How it works, and where it is guessing.
The short version: count how often a topic shows up in several kinds of public data, month by month, and notice when the early kinds rise before the late ones.
What it reads
Monthly counts from several public sources, going back a few years. Each source sits on one rung of the ladder:
Jobs, Reddit and money flows are not in yet.
The ladder
A real shift usually shows up first among people building things, then among people adopting them, and last in the news and with the general public. A rung counts as reached when that kind of source is running well above its own level from a year or two earlier. The stage of a shift is the highest rung it has reached.
Moving, spread and score
How fast a topic is climbing: the average of its last 3 months over the average of the 3 before. A source only counts as rising when that ratio is clearly above 1 and the counts are big enough not to be noise. Spread is how many different kinds of source are rising together; one source on its own is never enough. The score from 0 to 100 mixes spread, speed and how early it is.
Dominoes
A domino is a knock-on effect in a neighbouring field: cheap voice AI, then fewer entry-level call-centre jobs. Each one has odds and a rough lag in months. This is the most judgment-heavy part: we found no published, tested method for predicting spillover between industries, so treat dominoes as reasoned guesses that get graded like everything else.
Calls and grading
Every call is written with odds in steps of 5%, the date it was made, the date it will be checked, and a rule that decides it without our opinion. When the date comes, it is marked hit or miss and stays on the scorecard. We report the Brier score and a calibration chart, and compare against a plain base-rate guess.
The backtest
We replay past months, letting the method see only the data that existed then, and check whether it would have flagged shifts that later went mainstream, how early, and how often it raised false alarms. The misses are reported next to the hits.
Limits
- It only sees public traces. Work inside companies, private deals and non-English sources are mostly invisible.
- Popularity is not adoption. Stars and page views can rise on curiosity alone.
- Sources change how they count. A feed that changes its own collection can fake a rise (search-based flu tracking famously did this). A jump on one source alone deserves suspicion, which is one reason spread matters.
- The ladder order is a pattern, not a law. Some shifts skip rungs or start in the news.
- Odds further out are worse. Accuracy falls towards chance over long horizons, so we prefer calls that resolve within months, not years.