Everything people build on GitHub, on one timeline.
23 megatrends and 151 trends, month by month since January 2019, from every new public repo on GitHub (11.8M last month). Each line is a trend's share of what the world creates, so a line going up means builders are choosing it over everything else. Dashed lines are a 12-month forecast, checked against its own past calls. Since ChatGPT, Developer tools went from 2 to 23 of every 10,000 new repos.
What to do about it
Action score (0 to 100) is a fixed formula: Jev's priority for the role (50%, half weight when Jev is unsure), heat (30%: share growth over the year and three months, attention growth, size) and the 12-month forecast (20%).
Top 10 megatrends
Top 10 trends 40+ new repos a month
Ranked by heat. "Next 12 months" is the forecast for the share of new repos: Rising (+25% or more), Edging up, Holding (within 7%), Easing, Falling (-25% or more). Jev reads the same numbers and gives its own outlook and a priority per role, from Pass for now (0) to Top priority (4).
Builders vs attention: where the room is
Right: a bigger share of new repos than a year ago. Up: a bigger share of new stars. Bubble: new repos a month. An open lane (top left) has attention growing faster than builders; crowding (bottom right) is the reverse.
The rank race, every quarter since 2020
Megatrends ranked by their share of new repos at each quarter's end. Lines you pick above keep their colour here.
Every trend by phase
Our calls, scored in public
Each month the forecasts for the top 10 megatrends and top 10 trends are frozen here, then scored after 3 and 12 months: right if the direction (up, flat within 7%, down) matched. The first calls were frozen for Sep 2026; the first scores land in Dec 2026. Backtest of the same model from 4 past starting points: direction right 90% of 339 6-month calls (carrying the last 6 months forward: 84%), typical level error 22% (last value carried forward: 25%), 80% of outcomes inside the 80% band.
How this is measured
Builders: GitHub Search counts of new public repos (forks excluded) created each month with the trend's topic tag, divided by all new public repos that month, per 10,000. GitHub as a whole grew about 10x since 2019, so raw counts make everything look like it is rising; the share does not. A megatrend adds up its core tags (a repo with two of them counts twice). Tags are often added after a repo is created, so the newest month is left out until it settles.
Attention: daily new stars from GH Archive (via OSS Insight) on 4.8k repos: the 30 most-starred repos of all time per tag and the Pulse sample of today's leaders and newcomers. Shown as each trend's share of all new stars on those repos. 87 repos with a Check first or red Trust label are left out. Because the tracked repos are the ones that made it, older attention reads a little high for trends that later faded.
Phase: from the 3-month average of the share. Faded: 70%+ below its high with no real comeback. Emerging: under a tenth of today's level two years ago. Accelerating: up 30%+ in a year and faster in the last 3 months than the 3 before. Growing: up 10%+. Peaking: within 15% of its high and flat for 3 months. Cooling: 20%+ below its high and still falling. Reviving: fell 30%+ from a peak and up 10%+ in 3 months.
Forecast: damped trend (Holt) on the log of the share, fitted per line on up to 48 months, settings picked by its own one-month errors. The band widens with the horizon. A forecast is an extrapolation of the line, not knowledge of what launches next; big launches (the numbered markers) are exactly what it cannot see.
Jev (free lanes) reads only these numbers and the leading repos, and answers two things per item: an outlook for the next 12 months and a priority from 0 to 4 for a builder, an investor, a participant and a user. A priority is unsure unless 80% of Jev's probability sits within one level of it.
Refresh: counts and stars daily, the model daily, Jev weekly or when the numbers change, calls frozen once a month.
Not investment advice. These are measurements of what people build and star on GitHub, plus a statistical forecast and a model's reading of them. They say nothing about any company's revenue, valuation or stock. Investor priorities are about which fields deserve research time, never a recommendation to buy or sell anything.