Harbour seal, reviewed
Hind flippers are visible in the reviewed frame; this is the confirmed marine mammal example from the run.
- Source
- MaST Center Aquarium video bridge
- Evidence
- reviewed frame plus high-quality bundled video clip
Featured evidence
Reviewed solid border. Model output dashed.
Ocean sunfish, Mola sp.
A real and unusual animal in Puget Sound. YOLO-Fish called fish with high confidence; the open-vocabulary pass called harbor seal.
- Source
- MaST Center Aquarium video bridge
- Evidence
- reviewed frame plus high-quality bundled video clip
Black bears at Sunrise Mountain
A larger bear and smaller bear are visible together in the lower meadow on 2026-08-21 after source-window review.
- Source
- NPS Sunrise Mountain webcam
- Evidence
- annotated full frame plus neighboring frame sequence
Large fish, reviewed false marine-mammal label
A visible fish with dorsal fin was labeled harbor seal by the open-vocabulary pass.
- Source
- MaST Center Aquarium video bridge
- Evidence
- reviewed frame plus high-quality bundled video clip
Deer at Sunrise Mountain
MegaDetector surfaced the frame; the deer label comes from human review of the full-frame context.
- Source
- NPS Sunrise Mountain webcam
- Evidence
- annotated frame, source frame, and neighboring frame sequence
See the other 11 featured findings
Domain
Review status
Orcasound Lab SRKW encounter signal
A ten-minute Southern Resident bout at Orcasound Lab on 2026-08-26, 21:24 to 21:35 UTC. Fifteen consecutive 30-second windows are published, every one of them classified Southern Resident at 0.99 or above. The window shown here is the densest, with 13 positive segments. 29 of 30 windows in the local encounter cluster called SRKW above the site floor.
- Source
- Orcasound Lab hydrophone
- Evidence
- report evidence
First calls of the Orcasound Lab encounter
The opening window of the 2026-08-26 encounter at 21:24:44 UTC, three positive segments. The bout builds from here. Ecotype 1.00 Southern Resident, OrcaHello 0.97.
- Source
- Orcasound Lab hydrophone
- Evidence
- report evidence
Last calls of the Orcasound Lab encounter
The closing window at 21:34:45 UTC, ten minutes after the first. Nothing above the site floor follows it that day. Ecotype 1.00 Southern Resident, OrcaHello 0.93.
- Source
- Orcasound Lab hydrophone
- Evidence
- report evidence
North San Juan Channel Bigg's encounter signal
The August hydrophone analysis found a clustered Bigg's-class ecotype signal at North San Juan Channel that survived the site-specific floor and independent-record comparison. Six full 30-second windows from the 2026-08-15 cluster are now playable.
- Source
- Orcasound North San Juan Channel hydrophone
- Evidence
- ecotype rates, site-specific false-positive floor, daily presence chart, and independent sighting context
Black bear at Sunrise Mountain
SpeciesNet labelled this american black bear at 0.97 on 2026-08-21 19:02:59 UTC; MegaDetector drew the box at 0.88. Unreviewed model output.
- Source
- NPS Sunrise Mountain webcam
- Evidence
- classified crop from the source frame
Deer at Sunrise Mountain
SpeciesNet labelled this mule deer at 1.00 on 2026-08-21 14:17:19 UTC; MegaDetector drew the box at 0.88. Unreviewed model output.
- Source
- NPS Sunrise Mountain webcam
- Evidence
- classified crop from the source frame
Elk at Sunrise Mountain
SpeciesNet labelled this cervus species at 0.96 on 2026-09-04 14:40:39 UTC; MegaDetector drew the box at 0.87. Unreviewed model output.
- Source
- NPS Sunrise Mountain webcam
- Evidence
- classified crop from the source frame
Tightening the rules cut the review queue 15-fold
On 2,055 frozen detections from the same two hours of aquarium footage, correcting one conflict rule took the queue from 474 to 32. No detection changed, so the difference is the decision layer and nothing else.
- Source
- MaST Center Aquarium, 2026-08-10
- Evidence
- three passes over identical footage, each published as a queue
MaST aquarium source-quality break
Captured video from the Friday batch was already water-blocked, so this report treats the window as source-quality evidence rather than absence of fish or marine mammals.
- Source
- MaST Center Aquarium video bridge
- Evidence
- 1080p before/after video spot checks
Bush Point 4,816 Hz resonance
A persistent resonance explained most Bush Point Southern Resident detections, showing why public pages need floor checks and source-quality context.
- Source
- Orcasound Bush Point hydrophone
- Evidence
- resonance filter, spectrum chart, and per-node prevalence table
The most confident harbour seal call is a rockfish
The highest-scoring seal candidate in the widest run, at 0.79, has a visible dorsal fin and tail. Open-vocabulary labels are prompts for review, not species.
- Source
- MaST Center Aquarium, 2026-08-10
- Evidence
- annotated frame at the confidence the model reported
No featured finding matches that combination.
Latest reports
Newest data first. Each row links to the report and its report.json.
Source
October 2026
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- C Camera Model candidate
September 2026
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- H Hydrophone Model candidate
- H Hydrophone Model candidate
- C Camera Model candidate
- H Hydrophone Model candidate
- H Hydrophone Model candidate
- H Hydrophone Model candidate
- H Hydrophone Model candidate
- H Hydrophone Model candidate
- H Hydrophone Model candidate
- V Underwater video Quality gap
August 2026
- V Underwater video Reviewed
- C Camera Reviewed
- C Camera Model candidate
- H Hydrophone Classification
- H Hydrophone Classification
-
V Underwater video
Refining the marine detector on a fixed two hours series 1 of 4Model candidate
-
V Underwater video
MaST aquarium, 2026-08-10: pass 1, permissive rules series 2 of 4Model candidate
- V Underwater video Model candidate
-
V Underwater video
MaST aquarium, 2026-08-10: pass 3, wider vocabulary series 4 of 4Model candidate
Sources
H Hydrophone Orcasound hydrophone network
Continuous hydrophone audio from the Salish Sea, analysed for killer whale call presence and ecotype consistency.
OrcaHello. Binary call-presence detector. It does not identify pod, individual, or call type.
Ecotype classifier. Reports which population a call is more consistent with. It is a comparison, not a determination. Its authors found it less reliable on hydrophones other than those it was trained on; Orcasound is one of those it was trained on.
V Underwater video MaST Center underwater video
Underwater video from the MaST Center Aquarium pier in Puget Sound, analysed for animal presence.
YOLO-Fish. Marine mammal labels are review candidates from open-vocabulary prompts, not confirmed species.
MegaDetector. Detects that an animal is present, not which species it is.
C Camera Mount Rainier mountain cameras
Public National Park Service webcams, analysed for large mammal presence against a per-camera false positive floor.
MegaDetector. Detects that an animal is present, not which species it is.
SpeciesNet. Names the species in a MegaDetector box. At Sunrise its white-tailed deer labels are black-tailed deer, and an unnamed "animal" is usually a person crouching.
How to read these
- Every card is a model detection, not a confirmed sighting, unless it says a person reviewed it.
- Featured hydrophone evidence needs the full 30-second source window. Short review excerpts can remain counted, but they do not lead public evidence.
- Every label is shown only above its own measured error rate, at its own site. Floors are measured per hydrophone, never pooled across them.
- No detections is not no animals. Coverage and source-quality gaps are shown because a quiet day and an unrecorded day are not the same thing.
Full method, including false-positive floors, media rights, and approximate locations, is in METHODS.md. Machine-readable site data is in site-data.json.