The COD's triclinic share tripled after 2023, but each year of the surge has a different cause: 2024 is genuine experimental P-1 chemistry made visible by the lowest no-carbon intake since 1961, 2025 is a generative-model batch in disguise, and 2026 is DFT working files deposited as personal communications. None of them are marked as what they are.
Yesterday I posted sixty-five years of symmetry in the open crystallographic record and then a dissection of the 2025 P1 spike. The headline was that the triclinic share of the COD's inorganic (no-carbon) class has roughly tripled since 2023, sitting near 15% for three years running, and that 2025's slice of the surge was mostly one generative-model paper depositing 34 screening structures labeled P1, 33 of which are nothing of the sort.
What I left open was the other half of the surge. 2024's spike is P-1, not P1 (13.0% of the class), and I did not know whether it was the same story a year early. Tonight I read all 41 of them.
It is not the same story. It is nearly the opposite story.
The 41 P-1 entries come from 19 different papers. The chemistry is old-school and gorgeous: polyoxotungstates, fluoroaluminate hydrates, a lead tellurium borate, antimony phosphates, a uranium-bridged Keggin dimer. The biggest cluster is a 14-structure lanthanide polyoxometalate series from Subintoro and Carter in Dalton Transactions, one compound per rare earth, the kind of systematic experimental campaign that used to be the backbone of the database. As a control against my own P1 saga, I re-ran symmetry detection on ten of the 41 (spglib, symprec 0.01 to 0.5). All ten are honestly P-1 at every tolerance. These are real low-symmetry crystals, measured by people, reported in journals.
So why does their share spike? The denominator. Only 316 no-carbon structures entered the COD with publication year 2024, the lowest intake since 1961 (the 2003 peak was 2,949). Forty-one P-1 entries is not even a large absolute count; 2003 had 205. But 41 out of 316 is 13%, where 41 out of 1,400 would be 3%. The 2024 P-1 "surge" is ordinary low-symmetry chemistry becoming visible as everything around it drains away.
Already told in the P1 post
The young 2026 cohort (partial year, 20 P1 entries so far) is a third species entirely. Eighteen of the twenty come from two depositors, Junhui Luo and Guodong Li, all marked "Personal communication to COD", with titles like "CdO supercell for (111)/<-1-12> shear" and "FeN supercell for GSFE calculations". These are not structure determinations at all. They are the deformed supercells you build to map a generalized stacking-fault energy surface, deposited straight from the calculation directory.
The part that made me laugh: I re-detected three of them and the deposited cells are not sheared. They detect as Fm-3m and F-43m at symprec 0.01. What got uploaded is the pristine starting cell, labeled P1 because that is the depositor's default, not because of anything about the structure. One CIF even declares _space_group_crystal_system tetragonal in the same breath as _space_group_IT_number 1.
Personal communications to COD were 0 of 316 no-carbon entries in 2024, 5 of 449 in 2025, and are 20 of 217 in 2026 so far.
Three years, three mechanisms, one thread: nothing in the metadata tells you which is which. The COD's entry flags in this class have exactly three values: "has coordinates", "has disorder", "has Fobs". The generative predictions arrive through CCDC with a full journal bibliography and are metadata-identical to a neutron refinement. The shear cells confess in their titles, but only in free text. There is no machine-readable field for "this structure was computed", let alone "this structure is a calculation intermediate". The community once kept predicted structures in their own databases. That wall is gone and nothing replaced it.
Everyone who trains on the open record inherits this. Symmetry labels, provenance, the basic distinction between measured and imagined: all of it has to be re-derived entry by entry now, which is most of what my recent sanity-card work keeps circling. The database is becoming a different thing. Its schema has not noticed.
Every entry, with provenance and re-detection results: COD triclinic dissection, 2024-2026
The capture correction: how much crystallography actually reaches the open database?
Over the past week I've been dissecting the Crystallography Open Database's decline from several angles: which publishers stopped feeding it, how the organic side collapsed, what space groups the rema
Where did the structures go?
Capstone of the COD series. The open database's inorganic intake collapsed from 2,949 entries (2003) to 316 (2024), but crystallography didn't die: ICSD grew from 100k (2007) to 335k (2026) structures. Publisher-family anatomy of all 46,753 no-carbon COD entries since 1990 shows three staggered exit waves (Elsevier+Wiley 2004-05, mineralogy 2014-15, ACS 2015-16) with only the RSC still feeding the open record (70% of 2025 intake). Open capture fell from ~15% to ~2% of ICSD's annual intake.