Accuracy & Limits
Safety review: 3 August 2026 · Why a bite cannot identify the bug
Most identification tools tell you how confident they are with a number they cannot justify. Here is ours, measured, including the parts that are not good.
How we test
We score against 168 research-grade photographs from iNaturalist across 21 species — real phone snaps at bad angles, in bad light, identified by expert consensus rather than by us. The set is deliberately weighted towards cases where being wrong costs something: 48 photographs are of medically significant species, and 64 are of the harmless species those get confused with.
The exact photographs are pinned, so a change in the score means a change in the identifier and not a change in the test.
What our confidence levels actually mean
When we say “high confidence”, this is how often we were right about the genus:
| We said | Photos | Correct |
|---|---|---|
| High confidence | 46 | 95.7% |
| Medium confidence | 88 | 76.1% |
| Low confidence | 34 | 32.4% |
That gradient is the point. A low-confidence answer from us is genuinely unreliable, and we would rather tell you that than dress it up. If you see “low”, treat the answer as a starting point and not a conclusion.
What it still gets wrong
Across the full set we reach the most precise rank a photograph honestly supports 72.6% of the time. The failures are not random. They are almost entirely nondescript brown spiders — mature males in particular — where the character that settles the answer is the eye arrangement, and the photograph simply does not resolve it.
No identifier can read information that is not in the image. When we hit that wall we say so rather than guessing, which is why some answers come back as “we can’t rule this out” instead of a species name.
Why we sometimes refuse to say “harmless”
If the answer lands in a group that recluses, widows or bark scorpions are routinely mistaken for, and the photograph does not resolve the character that would rule them out, we will not publish “harmless”. You will get an explicit “we can’t rule this out” instead.
This is a deliberate asymmetry. Being told a harmless spider might not be costs you a moment of caution. Being told a brown recluse is a cellar spider costs considerably more.
What we will never do
- Give you a confidence percentage. We publish three bands and the measured accuracy of each, because a decimal implies a precision nobody in this field actually has.
- Give medical advice. We identify. If you have been bitten or stung and symptoms develop, see a doctor.
- Recommend pesticides. This is an identification site, not extermination advice.
How to read a result
Every answer lists the features it used. Check them against the animal in front of you — if they do not match what you can see, the answer is probably wrong, and the species pages exist so you can compare against the lookalikes yourself. See also how we identify.