Signals
What the news is talking about, and how well simple models predict that attention. This page describes patterns in research, technology and business coverage so you can learn how ideas travel. It does not say what anyone should do, and it is not investment, career or other advice.
Loading…
Loading…
How to read this page
- Attention is not truth. A phrase being everywhere shows what outlets chose to write about, not that it is important or correct.
- Ideas often travel from research to industry press. The stage label only says where a phrase appears now: research sources, industry press, or both.
- Everything is traceable. Each signal links to the stories behind it, and the method is written next to it. The raw data is in signals.json and forecasts.json.
- Small data, honest limits. The page collects about 450 stories a day. Early readings are noisy and are labelled as such.
Ideas
Topic momentum
Stories per topic in the last 7 days against the average week of the three weeks before.
Tone of business headlines
A word count, not a market forecast: the balance of upbeat and worried words in business headlines.
Stories several outlets cover
Headlines from different outlets that share most of their key words, with each outlet's rating (of the outlet, not the story). Reading across outlets is a good way to see how framing differs.
Forecast scoreboard
Once a week the page writes down questions about its own data, for example whether a phrase will keep appearing, together with a probability from a stated model (a Poisson model that assumes this week's rate fades a little). Questions are never edited afterwards, each one covers 7 days, and the result is scored with the Brier score, (probability − outcome)². Always saying 50% scores 0.25, so a lower score means better than that benchmark. This is the idea of strictly proper scoring rules from my paper You Should Be Properly Scoring Your Odometry, applied to forecasts. The git history of forecasts.json is the audit trail.