Real-time analytics

The collector writes each hit into a store that the live view queries directly, skipping the batch jobs that build daily reports. Refresh runs from a few seconds to a minute, and only a short trailing window is shown, often the last 30 minutes.

Tarsier's live board reads recent hits and presence signals, and the globe plots them by coarse location. The daily reports are built from the same hits later, after filtering, so the two can differ by a small margin at the edges.

A food truck posts its location at noon. By 12:04 the live panel shows 37 people on the site, 12 of them in the home city and 9 reading /menu. By 12:30 only 14 remain. The owner sees most clicks came from a lunch district two miles from where the truck is parked.

Tomorrow the truck parks there. No daily report could have shown the owner that mismatch while the lunch hour was still going.

Real-time analytics is the reporting mode. Active users is the number it usually shows. A session is finalised later. Event hits appear in the live feed as they arrive, before any filter that would remove test traffic or bots has run.

Using the live view for weekly reviews. Expecting its number to equal the daily total. Treating a three-minute spike as a trend. Forgetting that bots look like readers until filtered. Assuming the pipeline has no delay at all.

Keep it open on launch days, email sends and incident calls. Write down the window it uses. When you need to decide something that costs money, wait for the finished daily or weekly numbers.