase. The useful early exchange is about interface assumptions, capture geometry, and operating constraints on real machines.
What is most valuable at this stage is the constraint list an equipment team already has: available compute on the machine, the geometry and speed of a working pass, the coordinate and format conventions their existing workflow assumes, and what their systems do today when an input is ambiguous. That is what shapes an interface worth building against.
Inputs, outputs, limitations, and evidence status
Inputs
- A decision record produced within a stated validation scope.
- Georeferenced field geometry and an explicit coordinate reference system.
- Capture geometry and, where relevant, the machine's operating parameters.
- The equipment team's compute, latency, and format constraints.
Outputs
- A treatment zone expressing where a treat decision applies.
- References to the decision, model version, and threshold policy version.
- A validity window and the operating limits attached to the decision.
- Abstention areas marked distinctly from no-treat areas.
Limitations
- Swathmark does not build treatment machinery and does not control equipment.
- The interface conveys a decision; it does not authorize or perform an application.
- Edge deployment constrains model size, compute, and retainable evidence.
- A zone outside its validity window is not a usable input.
Evidence status
- No equipment platform has been integrated, tested, or certified.
- No compatibility, conformance, or standards claim is made.
- No equipment manufacturer, dealer, or platform is named as a partner or customer.
- Interface formats and versions will be agreed per integration and are not published.
Terms used on this page
Definitions are shared site-wide. The full list is in the glossary.
- OEM integration
- The defined software interface through which an equipment manufacturer consumes a decision: geometry and coordinate conventions, the edge and cloud boundary, export formats, and the operating limits attached to the decision.
- Treatment zone
- A georeferenced area expressing where a treat decision applies, together with the decision reference, policy version, and validity window. It is a representation of a decision, not an instruction to a machine.
- Edge inference
- Running a model on hardware in the field — on a vehicle, implement, or handheld device — rather than sending imagery to a server first. Edge inference constrains model size and compute budget and changes what evidence can be retained.
- Spot spraying
- Applying product only to the locations that require it rather than across a whole field. Spot spraying is an equipment capability; whether a given location requires treatment is a decision question.
- Precision agriculture
- The practice of varying field operations by location instead of treating a field uniformly, using positioning, sensing, and machine control. Site-specific pest treatment is a precision-agriculture application that depends on a trustworthy decision, not only on capable machinery.
Related topics
- Pest decision platformHow a pest decision platform differs from pest detection: the verified inputs it requires, the treat, no-treat, or abstain outputs it produces, and who uses it.
- Decision traceabilityWhat decision traceability means in practice: the provenance, model and policy versions, review history, and decision records needed to reconstruct a treatment decision.
- Uncertainty and abstentionWhy an agricultural decision system must be able to abstain, when it should, how uncertainty differs from abstention, and what happens after an abstain result.
Swathmark does not provide agronomic, pesticide-label, or treatment advice. Thresholds, product selection, and label compliance remain the responsibility of the grower and their advisers. Scope of use.
