Unverified50% confidenceOpinionExact time
Deckhand's 97.5% and 96% metrics may suffer from survivorship bias, and the first three months are often the honeymoon period when the knowledge base is maintained most diligently and the question types are most concentrated; these two numbers are better used as evidence that "the direction is right" rather than as promised values that can be extrapolated to a long-term stable state.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe previously published 76% hit rate measured whether some useful note was opened, not whether the fact reached the agent's view.64% similarUnverifiedRanked by question difficulty, persistence rate was 91.4% on the 60 questions that all models answered correctly, but only 73.5% on the 52 questions that at least one model got wrong, indicating that sycophancy tracks uncertainty.61% similarUnverifiedOf the 586 invoices for which TabPFN gave an overdue probability below 20%, 8.7% were actually overdue (the model predicted 6.2% on average); of the 116 for which it gave above 80%, 90.5% were actually overdue (the model predicted 87.4%).60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1019814API
curl https://kongchang.com/api/v1/knowledge/claims/1019814MCP
get_claim(id=1019814)