The classifier model
A single recording can carry more than one opinion about what it is. Your detector's own on-device classifier writes its guess into more than one metadata field, and you can add a manual correction on top — Fledermap keeps every one of these as a separate claim rather than overwriting earlier ones, and then picks a single "current best" answer to actually show you.
Why claims from different sources coexist
The detector's GUANO metadata, its wamd metadata, and its filename
convention can all encode an auto-ID, and they don't always agree with
each other — encoding differences, or one field simply being staler than
another. Keeping all of them, tagged by where each came from, means a
disagreement is visible and diagnosable instead of silently picked for
you by whichever field happened to get read first.
How the "current best" answer is picked
When several sources have a claim, one specific order decides which one
you actually see: your own manual correction first, then the detector's
on-device manual correction, then its automatic GUANO, wamd, and
filename-based guesses, in that order.
There's one important exception to "the first source in that order always wins": an automatic classifier's own No ID is treated as if it hadn't answered at all, and Fledermap keeps looking further down the list for an actual answer. A manual No ID, on the other hand, always wins outright — a human choosing "I can't tell" is a deliberate judgment worth keeping, not the same kind of gap as a classifier shrugging.
No ID, Noise, Unidentified, and Unmapped species — four different things
These are easy to conflate, but they mean genuinely different things:
- No ID — a source is confident this recording contains a bat call, but isn't confident enough to name a species.
- Noise — a source is confident this recording does not contain a bat call at all.
- Unidentified (a map filter option, not a verdict) — no source has made any claim on this recording yet, of any kind.
- Unmapped species — a source gave a specific species code, but that code isn't one Fledermap has a taxon for.
That last case is deliberate, not a bug: a code Fledermap can't confidently map to a species resolves to no answer rather than a guess, and shows up as Unmapped species in the map's Taxon filter for you to review. A wrong mapping — confidently pointing at the wrong species — is a worse failure than an honest gap, because a wrong answer looks just as credible as a right one.