One address, and it is read by a person

Reporting a wrong identification

This is the most useful message we get and the one most often sent in a form we cannot act on. "The app got my plant wrong" is a report that a photograph exists; it is not a report anybody can fix. What turns it into one is three things: the photograph itself, what the app said, and what you believe it actually is. The third is the important one even when you are not certain — "I think this is bindweed, not morning glory" tells us which pair the model is confusing, and confusable pairs are the unit of work.

Two details that sound minor and are not. Roughly where you were, to the level of a region rather than an address: a great many misidentifications are a model proposing a species that does not occur within a thousand miles of you, and location alone diagnoses them. And what you could see that the photograph does not show — a smell, a milky sap, a texture, whether the stem was square. Those are the characters a camera loses, and they are usually the ones that would have settled it.

Corrections to the free keys

The 23 browser keys hold 749 hand-written entries, and the diagnostic sentence attached to each is a judgement about what actually separates it from its look-alikes. Judgements can be wrong, ranges shift, and a character that works in one part of a species' range fails in another. If you know a key is misleading — particularly if you work with the group professionally — that correction is worth more than anything else on this page.

A citation helps and is not required. If you tell us that a character does not hold and why, we will go and check it against a primary source ourselves; what we will not do is quietly leave it. Corrections are made in the source data and appear at the next build, and the "Updated" date on each key says which build you are looking at.

Errors in the open dataset

The open field data file matches key entries to GBIF, iNaturalist, Macrostrat and the Paleobiology Database automatically. Automatic name matching produces a predictable class of error — a name that resolves to a homonym in a different kingdom, a common name mapped to the wrong species, a genus-level match presented alongside species-level ones. If you find one, the row identifier and what it should be is enough; it gets corrected in the next snapshot, dated 2026-09-04 at the time of writing.

Licensing questions about the dataset go to the same address. The short answer is that the compilation is CC BY 4.0 and you may use it commercially with a link back, while the underlying records carry their own sources' terms, which are listed on the data page.

The app: billing, accounts and deletion

Purchases and subscriptions are handled by Apple and Google, not by us, which means refunds and cancellations run through the store you bought from — we cannot process either, and any page that tells you otherwise is wrong. Apple's route is Report a Problem in your purchase history; Google's is the Play Store subscriptions screen. If you write to us about a charge we will point you at the right one, and we would rather you asked than assumed.

For account deletion, data access requests, or anything else covered by the privacy policy, write to the same address and say what you want done. Those are handled directly rather than through a portal.

What this address cannot do

Medical and veterinary questions. If somebody has eaten something, been bitten, or has a reaction, contact a poison centre or a doctor — in the United States, Poison Control on 1-800-222-1222, which is free and staffed around the clock. Email is the wrong medium for anything time-critical, and we are the wrong people.

Valuations. We are asked regularly what a coin, a card or a mineral specimen is worth. We do not know, and neither does any photograph: condition, provenance and authenticity set the price, and all three need the object in hand.

Legal identification for a permit, a survey or a court. Anything with a formal consequence needs a qualified person who will put their name to it. These keys are not that, and say so.

Suggesting a key that does not exist yet

The 23 keys published so far were chosen because the search results for those questions were unusually bad — listicles without a filter, image boards, PDFs from 2009. That is the criterion, and suggestions that fit it are genuinely welcome. What makes a good candidate is a question with a bounded answer: a group of thirty-odd things that people routinely confuse with each other, separated by characters an ordinary person can observe. "Identify any insect" is not that. "Which wasp is nesting under my eaves" is.

If you have the expertise as well as the question, say so — the slowest part of building a key is not the software, it is establishing which characters actually hold across a species' range, and that is exactly the knowledge that lives with people who work with a group rather than in the literature about it.

Partnerships, integrations and reuse

The dataset is openly licensed, so you do not need permission to use it — a link back is the whole requirement. If you are building something on top of it and want the schema to stay stable, or need a column that is not there, that is worth a message: the file is generated, so adding a field is usually cheap, and knowing somebody depends on a column is what stops it being renamed.

For anything involving the app — bundling, an institutional deployment, a research collaboration — the same address reaches the person who can actually decide. There is no sales team to route around.

Security and privacy reports

If you have found a vulnerability in the app or on this site, write to the same address with enough detail to reproduce it, and please give us a reasonable window before publishing. We will confirm receipt, tell you what we found, and say when it is fixed. Reports are read by the person who wrote the code, so there is no triage queue between you and somebody who can act.

Press and research enquiries

For anything written about the site or the dataset, the two things most often needed are the methodology and the licence, and both are on the open field data page in full rather than summarised. The short version for a citation: the compilation is CC BY 4.0 and may be quoted or redistributed with a link, while the occurrence counts belong to GBIF, iNaturalist, Macrostrat and the Paleobiology Database and carry their own terms, which travel with the numbers.

If you are a researcher who would find a different cut of the data useful — a column that is not there, a rank filter applied differently, the raw month histograms in a different shape — that is usually a small change to a generation script rather than a project, and it is worth asking. The file is regenerated rather than maintained by hand, so adding a field costs very little.

What happens to a message

There is no support tier and no routing. A message lands in one inbox and is read by the person who wrote the page you are writing about, usually the same day and always within two working days. That is the advantage of being small, and the disadvantage is the obvious one: there is nobody covering for a week off, so occasionally a reply is slower than it should be.

What we cannot promise is that everything gets built. Feature requests are read and a good number of them shape what happens next, but a reply saying "noted, not soon" is an honest answer rather than a brush-off, and you will get that rather than silence. Corrections are different: a factual error in a key or the dataset gets fixed, and the snapshot date on the page tells you when the fix went live.

One request in the other direction. If you write about an identification, please attach the photograph rather than describing it. A description is filtered through what you already think it is, which is precisely the thing under question, and half the useful information — the background, the scale, the light — is gone by the time it reaches a sentence.

Who you are writing to

Orvik is published by Vast Flow, LLP. The reference side of the site — the keys, the comparisons, the dataset — is compiled by one person, and the about page says who, what they are qualified in, and what they are not. That page also carries the honest version of what the app can and cannot do, which is worth reading before writing if your question is about accuracy.

Messages get a reply within two working days. If yours does not, it is worth sending again rather than assuming it was ignored — a small number of messages have been lost to spam filtering in both directions.