NaturecodeNon-profit · Sweden(01) The problem(02) Initiatives(03) Research → reporting(04) Partnerships

Conservation runs on numbers nobody checked.

We are a non-profit holding a measurement standard for environmental change: how much of a reported difference is real, and where no defensible answer exists. Our first initiative, the Naturecode Platform, puts it in the hands of the people doing the work — free.

The problem we exist for Our initiatives

The number on the datasheet is not the number in the field

Published, peer-reviewed, and almost never carried through into what a project reports to its funder.

0.791 → 0.414
Identifying birds from song
Mean average precision on clean recordings, against the same model on 286 hours of real soundscape. Conservation projects quote the first number.
95.6% → 68.7%
Identifying animals in camera-trap images
Accuracy at locations a model was trained on, against locations it was not. Species identification does not travel between places.
~90%
Of the variance between eDNA results is the laboratory
Absent one fixed protocol, which laboratory ran the sample explains more than the water does. There are still no proficiency providers for macrobial eDNA.
Refuted
Read counts as a measure of abundance
At equimolar input, metabarcoding read counts span three to four orders of magnitude. eDNA tells you a species was there. It does not tell you how much.

Why this matters beyond accuracy

Money follows reported change.

Restoration funding, biodiversity credits and protected-area decisions all rest on a reported difference between two surveys. When the measurement error is larger than the difference, the decision is being made on noise — and nobody involved is told.

What we do instead

State the threshold, or make no claim.

Every verdict carries the change it took to reach it, the method that produced the reading, and the number of occasions behind it. Where no defensible threshold exists, we say so rather than reporting a number that looks like one.

Initiatives1

A non-profit association registered in Sweden. No shareholders, and no incentive to report a larger change than the instruments support. One initiative is running; the others below are stated as intent, not as work already done.

Initiative 01 · Running

Naturecode Platform

The measurement infrastructure itself, at naturecode.ai. Per-site thresholds learned from each site's own repeat sampling rather than compared to a regional average, published method grades carrying the field accuracy and its citation, and 4 claims the system refuses to make. Free for conservation projects.

Open the platform

Why the platform is a project, not the mission

The mission is the standard, not the software.

A measurement standard outlives any one tool that implements it. The platform is how we make the standard usable today; the initiatives below are the rest of what holding one requires — and other organisations implementing the same standard would be a success, not a competitor.

What we do4

Four stages, in order. A report is only as defensible as the detection limit established before anyone went to the field.

01
Research — Establishing what each method can actually detect
The field accuracy of every method, with its citation rather than the manufacturer's figure. A method whose own error exceeds the effect being looked for cannot carry a claim, and we publish which ones those are. The largest open gap: no published minimal detectable change exists for eDNA metrics, and no documented noise floor for vegetation indices.
02
Monitoring — Protocols that survive contact with field conditions
Per-site repeat sampling designed so a site learns its own threshold rather than borrowing a regional average. Copernicus Sentinel-2 is free, full and open with commercial use explicit — ten-metre resolution every five days — and most projects do not know they already have it.
03
Analysis — Separating the signal from the instrument
The same statistical core the platform runs: successive-difference variance, a threshold learned from the site's own scatter, and an abstention where the occasions are too few to support a verdict. Roughly one taxon in five appears at a single time point and never again, so a threshold that clears analytical noise but not temporal turnover is not a threshold.
04
Reporting — Stating the uncertainty alongside the number
Every verdict carries the change it took to reach it, the method behind the reading, and the number of occasions. Where no defensible threshold exists we say so — 4 claims are refused outright, each published with the finding that forbids it, so the limits are checkable rather than asserted.

Partnerships3

We are early and say so. The platform is live; the field validation is not yet done, and that is exactly what partners are for.

fieldFree

Conservation projects

Use the platform at no cost. We need protocols tested against real field conditions, and the fastest way to find where ours break is for you to use them on a site we have never seen.

scienceOpen

Research institutions

Co-authored work on the detection limits that do not yet exist — a minimal detectable change for eDNA metrics, a noise floor for vegetation indices. Findings are published openly whether or not they suit us.

fundingAudit

Funders and assessors

We can tell you what a reported change would have to exceed before it means anything, given the methods a project actually used. No commercial incentive to widen a claim, because there are no shareholders to widen it for.

Open by default, with one exception

The exception is not a hedge — it is the one place openness does harm.

Open

Methods, grades, thresholds and refusals.

Anyone can check what we claim and what we decline to. A measurement standard that cannot be audited is not a standard — and a non-profit is the right shape for holding one, because there is no commercial reason to quietly widen a claim.

Not open

Where the threatened things are.

Precise locations of threatened species can be used to find them. Project data is private to its team by default, and location precision is reduced for taxa where disclosure is a risk. That decision belongs to the project, never to us.