Who compiles these keys, and what they are worth

Who writes this

Orvik is published by Vast Flow, LLP, a software company, and the reference side of it — the identification keys, the comparison pages, the open dataset — is compiled by Merey Tleugazin, who founded it. That is the whole masthead. There is no editorial board and no panel of consulting biologists, and inventing one would be the easiest possible thing to do on a page like this, which is exactly why it is worth stating plainly that there is not one.

I am not a biologist, a mycologist, a herpetologist or a geologist. What I am is a person who builds software, reads primary sources carefully, and has spent a long time on the specific problem of turning what an expert knows into something a non-expert can actually use in a field. Every claim on this site is either traceable to a named source, computed from a public database, or clearly marked as a judgement — and where the answer is "an app cannot tell you this", the page says so instead of guessing.

That framing is not modesty for its own sake. It changes what you should do with the output. A key here narrows a list; it does not certify a specimen. If the consequence of being wrong is a hospital visit — mushrooms, wild berries, snakes, anything a child might have eaten — the key is a tool for describing what you are looking at to somebody qualified, not a substitute for asking them.

What Orvik actually is

Three things, and they work differently from each other.

The app reads a photograph and proposes a name, with the plausible look-alikes listed underneath rather than hidden. It runs on iOS and Android and is free to download. What it does is pattern recognition on an image, which means it is good at the things that look distinctive in a photograph and unreliable at the things that do not — and a great many identifications turn on a character a camera cannot capture: a smell, a texture, whether the underside bruises blue, what the roots look like. We do not publish an accuracy percentage, because a single number across every subject would be close to meaningless: the same model that is dependable on a common garden flower is guessing on a small brown mushroom, and one figure covering both would mislead in the direction that matters most.

The free browser keys — 28 of them, holding 749 entries — are the opposite approach. Instead of a photograph, they ask you what you can see: the shape of the web, the colour of the spore print, whether a knife scratches it. You narrow, and the list shrinks. They need no account, they cost nothing, and every entry is rendered into the page source, so the whole catalogue is readable even with scripting switched off. Each entry carries the diagnostic sentence that separates it from its nearest look-alike, because that sentence is the part that does the work.

The reference layer is 255 articles across 9 sections, 28 comparison and alternatives pages, and one open dataset. The comparison pages carry no price anybody typed: every figure comes from a dated App Store snapshot, and the build fails if a currency amount appears in an editorial file. It is the same rule the field data lives under.

Where the numbers come from

Every count on this site is computed at build time from a source, not written by hand, and this is the single most load-bearing decision behind the whole thing. A number a person types is a number that goes quietly stale: it stays on the page long after it stopped being true, and nothing on earth notices.

For the living things in the keys, that source is GBIF and iNaturalist: 502 entries carry a GBIF taxon key, and the 420 of those that resolve to a single species stand on 634,096,345 georeferenced occurrence records between them, the oldest collected in 1877. Minerals get their formula, Mohs range and crystal system from Macrostrat; fossils get their first and last appearance from the Paleobiology Database. All four are free, public and institutional, and all four are credited on every page that uses them.

The whole snapshot is published rather than kept: it is downloadable as a single CSV on the open field data page under CC BY 4.0, so anybody can check a figure here against the source, or use the join for something else entirely. Data that cannot be checked is an assertion, and assertions are cheap.

What this site will not tell you

A short list, because the omissions matter more than the features.

• Whether something is safe to eat. No key here, and no photograph fed to any app, can carry that decision. Several deadly species are close visual matches for common edible ones, and the fatal cases are usually confident people rather than careless ones.
• Whether a bite or a rash is dangerous. The keys flag which species are medically significant and name the poison centre. They do not diagnose, and a symptom is a reason to call somebody, not to browse.
• What something is worth. The collectibles keys describe what a thing is. Condition, provenance and authenticity set the price, and none of the three survives a photograph.
• A confident answer where the evidence is thin. Where the record is too sparse to support a claim, the tables print a dash rather than a number, and the prose says the sample is too small. That is a deliberate choice about which kind of error to make.

How a key gets built

The order of work matters, so it is worth describing. A key starts from the question a person actually asks, not from a taxonomy — "what spider is this" rather than "Araneae of North America". That question sets the boundary: the entries are the things somebody in that situation could plausibly be looking at, which is usually thirty or so, and almost never the complete regional fauna. A key with three hundred entries is a checklist, and a checklist does not narrow anything.

Then the filters. Each one has to be something you can observe without equipment and without expertise: the shape of a web, the number of legs, whether the underside is pale, how big it is against a coin. A filter that requires a hand lens or a spore print is included only when nothing else will separate the group, and then the page says so. The single most common failure in published keys is a character that is obvious to the person who wrote it and invisible to everyone else.

Last comes the sentence that does the work. Every entry carries one line describing the thing that separates it from its nearest look-alike — not a description of the species, which any reference has, but the specific difference that resolves the specific confusion. Where two things genuinely cannot be separated on what you can see, the entry says that too. An honest dead end is more useful than a confident wrong turn, and it is the point at which the correct advice is to photograph it and ask somebody.

Only then do the numbers get attached. The occurrence counts, the seasonal curves, the hardness ranges are computed afterwards from the public databases and joined onto the entries — they inform how a shortlist should be ordered, and they never determine what is in it. Prior probability is genuinely useful and genuinely not identification.

Why the keys are free, and why that is not a trick

The keys cost nothing, require no account, and carry no advertising. There is an app behind them and the pages link to it, so the arrangement is not charity — but it is also not the usual bargain, where a free tool is deliberately crippled until you pay. These keys are the complete thing. Every entry renders into the page source whether or not scripting is on, which means you can read the whole catalogue with JavaScript disabled, save the page, or print it.

That is a deliberate decision with a cost attached: it means the keys are trivially copyable, and some of them will be copied. The reason to do it anyway is that a reference nobody can verify is a reference nobody should trust, and hiding the data behind an interaction is the fastest way to make a page useless to anyone reading it carefully — including the search engines and language models that increasingly answer these questions on our behalf. If a model is going to summarise a page about spiders, it ought to be summarising one where the numbers are real and the sources are named.

When we get it wrong

With 570 automated name matches in the dataset and 749 hand-written entries in the keys, some of this is wrong. That is not false modesty; it is arithmetic. What matters is whether it gets fixed, so: if a diagnostic character is misleading, a range is out of date, a taxon has been resolved to the wrong species, or a key confidently splits two things that cannot actually be split that way, write to [email protected]. Corrections go into the next snapshot, and the date at the top of every page tells you which one you are looking at.

The same address takes press enquiries, licensing questions about the dataset, and anything about the app itself. It is read by the person who wrote this page.