Issues with material types

[akthom]

The list of material types in the explorer does not look accurate compared to what we had in the past version of iSamples Central.

For example, there's a material type that's listed as "material". That doesn't make sense at all -- that's not a material. It seems like either a) the material type metadata in the geoparquet files is incorrect, or b) something is not being read correctly from the files.

[akthom]

Related to issue #260

[rdhyee]

You're right, Andrea — and your instinct (b), "something isn't being read correctly," is the closer one. Two things are going on:

1. A read bug (the main one): the build that produces the facet data picked the first entry of each sample's material list (p__has_material_category[1]). But source records — especially SESAR — carry the whole SKOS ancestry in that list, and the broad root concept (.../material/1.0/material, whose label is literally "Material") can sit in the first slot. So ~half a million samples got tagged with the meaningless root. That's exactly the "material" entry you saw.
2. Some genuinely-unclassified samples: a chunk of records are only ever classified at that top-level "material" concept upstream — those are real, just unclassified.

I verified against the 2026-01 wide data: 528,983 samples had the root in slot 1.

Fix (PR #271): resolve the full list, drop the root, and take the first real concept. It's deliberately conservative — only the affected samples change (~318k get a real material; ~210k that are root-only become unclassified and drop out of the facet), and zero samples keep "Material", so that bogus entry disappears entirely. Correctly-classified samples are untouched.

Two honest caveats:

• This picks the first non-root concept, not necessarily the most-specific one (the lists aren't clean hierarchy paths — some even end in an unrelated "rock"). Truly picking the leaf needs the SKOS hierarchy; I've noted that as a follow-up.
• It's a build-time change, so it shows up on the site only after the derived parquet is rebuilt and redeployed.

---

— 🤖 rbotyee (RY's bot). RY skimmed this; the answer is the bot's. Ping @rdhyee if it's off.