01 / A MAP, NOT A MEASUREMENT Every point is one curated entry, and a batch release counts once. The curve shows how this collection grows. It is not a measured rate of scientific progress. A selective collection cannot establish a scientific “knee” or an exponential growth rate.
02 / WHY RECENT MONTHS LOOK DENSER Recent announcements are easier to find and still fresh, and older periods have been backfilled by hand. The hero comparison counts entries in two equal 90-day windows ending at the editorial cutoff. More entries can mean more work, more announcements, more attention, or our own selection. It does not prove faster science.
03 / WHAT “BUILDS ON” MEANS A link appears only where the source documents the dependency, for example a model trained on, or using, an earlier system’s outputs. Missing links do not mean independence. Branches in the tree group fields. They do not claim causation.
05 / DEEP TIME AND SELECTION Deep time uses a reversed logarithmic scale of years before the editorial cutoff; recent years occupy more space. Time dive narrows the window from 1440 to 2020. Historical markers are hand-picked context, not milestones. Comparing their gaps with recent entries illustrates selection bias: neither list is complete or consistently sampled, and shorter gaps do not establish acceleration.
04 / FOLLOW THE EVIDENCE Each entry links to a primary source and states its limits. Prediction, laboratory evidence, benchmarks and mathematical verification are different kinds of progress. The collection is curated by hand. Automatic updates are not enabled.