What the internet is talking about

How the number is built

Each source is smoothed with an exponential moving average and divided by its own median over the previous 90 days. A median is used because a single viral day would raise a mean-based baseline.

The sources are combined with a weighted geometric mean. If a source did not report, the weights of the others are rescaled. With a geometric mean, one source cannot produce a high reading alone.

Full method

What the window changes

The window sets the period shown on the charts and the percentage next to each reading. That percentage is the change between the first and the last point on the chart.

It can differ slightly from the 7d column in the table. The table compares with the reading exactly 7 calendar days earlier. The chart starts from the first day the market has data inside the window.

The vertical scale of each chart is fitted to the visible period. The dashed line is always 1.00.

What Match orders by

Match sorts markets by how closely the price change and the index change agree over the selected window.

The score is min(|price|, |index|) − |price − index|, in percentage points. +33.6% price with +36.3% index scores 31. +52.7% with +77.2% scores 28. −1.6% with −3.8% scores −1.

A high score requires both changes to be large and close to each other. If price and index moved in opposite directions, the score is negative.

A high score does not show that attention caused the price change.

Where a reading is provisional

A market is marked provisional when it has less than 120 days of history. Its baseline covers a short period, so its readings are less reliable. Provisional readings are still shown.

No market is provisional today.