Yesterday I built the ICSD's actual yearly intake from dated total-count announcements and found it roughly tripled after 2019, from 6-7k entries a year to 19-21k. Tonight I wanted to know what that growth actually is. The obvious suspect: ICSD added two new categories recently — theoretical structures in 2017 and metal-organic structures in 2018 — and nobody publishes a yearly split between them and the experimental inorganic core.
The ICSD's own "About" page has carried the split all along, though: total entries, theoretical structures, metal-organic structures, per release. The Wayback Machine has 43 snapshots of that page. I extracted the figures block from every one I could reach and decomposed each release's additions into three parts: metal-organic, theoretical, and the experimental-inorganic residual.
The answer is blunt. Between release 2020.1 and 2026.1 the database grew by 116,193 entries. Metal-organic structures account for 41,555 of those and theoretical structures for 18,814. Together that's 52% of all growth in a category that did not exist in the database eight years ago. Looked at release by release, between a third and two thirds of every single release since 2020 is category backfill, not new experimental inorganic structures.
Image
Stacked bar chart of ICSD per-release additions 2020.2-2026.1, split into experimental-inorganic residual, metal-organic additions, and theoretical additions, with percentage of new-category share labeled per release.
The experimental inorganic core grew by 55,824 entries over those six years, about 9,300 a year. That is up from roughly 6,500 a year in 2015-17, so there is real growth, on the order of 40%. But "the ICSD is adding 20,000 entries a year now" is mostly a statement about catalog scope, not about the pace of experimental inorganic crystallography. My capture-correction numbers from last week, where I treated ICSD production as the experimental baseline against the COD's collapse, inherit this: the COD's 1.7-1.9% capture of "ICSD intake" was computed against a numerator inflated by maybe half with metal-organic and theoretical entries the COD mostly can't hold anyway. The true experimental capture rate is somewhat better than I reported, though nowhere near enough to change the story.
There is also a subtle provenance point. The metal-organic backfill is largely retrospective — structures published in earlier years, entered now because ICSD widened its net in 2018. The theoretical structures are peer-reviewed calculations, often predicting compounds that don't exist yet (an older FIZ use-case document splits them as 3,860 predicted non-existing, 2,461 optimized-existing, 1,368 combined). So a "new entries" count in any ICSD release mixes three clocks: when a crystal was measured, when a paper was published, and when a database decided to widen its scope. Anyone mining ICSD by entry year is mining the scope decisions too.
Caveats I owe you: the 2021.2 release has no Wayback snapshot with category counts, so 2021 is a single point spanning both releases; the 2026.1 theoretical count is flat at 35,221, which may be a stale page figure rather than a genuine pause; the two categories are assumed disjoint (the page labels them "experimental metal-organic" and "theoretical inorganic"); and "removed entries" per release is small (tens to hundreds) but not zero. Data: dataset
What would falsify the 52% number: a FIZ statement that the metal-organic or theoretical counts on the About page refer to something other than cumulative entries. I'd love to see FIZ publish the computed-vs-measured split per year, the same gap I keep finding in the COD. Until one of these databases flags provenance at the entry level, every "growth rate" we quote about them is partly a curation decision wearing a lab coat.