Some #183 candidates were never cultivated, and no metadata field says which

[realmarcin]

Found selecting the eleventh #183 record. Worth filing before anyone batches the remaining ~31, because the discriminator I looked for does not exist.

The case

Avena_Rhizosphere_CrossKingdom_SIP_Community came top of the candidate ranking: 95k of cached full text, 100% member coverage, methods keywords throughout.

It is a stable-isotope-probing study of native rhizosphere soil. Avena fatua was grown under ¹³CO₂ and paired rhizosphere/non-rhizosphere soil was collected at 6 and 9 weeks; the microbial community was never grown in a vessel by anyone. Giving it a cultivation_setup would assert a cultivation that did not happen — the same shape as Soil_Corrinoid_B12_Reservoir_Community, which was once scored 5/5 on growth conditions belonging to an E. coli reporter strain in a community that is metagenomic.

The plant had a growth setup. The community did not. CultivationSetup is about culturing the community — bioreactors, flasks, chemostats — not about the experimental system a field community was sampled from.

No field in the record distinguishes it

I checked all three candidates before assuming any of them worked:

community_origin — three NATURAL-origin records legitimately carry a setup, all verified as genuinely cultivated: GLBRC_UFMP_Fermentation_Community (in-house CSTR from a 4 L flask), Phormidium_Alkaline_Consortium (planar and tubular photobioreactors), Thiocyanate_Afipia_Thiobacillus_Bioreactor_Community (five CSTRs). Natural origin, cultivated afterwards.

ecological_state — the same two are STABLE, not ENGINEERED. Filtering on ENGINEERED would drop legitimate records.

community_category — a RHIZOSPHERE record already carries one (Lunar_Martian_Simulant_PGPB_Lettuce_SynCom, a stirred-tank bioreactor), as do CARBON_SEQUESTRATION and OTHER. Environmental-sounding categories are not environmental communities.

So: no combination of the record's own metadata separates "was grown in a vessel" from "was sampled from the world." The check has to be a human reading the source.

Why this matters now

The obvious next step on #183 is to batch the remaining records. That would be a mistake without this check, and the check is not automatable from the KB. The candidate ranking I have been using (member coverage + methods keywords + full-text cache) is a good filter for readable, and no filter at all for cultivated.

Not proposing a test

A guard here would need an allow-list covering every legitimate case, and the four above show the boundary does not follow any field. An allow-list that has to name each record individually is a list of decisions, not a rule — and it would give false confidence to exactly the batch run it is meant to protect.

Recorded here instead, so the next person to pick up #183 knows the ranking does not answer this question.

[realmarcin]

Two of the thirteen candidates are now checked, and both had to be refused. Recording it here so the list is not mistaken for a work queue.

record	verdict
Avena_Rhizosphere_CrossKingdom_SIP_Community	refuse — stable-isotope probing of native rhizosphere soil; the plant had a growth setup, the community was never in a vessel
OMM12_Gnotobiotic_Mouse_Gut_Community	refuse — its in vitro step is per-strain revival from glycerol cryostocks into 100 mL Wheaton serum bottles, before the mixture is inoculated into germ-free mice. The community lives in the animal. Those bottles are precultures, and a gnotobiotic mouse is not a CultivationSetup.
Cheese_Rind_InSitu_InVitro_Model_Community	partially curatable — the paper has an in situ survey and an in vitro reconstruction on cheese curd agar. Only the second is a cultivation; done in the PR that references this issue.

What the remaining ten look like

Not yet verified — this is a prediction from the record names and categories, offered so the next person checks rather than assumes:

• Likely refuse (community lives on a host): Lotus_LjSC3, ORNL_PMI_Populus_PD10_SynCom, PSY_Transgenic_Rice_Rhizosphere_Methane_Community, Suillus_Bacillus_Thiamine_Ectomycorrhizal_SynCom, hCom2_Complex_Gut_Microbiome. Plant SynComs and gnotobiotic gut communities have the same shape as OMM12: strains are grown individually in vitro, then the community exists on a plant or in an animal.
• Likely curatable: Acetylene_Fueled_TCE_Dechlorination_Groundwater_Enrichment (an enrichment is by definition cultured), Cyprus_Copper_Sulphide_Bioleaching_Consortium, Rifle_Aquifer_Bioanode_EET_Community (bioanode implies an electrochemical cell).
• Unclear: BioModels_MODEL2204300001_Kefir_Community_Model — a BioModels entry may be in silico, in which case there is nothing to record.

If that prediction holds, the tractable remainder of #183's full-text tier is roughly three or four records, not thirteen. Which is worth knowing before estimating the work.

The pattern underneath

In every refusal so far the paper does contain real cultivation numbers — vessels, volumes, temperatures — for the strains individually. That is why these records rank highly on any automated filter and why the refusal always takes reading. A SynCom paper almost always describes growing its members; describing growing the community is a different sentence, and sometimes it does not exist because the community was only ever assembled in a host.