Trust & provenance
How we prove a detection is what we say it is
A photo of an animal isn't evidence by itself. What makes it usable in a report, a claim, or an audit is everything around it: who or what identified the species, whether anyone checked, and whether the record for that night is even complete. That's what this page is about.
Most camera-trap tools show you a species label and stop there. If you're using that label in a sustainability report, a damage claim, or a permit application, "the AI said fox" isn't enough on its own — you need to know how sure it was, whether a person looked at it, and whether anything is missing from the record. Untamed Data tracks that alongside every detection, automatically, whether or not anyone ever asks for it.
What's recorded, per detection
- Which model made the call, and which version. Species recognition runs on a named model with a version number, stamped on the detection at the moment it was made. If the model is updated later, older detections still show what actually produced them — nothing is silently reattributed.
- Whether a person confirmed it, or the AI decided alone. Detections the model is unsure about go to a review queue; a person confirms or corrects them. That distinction — verified by a person, or AI-only — stays attached to the record, not just to a dashboard count.
- Whether the record for a period is actually complete. A camera that stopped uploading doesn't just go quiet — the gap is tracked and factored into effort calculations, so a dead battery in week three doesn't quietly inflate how active a site looks. Missing data is stated, not smoothed over.
Why it matters outside the platform
This isn't detail for its own sake. It's what a specific record needs to hold up once it leaves the dashboard:
- Sustainability reporting. Biodiversity impact reporting under frameworks like the EU's CSRD asks for auditable methodology, not just a number. Being able to show which detections were machine-only, which were human-verified, and what the survey effort actually covered is the difference between a defensible figure and one nobody can stand behind under review.
- Damage and compensation claims. Camera evidence is routinely used to support wildlife-damage claims. What holds up is a timestamped, unaltered record of presence — not a number someone could have edited after the fact. Camera footage proves presence; it doesn't by itself prove causation, and we don't claim otherwise.
- Government and conservation reporting. Programs feeding into Natura 2000 or similar reporting cycles need data whose provenance can be checked, not taken on faith. See how this connects to Camtrap DP export and Natura 2000 monitoring.
What this doesn't claim
We'd rather say this plainly than have someone assume too much. A confirmed detection means a person looked at that specific record and agreed with the label — it is not a forensic or legal certification, and we don't present it as one. Abundance figures (RAI) are a relative index built from independent events and survey effort, not a calibrated count of how many animals exist — useful for comparing activity across sites and seasons, not for stating a population number. And presence evidence is exactly that: proof an animal was there, not proof of what it did. Where a claim needs more than presence, that gap is real, and pretending otherwise would make the record less useful, not more.
Why we built it this way. A monitoring record that only looks credible gets challenged the first time someone actually checks it. One that states its own limits — what's verified, what's a proxy, what's missing — is the one that survives scrutiny. That's the same reasoning behind exporting in Camtrap DP, an open standard instead of a format we invented: provenance you can check is worth more than provenance you're asked to trust.
