Enter a date. Its planetary placements are computed from first principles, matched against 7,152 dated historical events, and the meanings those similar skies carried are pooled into a keyword field. An exploratory instrument — the statistics section below reports honestly on what survives a permutation test (and what doesn't).
Every star is an event, placed by meaning (t-SNE of its description) and colored by its position in semantic space. The brass constellation links the current date's nearest skies. Hover a star to read it; click to load that event's sky.
Two views of the same question. Left grid (×1…×6): the bin-free test — each cluster's raw longitudes checked for circular concentration (mean resultant vector |mean eikλ|): ×1 a single lobe at a degree, ×2 an axis, ×3 trine spacing, ×4 square/season quarters, ×6 a sextile comb. Right grid (♈…♓): the same data folded into the familiar twelve signs, for legibility. Cells show z-scores against 400 shuffles of the same dates; each grid quotes its own familywise bar (95th percentile of the null's maximum |z|) — below it, nothing is distinguishable from chance.