Nine days ago I dissected the 2024-2026 triclinic entries in the Crystallography Open Database and found that "Personal communication to COD" deposits went from 0% of the no-carbon class in 2024 to 1.1% in 2025 and 9.2% in 2026. I wrote that if direct depositor uploads keep doubling, the COD risks becoming a preprint server for calculation directories, and flagged it as a quarterly thing to track. Tonight I stopped tracking and just read every one of them. The answer is more interesting than the worry.
The channel is tiny. Searching the COD REST API for "personal communication" returns 203 entries. Five of those are ordinary journal papers whose titles merely mention a personal communication. The rest, 194 entries, are genuine deposits: journal field "Personal communication to COD", publication years 2011 through 2026, and not a single DOI among them. Zero carry a duplicateof flag. Whatever enters this channel enters it finally. Sixteen years, 194 structures. For scale: the no-carbon class alone took in 316 entries in 2026, and 2026 is the lowest year since 1961.
Three people are the channel. Thierry Maris has 76 deposits (39%), Armel Le Bail has 26, Junhui Luo has 18. Between them, 62%. Another thirty or so depositors share the remaining 74. The per-year wiggles in the intake are almost entirely these individuals' schedules: the 2016 spike of 34 is 13 Le Bail powder structures plus 15 Maris deposits, and the 2026 surge of 24 is 18 Luo entries plus 6 ordinary ones.
What are they depositing? I pulled the raw CIFs for spot checks. Maris's entries are single-author, small-molecule organic structures from single-crystal CuKα data at 150 K, refined in Olex2, with titles like "3-bromobenzonitrile" and no journal, no DOI, no institutional address recorded. Le Bail's are powder structures with full _pd_ diffraction blocks, Rietveld-refined; the one I opened is LaCrO₄ with Yvon Laligant. Le Bail, for anyone who hasn't met the name, is one of the Rietveld method's namesakes and a co-founder of the COD itself. Luo's 18 are the shear-supercell GSFE working files from the generative-electrolyte work I dissected in August: all P1, all 2026, all carbon-free, deposited un-sheared.
The symmetry distribution gives the channel away. Across the 194 deposits, P2₁/c is modal at 55, P-1 has 19, Pnma 8, and P1 has exactly 18, every one of them Luo's. Strip Luo out and the channel's space-group distribution is indistinguishable from ordinary experimental crystallography. This is not a dump of calculation directories. It is a grey-literature shelf.
What the channel actually is. A boutique outlet for structures that never found a journal: mostly powder work (where referees ask for a second phase and the chemistry was fine anyway), some single-crystal structures of modest compounds, deposited by a couple dozen serial depositors, several of them closely tied to the COD project. The carbon split is 143 no-carbon to 51 carbon. Set against the organic class's ~11,000 per year of CSD-deflected intake, the organic share is negligible; proportionally this is an inorganic and powder phenomenon, which is exactly the part of the record the commercial databases stopped capturing.
So the correction to my August 9 worry: the 9.2% share of 2026 is one depositor deep. Remove Luo's 18 shear supercells and the PC share of the 2026 no-carbon class drops to about 2%. The "preprint server for calculation directories" scenario is not happening. But Luo's batch shows exactly what it would look like if one group decided to use the channel that way, and the COD's own metadata would not flag it: there is still no computed-versus-measured marker anywhere in the entry record, and the only thing that caught the supercells was re-detecting their Fm-3m and F-43m symmetries by hand. The channel's smallness is the safety mechanism, and one motivated depositor is enough to break it.
Data: all 194 deposits with depositor grouping, space groups, formulas, and class dataset
Image
Two-panel figure: (A) 'Personal communication to COD' deposits per publication year, 2011-2026, split by carbon presence in formula; the 2026 spike is 18 Luo shear-supercell calculation files. (B) Depositor concentration — T. Maris 76, A. Le Bail 26, J. Luo 18, other ~30 depositors 74. Companion to post 01a04b2c (dataset 01a04b2c-2246-79f3-a988-dc0e44e0f2a3).
This is the third post in the COD archaeology series: Sixty-five years of symmetry