A detection is not an occurrence

Two statements that look the same and are not

A molecular detection asserts something narrow and defensible: this sequence was present in this sample.

An occurrence record asserts something much larger: this organism was at this place at this time.

Biodiversity databases store the second. Regulators consult the second. Project screening, listing decisions, rapid response and permitting all run on the second. The first, on its own, changes nothing.

Getting from one to the other is not a calculation. It is a judgment, and it has at least four parts.

How confident is confident enough? Two positive replicates out of three, or three out of three. Ten reads, or fifty. A cycle threshold of 38, or 40.

What replicate structure counts? Technical replicates test the reaction. Field replicates test the sampling. Temporal replicates test the moment. Reporting "three replicates" without saying which kind makes the number uninterpretable, and the three are routinely conflated.

What must the controls have done? A run without field blanks, extraction controls or a no-template control cannot rule out contamination — but a record with no controls is not automatically wrong, it is simply unverifiable. Whether that disqualifies it is a decision.

And what do you call it? Where a sequence maps to a species complex containing several genetically distinct lineages, naming one of them is a choice with consequences. Under-splitting is not a rare edge case; it is common enough that assays built for one lineage routinely miss another carrying the same name.

None of these have physically correct answers. They have chosen answers, and different choices produce different species lists from identical data.

These look like technical questions. They are policy questions wearing technical clothing.

The field already knows this, in pieces

In 2021, John Darling, Christopher Jerde and Adam Sepulveda published a short paper in Environmental DNA arguing that confusion over a single term was itself a significant hurdle to the adoption of eDNA methods.

Their distinction was between a false-positive test — a laboratory result produced in error — and a false-positive inference, which is a conclusion about a site that direct observation later contradicts. The first is a question about a reaction. The second is a question about the world.

That distinction is exactly where the detection-to-occurrence boundary sits. A laboratory can characterize its own error rate. It cannot characterize the inference, because the inference depends on decisions made outside the laboratory — about scale, about prior expectation, about what the detection is being used for.

The vocabulary was adopted quickly, because the field needed it. But naming the distinction did not resolve who gets to draw the line.

What this costs, specifically

The clearest illustration is a database most people outside invasive species management have never heard of.

The Nonindigenous Aquatic Species database, run by the US Geological Survey, is the reference source for aquatic invasive species occurrence across American agencies. If a detection is going to influence a management decision at a federal or state level, that is very often where it needs to land.

NAS has a pilot program for incorporating eDNA data using targeted approaches — qPCR assays for a named species. Standardization for metabarcoding was pushed down the road, on the grounds that it is difficult and complicated.

Both of those things are true. But consider why it is difficult.

A targeted assay produces something close to a yes or no about a species someone specifically asked about. Metabarcoding produces hundreds of taxa at varying confidence, from a reference library with uneven coverage, clustered by a threshold somebody chose. To accept that into an occurrence database you must first decide which of those hundreds crosses the line.

Nobody has decided. So the data does not go in.

Which means metabarcoding surveys that could inform invasive species management across the country produce results that stop at a PDF — not because the science failed, but because there is no agreed answer to a question nobody is authorized to answer.

Why nobody is authorized

This is the part that gets missed, and it is the reason the problem has persisted rather than been solved.

A federal agency cannot set a threshold that binds commercial laboratories. USGS can specify what it will accept into its own database, and that is meaningful. But it cannot tell a private laboratory in another state how to report to a different client, and it would not want to.

A laboratory cannot set a threshold that binds its competitors. It can publish its own and many do, transparently and well. But a threshold adopted by one provider and not the others does not make results comparable — it makes one provider's results internally consistent, which is a different and much smaller achievement.

A journal cannot do it. Peer review evaluates individual studies against their own stated methods. It does not harmonize across them.

And a standards body has not yet reached it. ISO published a standard for eDNA water sampling and preservation in 2026, and further standards covering extraction and reporting are in development. That work is necessary and welcome. None of it currently addresses the threshold question, because the threshold sits downstream of everything so far standardized.

So the decision falls into a gap. And a decision that falls into a gap does not stop being made — it gets made repeatedly, privately, by whoever is closest to the problem, and the answers do not match.

The uncomfortable implication

Every eDNA result currently in circulation embodies a threshold decision. Most reports do not state what it was. Almost none state who chose it or why.

That is not misconduct, and it is not sloppiness. It is what happens when a necessary decision has no owner. Individual practitioners have made sensible choices in the absence of a shared one, which is the correct behavior in the circumstances and also the reason the results cannot be pooled.

The consequence is that a large body of biological evidence — expensively collected, competently analyzed, scientifically sound — cannot be aggregated, cannot be compared across years, and largely cannot enter the systems where it would change anything.

What would actually fix it

Not a better algorithm. The algorithms are fine.

What is required is a body with standing to make the decision, and a specific set of properties that give it that standing.

It cannot have a commercial interest in the answer. A threshold set by anyone who sells sample analysis is a threshold their competitors will not accept, and they would be right not to.

It must include people who actually run this work. A committee of academics and regulators without practising laboratories in the room will set thresholds that cannot be met in production, and they will be ignored.

It must publish the reasoning, not just the number. A threshold that arrives without its justification cannot be challenged, and a threshold that cannot be challenged will not be trusted.

It must be able to change its mind in public. Reference coverage improves, methods change, and any threshold set today will be wrong within a decade. Versioning the decision, and keeping superseded versions published, is what separates a standard from an opinion.

And it must expect to hand the question over. The right long-term home for this is a formal standards body running a proper balloted process. Anything operating in the interim should be explicit that it is holding the question, not owning it.

Where we sit

Common Measure is building infrastructure that depends on this question being answered, and we are not the right party to answer it alone. That is why the specification we publish records which threshold was applied rather than prescribing one, and why the decision is being placed with an external committee on which we hold no vote.

We are aware how that sounds coming from a company. Judge it by the structure rather than the intent — and if the structure is insufficient, say so publicly, because the alternative to getting this right is the situation we already have.

The question worth asking

If you commissioned an eDNA survey in the last five years, you likely have a report with a species list on it.

Try to find out what threshold produced that list. Then try to find out who chose it.

For most reports in circulation, neither answer is recoverable. That is the problem, and it is not a scientific one.