Platform

Pest decision platform: from detection to a decision that can be defended

Definition

A pest decision platform is a system that converts verified field observations into an explicit pest-management decision and retains the evidence behind it. It combines registered image capture, provenance and quality verification, versioned detection models, biological threshold policies, and uncertainty handling to produce a treat, no-treat, or abstain outcome that can be audited after the event.

This page describes the intended workflow end to end: what has to go in, what comes out, who it is for, and where the boundaries sit.

Published: August 6, 2026
Last updated: August 8, 2026

What is a pest decision platform?

It is the layer between sensing and action. Where detection software reports what a model saw, a decision platform determines whether what was seen justifies action, records why, and expresses the result in a form another system can consume.

The distinction matters because the two failure modes are different. A detector fails by mislabeling an image. A decision platform fails by acting on a mislabeled image, on an unverifiable image, or outside the conditions its models were evaluated for — and by leaving no record that would let anyone find out.

What do growers and crop consultants call this?

They mostly do not call it anything. Practitioners talk about scouting a field, reading a threshold, and deciding whether to spray. "Pest decision platform" is our term for the software layer underneath that existing practice, not a new activity.

The vocabulary gap is worth stating plainly, because it is the source of most confusion in conversations about this product. A grower or crop consultant already makes threshold-based treatment decisions every season; the work is scouting, judging pest pressure and life stage, comparing that against an action threshold from their extension programme or adviser, and choosing whether and where to apply. None of that is new, and none of it needs software to exist.

The same activity, in two vocabularies
What practitioners say	What this site calls it
Scouting a field	Registered capture and verification
Reading a threshold	Evaluating a versioned threshold policy
Deciding whether to spray	A treat, no-treat or abstain decision
Where to spray	A machine-readable treatment zone
Not sure, come back and look again	Abstention with a recorded reason
Keeping spray records	Decision traceability

What changes is the input and the record, not the decision. Scouting is sampled, skilled and time-limited; imagery is broader in coverage and repeatable but blind to what the frame does not contain. And a scouting note in a spiral binder cannot be re-read against a revised threshold three seasons later, whereas a decision record can.

Swathmark does not provide agronomic, pesticide-label, or treatment advice, and publishes no threshold values. Which threshold applies, and what product to use if a treatment is warranted, stays with the grower and their advisers.

How is a pest decision platform different from pest detection software?

Detection estimates what is in imagery. A decision additionally requires verified inputs, biological thresholds, uncertainty handling, and traceability — four things a detector does not supply and is not accountable for.

Detection and decision compared
Question	Detection software	Decision platform
Is the input trustworthy?	Assumed. The model runs on whatever it is given.	Verified before inference; failures return an invalid-input result.
Does the observation matter?	Out of scope.	Evaluated against a versioned biological threshold policy.
What if the evidence is weak?	Returns a low confidence score.	Abstains explicitly and records why.
Can the result be reconstructed?	Depends on the caller's logging.	A decision record retains assets, versions, and review history.
What does the output drive?	A label or a bounding box.	A georeferenced treatment zone with stated operating limits.

What decisions can the platform produce?

Three outcomes, and only for an area and time within the stated operating and validation boundaries: treat, no-treat, or abstain. Abstain is a first-class result, not an error path.

The three intended outcomes
Outcome	Meaning	What is retained
Treat	Evidence and policy together indicate action is warranted for a defined area.	Zone geometry, policy version, model version, source assets, uncertainty.
No-treat	Evidence is sufficient and the policy is not met for that area.	The same record. A no-treat decision is evidence too.
Abstain	Evidence or operating conditions are insufficient to decide either way.	The reason for abstaining, and what would resolve it.

A decision is scoped to an area and a time. It does not carry forward: conditions and pest development change, and a stale decision is an unsafe one.

What inputs does a treatment decision require?

Imagery alone is not enough. A decision needs registered capture context, confirmed data rights, known calibration, measured image quality, biological context including life stage and crop stage, and an applicable threshold policy version.

1Registration: which farm, field, crop, and capture session the imagery belongs to.
2Rights: what the imagery may be used for, recorded as an enforceable state rather than an understanding.
3Quality: measured focus, exposure, motion blur, occlusion, and resolution at the target.
4Calibration: the camera calibration that lets image positions map to field geometry.
5Biology: apparent life stage, density within the imaged area, and crop growth stage.
6Policy: the threshold policy version applicable to that crop, pest, and stage.

If any of these is missing the platform is designed to abstain rather than to substitute a default. A decision produced from an incomplete input set is indistinguishable, downstream, from one produced properly — which is precisely why it must not be produced.

Who uses a pest decision platform?

Four groups with different needs: farm operations deciding where to act, research and extension specialists validating the biology, capture and sensing teams supplying imagery, and equipment teams consuming the output.

Intended users and what each needs from the platform
User	What they need
Commercial specialty-crop farms	A defensible answer about where action is and is not warranted, and a record of it.
Research stations and extension specialists	Transparent biology, versioned policies, and datasets whose lineage can be examined.
Entomologists and agronomists	Life-stage detail, adjudication of ambiguous labels, and honest uncertainty.
Field-capture and sensing teams	Explicit capture requirements and immediate feedback on invalid inputs.
Agricultural equipment and OEM teams	A stable interface, machine-readable zones, and stated operating limits.

What does the platform output?

A decision record and, where the outcome is treat, a georeferenced treatment zone expressing where the decision applies. Both are machine-readable and both carry the versions and limits that produced them.

A treatment zone is a representation of a decision, not a command to a machine. What an implement does with it — whether it can act at that resolution, at that speed, with that product — is an equipment question governed by the integration interface and by the operator.

What is not part of the platform?

It is not treatment machinery, not a farm-management dashboard, and not a source of agronomic or pesticide-label advice. It does not select products, set economic thresholds, or authorize an application.

No dedicated treatment machinery. Swathmark builds the decision layer; application equipment is third-party.
No product selection, rate calculation, or label interpretation.
No agronomic threshold authorship — thresholds come from agronomic authorities and the grower's own program.
No general farm-management, accounting, or compliance-reporting function.
No autonomous action. A decision is an output that people and equipment act on, or decline to.

Inputs, outputs, limitations, and evidence status

Inputs

Registered capture sessions with farm, field, crop, rig, and geometry recorded.
Confirmed data rights covering analysis and, separately, training use.
Measured image quality and known camera calibration.
Biological context: apparent life stage, density in the imaged area, crop stage.
An applicable, versioned biological threshold policy.

Outputs

An explicit treat, no-treat, or abstain outcome for a defined area and time.
A machine-readable decision record with every version that contributed to it.
A georeferenced treatment zone where the outcome is treat.
Stated uncertainty and the operating limits attached to the decision.
A recorded reason wherever the platform abstains.

Limitations

Decisions are bounded by the stated crop, pest, and validation scope.
An incomplete or unverifiable input set yields abstention, not a best guess.
A decision applies to a defined area and time and does not carry forward.
The platform does not act; equipment and operators do.
No agronomic, pesticide-label, or treatment advice is provided.

Evidence status

No performance, accuracy, savings, chemical-reduction, or yield figures are published.
No customer, acreage, field, or treatment counts are published.
Equipment compatibility and OEM integration are not yet demonstrated with a named platform.
The first validation domain is Colorado potato beetle in potatoes.

Terms used on this page

Definitions are shared site-wide. The full list is in the glossary.

Treatment decision: An explicit output stating treat, no-treat, or abstain for a defined area of a field at a defined time, together with the evidence and policy version that produced it. A detection is not a treatment decision.
Abstention: An explicit output declining to produce a treat or no-treat decision because the evidence or the operating conditions are insufficient. Abstention is a designed result, not an error and not a silent default.
Threshold policy: The reviewable, versioned rule set that turns biological observations into a management decision: which life stages count, how density or defoliation is assessed, which crop stage applies, and what uncertainty is tolerated. Held separately from model code.
Decision record: The durable, machine-readable artefact retained for each decision: source assets, rights status, capture context, model and calibration versions, policy version, uncertainty, review history, outcome, and export. It is the object an audit reads.
Crop scouting: Structured in-field inspection of a crop to estimate pest presence, life stage, and pressure, typically by sampling plants along a route. Scouting is skilled, sampled, and time-limited — which is why its coverage and repeatability vary.
Crop consultant: An independent adviser who scouts fields and recommends management actions to a grower. Often the person who actually shapes a treatment decision, which makes them distinct from both the grower and an extension specialist — and a distinct audience for a decision platform.
Economic threshold: The pest level at which intervention is considered justified because expected crop loss outweighs the cost of acting. Economic thresholds are set by agronomic authorities and local programs, not by a software vendor.