Frontier data · Accession series MOOSE
Tine is the data foundry frontier labs trust to label, rank, and benchmark moose at scale. Eighteen hundred pounds of training signal, annotated down to the antler tine, with inter-annotator agreement we will put in the contract.
A vetted workforce of field biologists draws bounding boxes, antler-tine keypoints, and pelage masks around an animal that is, on average, the size of a delivery van and visible from a moving train. Inter-annotator agreement holds above 98.4%.
We collect pairwise preference data directly from the herd. Two forage options, one moose, one ranked judgment, logged. Your model learns to align with what a fourteen-hundred-pound ruminant actually prefers, which turns out to be willow.
Twelve thousand held-out examples measuring exactly one capability: can your model tell a moose from a horse wearing a winter coat. Most frontier models score below 60%. We send you the number on a card.
“Our model confidently labeled three elk as moose the week before launch. Tine caught all three. The fourth one was an actual moose, and they caught that too.”
“Our last vendor outsourced the labeling to people who had never met a moose. Tine’s annotators have all met a moose. Several of them more than once.”
“I asked for inter-annotator agreement above 95 percent. They returned 98.4 and a photograph of the one moose that disagreed.”
For teams labeling their first moose.
Most teams land here. Their reasons vary.
For organizations operating at herd scale.