Relevance is deliberately not in this list. A score answering “is this moving” must not be partly “do we care”, or the radar can never tell us something irrelevant is exploding. It became the Fit score instead.
These are not editable here, and that is the honest position: they change what every stored snapshot means, so they belong in a commit somebody can review and revert.
27 runs carry a status, which is enough to read this seriously. The percentage is the share that held its ground — emerging that is still emerging or has become rising. 293 items appear in both runs, and the median trend score moved -2.0. Items the later run never saw are counted as “gone” rather than as a failed prediction — falling out of a search window is not evidence about a threshold.
The cadence lives in .github/workflows/snapshot.yml; four runs a day is a deliberate trade between resolution in time and the size of the repository, since the history is committed to it. What the radar searches for is a data file — the Topics screen shows it, what each direction returned, and how much of GitHub's request budget is spent.
A colleague opening the same page sees none of this, and it does not follow you to another device. Sharing it needs a store the server can read, which is the one thing here that genuinely needs a database.