In 1871 Dmitri Mendeleev had a table of the elements with holes in it. Where a column and a row crossed at an empty square, he did not shrug — he described the element that would fill it: its atomic weight, its density, the formula and colour of its oxide, even how it would be found. He named these unborn elements with a borrowed Sanskrit number-word: eka-aluminium, eka-boron, eka-silicon — one-beyond the element above each gap. Within fifteen years gallium, scandium and germanium were pulled out of ores and lit up in flames, and the measured numbers matched his guesses to within a few percent. The prediction was the test, and the world passed it. Underneath that triumph is a quieter, deeper one: the axis his table was really sorted on was never the weight he used — it was a count nobody could see until 1913.
Mendeleev's rule was simple and brutal: line the known elements up by atomic weight, and a family likeness recurs at regular intervals — the periodic law. But a strict weight-line had bumps that broke the families, so rather than force a known element into the wrong column he left the square empty and predicted its tenant from the neighbours: the average of the elements above, below, left and right. Tap a dashed gap below to read what he wrote in 1871 about an element that did not yet exist.
This is the whole point of a science, and the rarest thing a theory can do: say a number before anyone measures it, and be right. Below, for each of the three, Mendeleev's 1871 prediction sits beside the value found after the element was isolated. The thin bar reads the error off the centre line — how far the world's answer fell from his guess. Switch between them. Seven of the nine numbers land within three percent; all nine within ten. The two loosest are named where they sit: scandium's oxide density (about nine percent out) and germanium's specific heat (about four).
Mendeleev's ordering principle was atomic weight — and yet he broke it, on purpose, in a handful of places. Tellurium is heavier than iodine, but he placed tellurium first, because its chemistry belongs with oxygen and sulphur and iodine's with the halogens. He trusted the family over the scale. For forty years this looked like a fudge. Flip the strip below from order-by-weight to order-by-atomic number:
In 1913 a young physicist, Henry Moseley, fired electrons at each element and measured the frequency of the X-rays they fired back. He found that the square root of that frequency climbs in a dead-straight line — not with atomic weight, but with the element's position in the table, an integer he called the atomic number (which we now know counts the protons in the nucleus). The line below is real: each dot is an element, plotted by the square root of its characteristic X-ray frequency. Toggle the horizontal axis between weight and number and watch which one the points actually fall on.
So the table's true axis was a whole number all along — a count of protons no one could see in 1871. Mendeleev's "errors" weren't errors; they were the periodic law reaching past the weights he had to the numbers he couldn't, and getting there first. Moseley was killed at Gallipoli in 1915, twenty-seven years old.
A page that only showed the hits would be lying by omission. Mendeleev predicted more than came true, and was wrong in ways worth naming. The method was powerful because it was periodicity, not prophecy — and periodicity by weight was itself only an approximation to the deeper law. A few of the misses and the caveats:
The honest summary: a real predictive theory, batting well above chance and far above any rival of its day — but a theory, with a real error bar and real failures, not a crystal ball. That it could be wrong in checkable ways is exactly what made being right mean something.
What this is. Standard history of chemistry, made operable: Mendeleev's published predictions (1871) for the three then-undiscovered elements beside the measured values reported on their discovery and refined since, the famous tellurium–iodine inversion, and Moseley's 1913 X-ray ordering. Nothing here is an original chemical finding. The only thing new is the form — the prediction set beside the measurement so the error is a bar you can read, and the weight/number axis a switch you can flip.
How it is kept honest. Three kinds of claim are marked apart. (a) The predicted and measured numbers are sourced, each carrying its citation; the percent errors shown are recomputed live in your browser from those two numbers, not typed in. (b) The atomic weights and atomic numbers in the ordering strip, and the X-ray frequencies in the Moseley plot, are sourced modern values; the straight-line fit (slope, intercept, R²) is computed live by least squares so you can see it, not take it on faith. (c) Historical judgments — what counts as a "hit", whether a given prediction was Mendeleev's or a successor's — are reported with their sources and flagged where scholars differ (the Sanskrit-as-homage-to-Pāṇini story among them).
What is and isn't proved. The verifier in research/eka/ runs — internal-consistency assertions: that every percent error the page can show is reproduced from the predicted/measured pair; that all nine quantitative predictions land within ten percent and seven of nine within three, with the two looser ones (scandium's oxide density, germanium's specific heat) the exact two the page names; that under weight-order the flagged pair inverts and under number-order it doesn't; that √frequency is linear in atomic number (R² > 0.999) and visibly less linear in weight, the same tellurium–iodine inversion stepping backward on the weight axis; and that the page embeds this dataset byte-for-byte (the drift guard). It does not re-measure the elements — the measured values are sourced, not re-derived — nor adjudicate the open historical questions, which it reports rather than settles.
Sources, per claim, live in the dataset and the verifier. Built one night in the Wasteland; every figure checked offline before it shipped.