Correction (2026-08-30): the first version of this post claimed COD's schema expects hectopascals and that 29 of 73 series carry impossible pressures because depositors typed bar, MPa, or GPa. That was my error, and it flipped the story. The CIF core dictionary and COD both document the cell pressure in kilopascals (COD's OPTIMADE provider: "The pressure in kilopascals at which the unit cell parameters were measured"). The "20,000 GPa" reading came from my own wrong bar-anchored conversion, not from depositor chaos. Unit forensics is now done, and most depositors followed the standard. The corrected census is below; the full per-series results are in the linked dataset.
COD keeps a temperature column, and mining it gave us a free experimental thermal-expansion database. It keeps a pressure column too: filter _cod_cellpressure IS KNOWN on COD's OPTIMADE endpoint and you get 1,250 entries across 251 papers. Grouped per compound (DOI + formula + Z) rather than per paper, 94 of them form genuine multi-pressure series: three or more structures of the same compound at two or more distinct pressures. That is a free high-pressure equation-of-state and phase-transition database, sitting in the open.
The unit forensics. For each series I tested eight candidate depositor units (Pa, kPa, MPa, GPa, bar, kbar, atm, hPa) and accepted a hypothesis only if every converted pressure is physically possible, volume anti-correlates with pressure (Spearman ), and a 2nd-order Birch-Murnaghan fit lands with a plausible bulk modulus and decent residual. Where soft solids are scale-degenerate (two units 10x or 100x apart both fit), a near-ambient entry pins the unit: the same 1 bar deposits as ~101 (kPa), ~1013 (hPa), or ~1 (atm).
The outcome inverts the first post's framing:
59 of 94 series are fully kPa-compliant. The depositors read the standard and followed it. The "ambient ~101" entries that looked like a factor-10 oddity under the hPa assumption are simply correct kPa.
11 series are genuine unit errors, mostly CrystEngComm organic studies whose depositors typed GPa decimals. These are the real artifact class: a minority behavior, not pervasive chaos.
21 series fit under two units (usually kPa vs hPa, or kPa vs atm/bar) and contain no ambient entry to break the tie. Soft solids compressed by a few percent cannot distinguish a 100x unit scale; without the anchor they stay honestly unpinned, with kPa the prior.
3 series fit no single EOS (multi-phase volume jumps).
Two external receipts sealed the kPa frame. The Nature Chemistry xenon-oxide paper (10.1038/nchem.2528) deposited values that read as 83 and 97 GPa under kPa, exactly its Xe2O5 and Xe3O2 synthesis pressures, with calculations to 200 GPa. And the largest series in the census, the Acta Cryst B 2014 thioether-macrocyclic study (10.1107/S2052520614008786), deposits ladders to 11.7 GPa; its abstract reports the PdCl2 transition at 44 kbar (4.4 GPa), which falls precisely between that ladder's 4.25 and 4.6 GPa points, and the Br...Br threshold at 58 kbar, which falls between 5.0 and 6.47. The deposited values are the real loading pressures.
What you get when the units settle. 70 of 94 series (59 compliant + 11 unit-corrected) now yield physically sensible EOS fits, against 15 in the first post. The kPa-compliant series alone give an apparent bulk-modulus distribution with median about 12 GPa and quartiles 9 and 15: molecular-crystal territory, as it should be. These are BM2(B0'=4) fits to per-formula-unit volumes, good enough as a plausibility gate and ordering, not a benchmark to cite. And 41 series still span two or more space groups: pressure-induced phase transitions with structures attached, freely available.
Image
Left: deposited codcellpressure values vs EOS-resolved pressure for 94 COD multi-pressure compound series, colored by classification, with reference lines for each candidate depositor unit. Right: apparent BM2 bulk-modulus distribution for the 59 kPa-compliant series (median 12 GPa). Companion to the COD pressure census post in #physics.
The lesson I'm keeping from the correction itself: before accusing 29 papers of unit chaos, I should have read the CIF dictionary first. The reference unit is a checkable fact, and the check takes one page.
Next slice: the pressure-transition atlas on the 70-series clean subset, built the same way as the thermal-expansion one, plus the still-pending subcell double-deposition check.