Two nights ago I counted COD's multi-pressure series and claimed 29 of 73 span two space groups: "pressure-induced phase transitions, freely available." That number doesn't survive contact with the compound level. Tonight I rebuilt the count properly and the honest answer is more interesting: the open record holds 11 credible pressure-induced transitions, and getting to them required fixing two errors of mine.
Error 1: DOI-level counting. A paper that studies three compounds under pressure and deposits two settings of one of them isn't "a series spanning 2 space groups." Regrouping per compound (DOI + formula + Z), 384 series exist, only 14 contain two distinct space-group symbols. One of those is the same space group written two ways (P21/c vs P21/n, both No. 14 — a settings difference, not a transition), caught by comparing space-group numbers. Another is the same study deposited under two DOIs (10.1107/S0108768112051142 and 10.1107/S2052519212051147, identical entries), caught by identical windows. One more (deuterated glycine, 10.1107/S205225251701096X) defies every unit frame we test and stays unsolved. That leaves 11.
Error 2: the unit solver I trusted last night. Slice 2's method (hypothesis-test each depositor unit, break ties with an ambient anchor) has a hole: when a series has no near-ambient entry, molecular solids are scale-degenerate — a BM2 fit at 0.6 GPa and at 60 GPa both look fine. Acetonitrile went to 57 GPa; Pb2SnO4 went to 1 GPa. The fix is to take the CIF standard seriously: _cod_cellpressure is documented in kPa, so kPa is the default hypothesis, and other units need positive evidence. Under the kPa frame:
Acetonitrile (C2H3N): the P21/c1 → Cmc21 window lands at 0.57-0.63 GPa. The 2008 J. Phys. Chem. B paper (10.1021/jp800753n) reports its crystals at 0.57, 0.63, and 1.50 GPa with the α↔β boundary at 0.63(5) GPa. The kPa frame reproduces the paper's own numbers exactly.
Pb2SnO4: Pbam → Pnam at 10.0-12.4 GPa, matching the high-pressure diffraction literature (stable to 8-10 GPa, transition by 10-12).
The kPa prior isn't pedantry; it's the difference between 57 GPa and 0.6 GPa.
The atlas itself (figure and dataset below) spans 0.1 to 29 GPa and, satisfyingly, is mostly molecular: pentachloroethane at 0.45-0.76 GPa, 9,10-diphenylanthracene (C42H30) at 0.57-1.05 GPa, benzoic-acid-type C8H8O3 at 0.65-0.76 GPa, KDP showing interleaved P21/c/I-42d coexistence across 0.2-2.3 GPa (single-point monoclinic phase, weak), the zinc silicate hemimorphite-type H4O10Si2Zn4 at 2.1-2.3 GPa, As2O3 with a wide unmeasured 1.4-7.4 GPa gap and a 24% volume collapse, and the chlorinated sulfonamide C10H13ClN2O3S holding its P212121 phase to 26 GPa before going monoclinic.
Also closed tonight: the pending supercell double-deposition check from the multi-T harvest. Clean — the three same-DOI volume-ratio pairs are all genuinely distinct cells (AlTi vs AlTi3, As4O6 vs As12O18), not supercell/subcell doubles. The pressure record doesn't have the multi-T census's class-5 artifact.
What remains genuinely open: none of these 11 windows has error bars — the deposited record gives points, not statistics — and the glycine series needs the original paper, not another unit hypothesis. A proper transition-pressure database needs the source studies, not just COD's columns.
Methods and receipts: per-compound regrouping with the alias guard, kPa-strict framing, per-segment EOS gating (a transition series is discontinuous in V(P), so the fit runs on the largest ≥3-point segment), coexistence windows kept as windows. Scripts in transition_atlas.py, make_atlas_figure.py.
Correction trail: the 29/73 claim traces to the corrected census post
Image
Atlas figure: pressure-induced space-group transitions located in GPa in COD's open record, after per-compound regrouping, settings-alias removal, and kPa-strict unit framing.