[kavanase]

Spotted this issue when testing out the latest mp-api release and associated MP database release.

Summary

MPRester.get_entries_in_chemsys(..., property_data=[...]) (default mixed thermo type) raises KeyError for any chemical system whose pre-built phase diagram contains a GNoMe entry, for API keys without GNoMe access. E.g. Cu-Se-Si (database v2026.09.28):

MPRester().get_entries_in_chemsys("Cu-Se-Si", property_data=["energy_above_hull"])
# KeyError: 'mp-3336429'

mp-3336429 (Cu3SiSe4, on the hull) is a GNoMe material (gnome_r2scan_statics, license: BY-NC).
The pre-built phase diagrams (S3 objects/phase-diagrams) include GNoMe entries for every API key, but
thermo.search (like all the access-controlled REST collections) does not return GNoMe docs to API keys without GNoMe access, so the property_data decoration fails on the missing doc.

Changes

• ThermoRester.get_phase_diagram_from_chemsys (used by get_entries_in_chemsys, get_stability,
  the oxygen evolution helper and the MCP phase diagram tool): for keys without GNoMe access, GNoMe entries are ommited and the (mixed) phase diagram is rebuilt.
• Leaving the entries out alone is not enough: in r2SCAN-anchored systems, MaterialsProjectDFTMixingScheme
  places GGA(+U)-only entries onto the r2SCAN hull including the GNoMe entries, so these are
  re-referenced to the r2SCAN hull without them (_accessible_entries).

Additional details from Claude:

Scope (served GGA_GGA+U_R2SCAN phase diagrams, read from the public S3 Delta tables)

v2026.09.28: 21,694 of 117,172 chemical systems (18.5%) have GNoMe entries in their served phase
diagram (KeyError without access). 29,203 GNoMe materials, 27,977 of them on a hull.

Checklist

• Google format doc strings added. Check with ruff.
• Type annotations included. Check with mypy.
• Tests added for new features/fixes.
• If applicable, new classes/functions/modules have duecredit @due.dcite decorators to reference relevant papers by DOI (example)

[kavanase]

From what I can tell, the test failure here is unrelated to the PR (test_mock failing due to my fork not having the MP_API_KEY secret?)

[kavanase]

Also just noting something that was spotted when investigating this: the v2026.09.28 MP mixed phase-diagram table has 3,536 duplicate chemical-system rows, all identical. mp-api just takes the first, so it makes no difference, just fyi!