Samples Non Detects

Include non-detects

A checkbox in the Samples filter panel. Ticked by default. It decides whether the answer counts clean results as well as contaminated ones, and it moves the numbers more than any other filter on the page, because non-detects are most of the data.

What a non-detect is

A non-detect is a measurement, not a gap. The lab tested for the substance and found less than it is able to see. The record carries the detection limits instead of a value:

	A detect	A non-detect
coso:measurementUnit	NanoGM-PER-L	absent
numeric value	0.632	absent
methodDetectionLimit	0.486 ng/L	0.449 ng/L
reportingLimit	1.93 ng/L	1.78 ng/L
typed coso:NonDetectQuantityValue	no	yes

The missing unit is the important row. There is nothing to put a unit on, because no quantity was measured. That absence used to delete the record from every query. See Debugging, 2026-09-17.

How common they are, measured on federation:

Scope	Detects	Non-detects	Non-detect share
York County, all substances	7,110	16,420	70%
Maine, all substances	134,572	453,476	77%

What the checkbox does

It filters at two levels, and both matter.

1. Which sample points appear on the map. A point survives only if it has at least one detection left.
2. Which rows appear inside a popup. Points that survive still lose their non-detect rows.

York County, no substance selected:

	Ticked (default)	Unticked
Sample points on the map	451	367
Observations in popups	23,530	7,110
Points with at least one detect	451 includes all 367	367
Points where everything came back clean	70 shown	dropped
Points with no measurements in the graph	14 shown	dropped
Non-detect readings	16,420 shown	none shown

The second row is the one people miss. The 367 surviving points hold 21,822 observations between them, but unticking shows only the 7,110 detects among those. A well with 30 readings of which 2 were detections shows 2 rows, not 30.

With a substance selected the same thing happens inside that substance. York County PFOS: 361 points ticked, 230 unticked.

What a dot on the map means

This is the part worth saying out loud, because the two readings of the map are different claims:

• Ticked: a dot means someone sampled here.
• Unticked: a dot means PFAS was found here.

The misleading part is the absence of a dot. With the box unticked, 70 York County locations that were tested and came back clean look exactly like ground nobody ever visited. For a contamination map that is usually the wrong default, which is why the box starts ticked.

Every sample point, classified

Mutually exclusive, York County, all substances:

State	Points	In the answer when unticked?
No measurements in the graph at all	14	no
Tested, everything was a non-detect	70	no
Mixed: some detects, some non-detects	366	yes
Only detects	1	yes
Total	451	367

The 14 with no measurements are real registered locations, not bad rows. They are all soil sampling points (SB-*, SS-*, YO-* in EGAD) with coordinates, a parent site and a feature type of SOIL AUGER, whose lab results are not in the graph. At their parent site every drinking-water and monitoring well has 24 to 30 observations and every soil point has zero, so the gap is systematic rather than random. They currently render with an empty popup.

There is a fourth possible result state in the code, "non-quantified", a result that is neither a number nor typed as a non-detect. It is defensive: zero occurrences in York County and zero across Maine.

Opening the filter panel changes nothing

The checkbox only exists once + Add Filters is expanded, but expanding it is a display-only action:

<button className="btn-link" onClick={() => setShowFilters(true)}>
  + Add Filters
</button>

setShowFilters is local component state and never touches the question. The box draws as ticked because it reads value?.includeNondetects !== false, and with no filters set at all that is true. The emitted SPARQL is byte-identical before and after, and Apply stays disabled.

The SPARQL

Ticked, with no other sample filter, the whole samples block is two lines. Nothing asks about measurements, so nothing can filter on them:

?spC rdf:type coso:SamplePoint ;
     spatial:connectedTo ?s2target .

Unticked adds the result chain and one explicit filter:

?observationC coso:observedAtSamplePoint ?spC .
?observationC coso:hasResult ?resultC .
OPTIONAL { ?resultC qudt:quantityValue/qudt:numericValue ?numericResultC }
OPTIONAL { ?resultC qudt:quantityValue ?_qvC .
           ?_qvC rdf:type coso:NonDetectQuantityValue .
           BIND(true AS ?nonDetectC) }
BIND(COALESCE(STR(?numericResultC),
       IF(BOUND(?nonDetectC), "non-detect", "non-quantified")) AS ?result_valueC)
FILTER(BOUND(?numericResultC))

FILTER(BOUND(?numericResultC)) is the exclusion. Note it is written as an explicit filter rather than achieved by requiring a field non-detects lack, which is what the old code did by accident.

Two details that bite anyone writing these by hand:

• Do not use coso:measurementValue. It is multi-valued on non-detects, returning both "non-detect" and "non-quantified", which duplicates every non-detect row and makes MAX() return a string. The app derives its displayed value from qudt:quantityValue instead.
• Do not require coso:measurementUnit unless a concentration range needs it. Requiring it is indistinguishable from unticking this box, except silent.

If the numbers look wrong

• Counts jumped up after a deploy. Expected. Before 2026-09-17 the checkbox did nothing in either position: a required unit join removed non-detects regardless. Any figure taken off the app before that understates PFAS presence.
• A popup is empty. Likely one of the 14 points with no measurements in the graph.
• Unticking removed more than expected. It removes readings, not just points. Check the popup row counts, not only the dot count.

In the code

SampleFilters.tsx renders the checkbox. fusedQueries.ts decides which joins each filter needs, in sampleJoinsNeeded. samples.ts holds needsUnitJoin and resultValueClauses. scripts/check-query-joins.mts asserts the rule across all 108 question shapes in CI.