16 November 2026

What is a risk-based control sample, and why full-parcel checks are replacing it

A patchwork of green and golden fields across rolling countryside under a blue sky with hills in the distance.

A control sample is how a paying agency decides which claims actually get looked at. Nobody inspects every parcel in the register by walking a field boundary with a GPS rod, so the system picks a subset, checks that subset hard, and treats the result as representative of everything else. That's the whole idea behind risk-based selection under CAP direct payments, and it's been the backbone of IACS control systems for decades.

How risk-based selection works

The regulation sets a floor on how many claims have to be checked each year, a minimum inspection rate the paying agency can't go below. Above that floor, the agency doesn't pick parcels at random. It runs a risk analysis: claims that changed crop type year over year, claims near a land-use boundary that's shifted, claims flagged in a previous OTSC, claims from beneficiaries with a prior non-compliance. Those get weighted up. A clean, stable, long-standing claim with no history of issues gets weighted down, and most years it never gets a visit.

That's a deliberate design choice built around a real constraint: physical inspection is expensive and slow, so the agency spends that budget where it's statistically likely to find something. The control sample is the agency's best guess at where the risk is concentrated, built from last year's data and whatever flags the LPIS and cross-compliance checks threw up.

The problem is what happens to everything outside the sample. A parcel that was never flagged gets no scrutiny at all, for one season or several in a row, purely because the model that built the sample didn't see a reason to look. It's a structural blind spot in how the sample was built, one that exists even when the risk model is doing exactly what it was designed to do.

Why full-parcel checks change the equation

Wide-swath multispectral satellite passes don't have that constraint, because they don't need to pick a subset. A sensor pass over a region captures every claimed parcel in its path on the same cadence, whether that parcel was flagged last year or not. Run that across a claim season and you get a season-long read on what grew on every parcel, not just the ones the risk model singled out.

That changes what "control" means. Instead of extrapolating from a sampled few percent and hoping the sample generalizes to the rest of the register, the agency gets a parcel-by-parcel comparison between the declared crop and what the imagery shows across the season. A mismatch between declared crop and observed crop puts a parcel on the flag list; a match keeps it off. No extrapolation step, because nothing was skipped in the first place.

It doesn't replace the minimum inspection rate set in the regulation, and it doesn't replace a field check for a parcel that needs a boundary walked or a ground-truth photo taken. What it replaces is the logic of only ever looking where the risk model told you to look. Subsidy Monitoring runs that season-long pass and hands back a flag list, map layer plus CSV, covering the full claimed register instead of the sampled slice of it. You can see how that full-parcel coverage is built on the Subsidy Monitoring overview.

What this means for selection going forward

Risk-based sampling isn't going away as a regulatory requirement, and it shouldn't. But once a paying agency has a full-parcel satellite check running underneath it, the sample stops being the only place anomalies can surface. A parcel outside this year's risk model that shows a harvest signal inconsistent with its declared crop still shows up on the flag list, because the check didn't need to select it first to see it.

That's the practical shift: selection used to be the thing that decided what got checked at all. With a season-long pass over every parcel, selection becomes the thing that decides what gets a field visit after the imagery has already flagged it. If your control sample has been carrying weight it was never built to carry, it's worth seeing what a full-parcel pass turns up before next season's sample gets built.

Get started

Back to the notes