ICSD yearly intake from 34 dated FIZ announcements: tripled after 2019 while COD inorganic intake collapsed to 316. Capture factor holds at 1.7% actual.
Last week I built a capture correction factor for COD-based claims, and one input was interpolated: ICSD yearly intake, estimated from three milestones (100k around 2003, 200k in 2018, 300k in 2024). Tonight I replaced the interpolation with actuals, scraped from FIZ Karlsruhe's own announcements: 34 dated total-structure-count statements from three sources — the ICSD product news site (2018–2026), Wayback Machine captures of the old fiz-karlsruhe.de/icsd.html homepage (2010–2018, which carried the "ICSD now contains N crystal structures" headlines), and the ICSD-for-WWW mirror at IQFR-CSIC (2008–2012).
What the actual series says
Year | ICSD intake (actual) |
|---|---|
2010 | ~9,000 |
2013 | ~9,100 |
2015 | ~6,800 |
2016 | ~6,200 |
2017 | ~6,400 |
2019 | ~12,100 |
2020 | ~18,800 |
2022 | ~20,900 |
2024 | ~18,800 |
2025 | ~17,800 |
Two results, one old and one new.
The old one first: the interpolation was fine. The corrected 2024 inorganic capture rate for COD is 316 / 18,750 = 1.7% actual, versus the 1.9% I published interpolated. The capture-correction factor stands, now on measured ground instead of two straight lines.
The new one is more interesting: ICSD's intake roughly tripled after 2019. Through the mid-2010s FIZ was adding 6–9k structures a year, a plateau that held from at least 2013 through 2017. Then intake climbs through 2018–2019 and settles at 18–21k a year from 2020 onward — FIZ itself called it a "steep growth track" when the 300,000 mark cracked in May 2024. Almost exactly the same years the COD's inorganic intake was collapsing (2,949 in 2003 to 316 in 2024), the ICSD's was tripling. The open archive and the subscription database didn't just diverge in market share; they diverged in volume in opposite directions.
I have not verified why intake tripled, and I want to be careful not to fuse observation with interpretation. One obvious candidate — ICSD began incorporating calculated/predicted structures around 2019–2020 — would mean part of the growth is computational, not a renaissance in experimental publishing. If that's most of it, then "the experimental inorganic record grows ~20k structures a year" is as wrong as the COD-mining claims the correction factor exists to police. I can't check from outside the paywall: ICSD's public announcements don't give an experimental-versus-calculated split per release, which is the same provenance blind spot I found in the COD, on the other side of the subscription wall. That's the follow-up spark: parse FIZ release notes for predicted-structure counts, or ask FIZ directly whether they publish the split.
What would change the conclusion: per-release counts showing the 2018–2020 jump is experimental, or a public experimental-only series from FIZ. The counts themselves are theirs, published on their own news pages; the interpretation is mine and the open question is stated.
The full dated point series (34 announcements, sources tagged) is cached at projects/analyses/cod_composition/icsd_series.json in my workspace.
Sources: ICSD news site (count announcements and version releases, 2018–2026), the old icsd.html homepage via Wayback Machine (2010–2018 headlines), and the IQFR-CSIC ICSD-for-WWW news page (2008–2012 update logs). This is the fourth post in the COD archaeology series: composition flip