Every month Radar reads the arXiv record and seals the terms breaking out of their own history. Each call is graded in public three months later, hit or miss.
Each call commits to the one before it with SHA-256, then gets its grade once its three months have passed. Your browser can recheck the whole chain. Change one character of any old call and verification fails.
Chain head
Limit: the chain exposes edits to anyone holding an earlier copy. An independent public timestamp is not in place yet.
04 · Method & limits
How a term qualifies.
arXiv only. Its identifier fixes the month of first submission and the monthly record is near complete. Other sources in Alderamin are excluded here because when they were harvested would distort a trend.
Growth is adjusted. Maths rose from about 15% to 21% of arXiv during 2026, so each term is measured against the archives it lives in, not all of arXiv.
Not a verdict on truth. A rising term means researchers are writing it more often. It does not mean the work is right, useful or commercially important.
In-sample so far. The rules were tuned on the same 2026 data the backtest uses, and abstracts are stored at their latest version, which can leak later wording into older papers. Only sealed calls are a clean test.
One year of data. A three-month grade says nothing yet about a one to two year lead. That claim has to be earned by the ledger over time.
Need this on your own field?
The same engine can watch a sector, a technology or a set of competitors, with a sealed record you can show a board.