Microbial disease granularity review (2026-08-29)

Question. Cancer entries got a ratified granularity ladder in August 2026 (design decisions §3a, from cancer-taxonomy-granularity-review-2026-08-28.md). Mendelian disease has one too (§3, issues #306/#7082). Infectious disease has neither. This report reviews how the ~146 microbial entries currently in the KB are split and lumped, identifies where the implicit practice is inconsistent, and proposes a decision ladder for ratification.

TL;DR. The KB is splitting microbial disease along six different axes at once — organism taxonomy, organ syndrome, clinical eponym, phase of infection, transmission setting, and neoplastic sequela — with no recorded rule saying which one wins. Most individual calls are defensible; the problem is that nothing says why, so the same situation is resolved two different ways in different parts of the KB. Concretely: Legionella is split by syndrome (Legionnaires' disease / Pontiac fever) while Neisseria, Streptococcus, Haemophilus and Listeria are lumped by syndrome into one Bacterial_meningitis entry with no subtypes; Rocky Mountain spotted fever and boutonneuse fever exist simultaneously as full standalone entries and as subtypes of Spotted_Fever_Rickettsiosis, both carrying the same MONDO IDs; and ten lumped entries anchor to a MONDO term far broader than the entry (Travelers_Diarrhea → MONDO:0001673 diarrheal disease). The single exemplary artifact in the whole area is the Treponematoses grouping, which does record its lump/split reasoning explicitly — and it is the only one of 99 groupings covering infectious disease.

On identifiers: NCBITaxon is the sole vocabulary in use, and that is the right call. It reaches serovar and serotype rank, so the unbound agents are curation misses rather than ontology limits. GTDB and GenBank assembly accessions are absent and should stay absent — GTDB classifies Shigella as Escherichia coli, which would erase a distinction the KB depends on. The one thing no taxonomy can carry is pathotype, which §4 rung 3a handles structurally instead.

Post-publication note (2026-09-16). The ladder in §4 was ratified as design decisions §3e — not §3b, which was taken by the time it landed — with three corrections that the discussion on #10115 produced after this report was written, all recorded there rather than edited into the text below:

1. §2.3 and §5 item 4 are withdrawn for Spotted_Fever_Rickettsiosis. Its RMSF and boutonneuse fever rows are pointer subtypes doing what §3a L4 prescribes for a promoted stratum, not a second copy of the disease; the pointer is now machine-readable (Subtype.curated_in). The finding stands for Viral_Hemorrhagic_Fever.
2. §5 item 6's pathotype: slot is not added. A pathotype is a disease-level has_subtypes row with classification: pathotype; agent-level strata are taxonomic and bind NCBITaxon once PR #10353 lands.
3. The case catalog proposed in the follow-up discussion is not created; the checker just check-granularity is the computed half, and a decision about a lump is recorded in the entry itself.

---

1. What the KB does today

1.1 The set

Set	Count
Disorder entries carrying an infectious_agent: block	125
Further entries whose etiology is a transmissible agent but which carry no infectious_agent: block	≥21
Total microbial entries (working figure)	~146
Groupings over any of them	1 (Treponematoses) of 99
Modules for infection biology	~15 (intracellular_pathogen_persistence, granuloma_formation, innate_antiviral_interferon_response, the antibiotic/antiviral target modules, …)
Infectious stubs still queued	40, all entry_type: UNDECIDED

The 21+ entries missing infectious_agent: are not edge cases. They include COVID-19, Lyme_Disease, Pertussis, Legionnaires_Disease, Q_Fever, Psittacosis, Middle_East_Respiratory_Syndrome, Paralytic_Poliomyelitis, Congenital_Zika_Syndrome, Mycetoma, Noma, Dental_Caries, Pelvic_Inflammatory_Disease and all five pneumonia entries. Nine of them share the parent strings Bacterial Respiratory Infection / Viral Respiratory Infection / Fungal Respiratory Infection — a coherent curation tranche that simply never populated the pathogen slot. The causative organism is therefore not machine-queryable for roughly one microbial entry in seven, including the most-read entry in the KB.

1.2 Six granularity axes, none declared

Axis	Examples	Consistent?
Organism (1 species ↔ 1 disease)	Cholera, Measles, Rabies, Tetanus, Buruli_Ulcer, Melioidosis, Trachoma	yes — the stable core
Organism genus/complex, lumped	Babesiosis (5 taxa), Brucellosis (3), Leptospirosis (2), Aspergillosis, Leishmaniasis, Human_African_Trypanosomiasis (2 subspecies)	inconsistent — see §2.2
Organ syndrome across unrelated pathogens	Bacterial_meningitis (4 taxa), Infective_Endocarditis (3), Endophthalmitis (4), Folliculitis (4), Choroiditis (4), Chorioamnionitis (6), Travelers_Diarrhea (8), Viral_Encephalitis (5), Viral_Hemorrhagic_Fever	inconsistent — see §2.1
Clinical eponym / syndrome from one organism	Legionnaires_Disease vs Pontiac_Fever; Hantavirus_HFRS vs Hantavirus_Pulmonary_Syndrome; Scarlet_Fever vs Toxic_Shock_Syndrome	inconsistent — see §2.3
Phase of one infection	Acute_Hepatitis_C_Virus_Infection vs Hepatitis_C; Acquired_Immunodeficiency_Syndrome with no HIV-infection entry; Long_COVID vs COVID-19	inconsistent — see §2.4
Infection-attributed neoplasm	Cervical_Cancer, HPV_Positive_Head_and_Neck_Cancer, EBV_Associated_Gastric_Cancer, Burkitt_Lymphoma, Kaposi_Sarcoma, Merkel_Cell_Carcinoma, Adult_T_Cell_Leukemia_Lymphoma, Aflatoxin_Related_HCC	governed by §3a, not by any infection rule

1.3 Measurable symptoms

Symptom	Measurement
Hierarchy is free text, not a foreign key	parents: is multivalued: true, range: string. The 125 flagged entries use 131 distinct parent strings — more distinct strings than entries
…and it is not even case-normalised	Bacterial Infection (17) / Bacterial infection (2) / Bacterial infectious disease (2) / bacterial infectious disease (2) / primary bacterial infectious disease (2) are five spellings of one concept
Life-cycle modelling is curated for one clade only	agent_life_cycle: 15/125 (12%). Where present it is well populated (hosts 15/15, life_cycle_stages 12/15, vectors 8/15) — but all 15 are vector-borne parasitic or arboviral entries. At least 24 further zoonotic or vector-borne entries have no life-cycle block at all (§2.6). transmission: 70/125 (56%)
Subtypes are mostly unbound	181 subtypes across the infectious set, only 54 (29%) carry a subtype_term ontology binding
Over-broad MONDO anchoring on lumps	10 entries have ≥3 agents and 0 subtypes; several anchor above their own scope — Travelers_Diarrhea → diarrheal disease, Soil_Transmitted_Helminthiases → helminthiasis, Lymphatic_Filariasis → filariasis, Rhinovirus_Infection → common cold
Mapping discipline is nearly absent	17 mondo_mappings entries total across 125 files (12 exactMatch, 3 narrowMatch, 1 broadMatch, 1 closeMatch)
No disease_term at all	Adenovirus_Respiratory_Infection, Lone_Star_Virus_Infection
Unbound pathogen taxa	7 entries name an agent with no NCBITaxon binding (Tetanus → Clostridium tetani, Pinta → Treponema carateum, Cytomegalovirus_Retinitis → Human cytomegalovirus, …)
One entry is the ontology root	Infectious_Disease is a 184-line Disease entry anchored to MONDO:0005550 infectious disease, with one generic pathophysiology node ("Pathogen Invasion and Replication")

---

2. Specific defects

2.1 Organ-syndrome entries lump unrelated pathogens with no internal structure

Bacterial_meningitis (1,352 lines) names Neisseria meningitidis, Streptococcus pneumoniae, Haemophilus influenzae and Listeria monocytogenes as its four agents and carries zero subtypes. These four organisms differ in age distribution, vaccine availability, capsular vs listerial pathogenesis, empiric antibiotic choice (ampicillin is added specifically for Listeria), and MONDO identity — each has its own MONDO class. The entry's pathograph therefore has to describe a mechanism generic enough to cover all four, which is exactly the failure mode the cancer review identified for organ-umbrella entries.

The same pattern holds for Infective_Endocarditis (3 agents / 3 unbound subtypes), Endophthalmitis (4 / 6 unbound), Folliculitis (4 / 8 unbound), Choroiditis (4 agents, none NCBITaxon-bound, 10 subtypes), Chorioamnionitis (6 / 3) and Travelers_Diarrhea (8 agents, 0 subtypes, anchored to diarrheal disease).

Travelers_Diarrhea also exposes a representation limit worth recording: its three E. coli pathotypes (ETEC, EAEC, EPEC) all bind to NCBITaxon:562 Escherichia coli, because NCBITaxon has no pathotype rank. Pathotype is a real mechanistic distinction (heat-labile/heat-stable enterotoxin vs aggregative adherence vs attaching-effacing lesion) that the current binding cannot carry.

2.2 Genus-level lumps are decided ad hoc

Coccidioidomycosis lumps C. immitis and C. posadasii — correct; they are clinically indistinguishable. Human_African_Trypanosomiasis lumps T. b. gambiense and T. b. rhodesiense — questionable; they differ in reservoir (anthroponotic vs zoonotic), tempo (chronic West African vs acute East African), and staging/treatment pathway, which is precisely the profile the Treponematoses grouping used to justify splitting four near-identical organisms. Malaria lumps P. falciparum and P. vivax, which differ in hypnozoite biology and therefore in radical-cure requirements. Nothing records which way the call went or why.

Note what makes this hard to audit: reservoir host and arthropod vector are among the strongest arguments for splitting a genus lump, and the agent_life_cycle block that carries them is populated on only 15 entries — none of them the ones in dispute here. Human_African_Trypanosomiasis is one of the 15 and does record two hosts and a vector, which is precisely why its split question is answerable at all.

2.3 Rocky Mountain spotted fever is modelled twice, at two levels

Concept	As standalone entry	As subtype of Spotted_Fever_Rickettsiosis
Rocky Mountain spotted fever	Rocky_Mountain_Spotted_Fever.yaml, 2,119 lines, MONDO:0019359	subtype RMSF, subtype_term MONDO:0019359
Boutonneuse fever	Boutonneuse_Fever.yaml, 1,273 lines, MONDO:0024472	subtype MSF, subtype_term MONDO:0024472

Spotted_Fever_Rickettsiosis (1,714 lines, MONDO:0001195 spotted fever) is in substance a grouping curated as a Disease: its six subtypes are six named diseases, two of which are already full entries. Under §3a's rule for promoted strata, this should be a Grouping with the members as Disease entries — the Treponematoses shape exactly. The same test flags Viral_Hemorrhagic_Fever (8 unbound subtypes, one of which duplicates the standalone Hantavirus_Hemorrhagic_Fever_with_Renal_Syndrome entry).

2.4 Phase-of-infection and pathogen-syndrome splits disagree with each other

• Legionella is split by syndrome into two entries (Legionnaires_Disease, Pontiac_Fever) — reasonable, they are genuinely different illnesses. But Pontiac_Fever carries infectious_agent: and Legionnaires_Disease does not, so the shared etiology is invisible to any query.
• HCV is split by phase: Acute_Hepatitis_C_Virus_Infection (MONDO:0100371) and Hepatitis_C (MONDO:0005231). The schema already has progression: (keyed on phase) and stages: for exactly this.
• HIV is split the other way: Acquired_Immunodeficiency_Syndrome exists, but there is no HIV-infection entry, so untreated chronic HIV infection has nowhere to live except inside the AIDS entry.
• SARS-CoV-2 spreads over four entries — COVID-19, Long_COVID, Multisystem_Inflammatory_Syndrome_in_Children_MIS-C, Seasonal_Coronavirus_Infection — of which only the middle two declare the virus.

2.5 Coverage gaps that follow from having no ladder

Because nothing says what level a microbial entry sits at, the queue has been filled opportunistically. Absent from both kb/ and stubs/:

typhoid fever (while Paratyphoid_Fever is curated), herpes zoster / shingles (while Chickenpox is), anthrax, diphtheria, amebiasis, histoplasmosis, listeriosis, salmonellosis, West Nile fever, yellow fever, rotavirus and norovirus gastroenteritis, sepsis, urinary tract infection, cellulitis, infectious osteomyelitis. Cryptosporidiosis and Zika exist only as stubs. All 40 infectious stubs are entry_type: UNDECIDED, meaning the lump/split call — the curator's first job per the stub queue rules — has not been made for any of them.

2.6 Life-cycle modelling stops at the parasitology entries

agent_life_cycle — which carries hosts (with a role for definitive / intermediate / reservoir / accidental / paratenic), vectors, and life_cycle_stages (OPL-bound) — is present on 15 entries and well populated on all 15: Babesiosis, Chagas_Disease, Chikungunya, Dengue, Human_African_Trypanosomiasis, Leishmaniasis, Lymphatic_Filariasis, Malaria, Myiasis, Nipah_Virus_Disease, Onchocerciasis, Schistosomiasis, Soil_Transmitted_Helminthiases, Taeniasis_Cysticercosis, Toxoplasmosis.

The pattern is right and the block is not the problem. The problem is that it stops at classical parasitology. At least 24 further entries with a non-human reservoir or an arthropod vector carry none: Plague (flea, rodent), Rocky_Mountain_Spotted_Fever, Boutonneuse_Fever, Spotted_Fever_Rickettsiosis, Scrub_Typhus, Murine_Typhus, Lone_Star_Virus_Infection, Tularemia, Brucellosis, Leptospirosis, both hantavirus entries, Rabies, Monkeypox, Ebola_Virus_Disease_EVD, Cat-Scratch_Disease, Glanders, Oroya_Fever, Melioidosis, Dracunculiasis, Cutaneous_Larva_Migrans, Polycystic_Echinococcosis, Hepatitis_E, Viral_Encephalitis, Viral_Hemorrhagic_Fever — plus Lyme_Disease, Q_Fever, Psittacosis and Southern_Tick-Associated_Rash_Illness, which have no infectious_agent block either.

This matters for the ladder specifically: reservoir host and vector are two of the five axes rung 4 promotion is decided on. Where they are unrecorded, the promotion question cannot be answered from the entry.

2.7 Identifier policy: NCBITaxon only, and that is mostly correct

What is in use. NCBITaxon, exclusively — 121 of the 125 flagged files, 411 occurrences, 165 distinct CURIEs, of which 151 are binomial or finer and 14 sit at genus (Aspergillus, Babesia, Brucella, Candida, Legionella, Leptospira, Shigella, Norovirus, …). It is properly wired: OrganismDescriptor binds term.id to the OrganismTerm dynamic enum (reachable_from NCBITaxon:1 plus NCBITaxon:10239 for viruses) at obligation_level: REQUIRED, validated cache-first against cache/ncbitaxon/terms.csv and cache/enums/organismterm_*.csv, with NCBITaxon: ols:ncbitaxon in conf/oak_config.yaml (OLS rather than a local build — the SQLite is ~13.5 GB, issue #5160).

GTDB is not used anywhere. No prefix in the schema, no adapter in oak_config.yaml, zero occurrences in kb/ or src/. Its only appearance in the tree is a semsql registry file inside .venv — a dependency's ontology catalog, not project configuration.

GenBank / RefSeq genome identifiers are not used for microbes. The GenBank-shaped strings in kb/ are all human reference sequences inside evidence snippets and prose in Mendelian entries — NM_007241.4, NC_000015.10, NG_016284.1, NC_012920.1 (mtDNA) — and none are structured identifiers in a slot. There are zero assembly accessions (GCA_/GCF_) anywhere in the repository.

Where NCBITaxon is sufficient. It reaches well below species. Every one of the seven unbound agent names in §1.3 resolves against NCBITaxon, including at serovar and serotype rank:

Unbound in the KB	Exists as
Clostridium tetani	NCBITaxon:1513
Treponema carateum	NCBITaxon:3712027
Human cytomegalovirus	NCBITaxon:10359 (Human betaherpesvirus 5)
S. enterica serovar Paratyphi A	NCBITaxon:54388 — serovar rank
E. coli O157:H7	NCBITaxon:83334 — serotype rank

Those seven are therefore curation misses, not ontology limits, and the fix is mechanical.

Where NCBITaxon genuinely runs out: pathotype. ETEC, EPEC and EAEC have no NCBITaxon class — a search returns only individual sequenced isolates (e.g. Escherichia coli 042, the EAEC reference strain). This is why all three Travelers_Diarrhea pathotypes collapse onto NCBITaxon:562. It is not a fixable gap in NCBITaxon: a pathotype is defined by virulence gene content, not by phylogeny, so no taxonomy will ever carry it. Rung 3 of the ladder handles this explicitly.

GTDB would not fix this and would make things worse. GTDB is a genome-based taxonomy that re-splits and re-merges NCBI taxa on average nucleotide identity. It has no pathotype concept either, and its species assignments diverge from the clinical names the KB is built on. Queried live against the GTDB API:

Shigella dysenteriae 1012 → gtdbTaxonomy: … g__Escherichia; s__Escherichia coli
Shigella sonnei 53G       → gtdbTaxonomy: … g__Escherichia; s__Escherichia coli

GTDB classifies Shigella as Escherichia coli. Adopting it as a binding vocabulary would collapse the agent of Shigellosis into E. coli and erase exactly the distinction the ladder exists to preserve. GTDB is the right tool for microbial phylogenomics and the wrong tool for naming what a patient has.

A GenBank/RefSeq assembly accession names one isolate, not a stratum. Recording GCA_000027125.1 for EAEC commits the KB to a single sequenced genome standing in for a pathotype defined by a gene repertoire that varies across isolates. That is a stronger claim than the evidence supports.

Recommended policy.

1. NCBITaxon remains the sole binding vocabulary for infectious_agent_term, hosts, and any organism reference. Bind the most specific NCBITaxon node that is correct, down to serovar or serotype where one exists.
2. Do not add GTDB as a prefix, adapter, or binding vocabulary. If a future need arises for genome-phylogenetic context, it belongs in a notes: line or a dedicated annotation, never in term.id.
3. Do not add genome or assembly accessions to infectious_agent. A specific sequenced isolate belongs in a datasets: record (where bioproject: / sra: prefixes already exist) or in an evidence snippet, not as the agent's identity.
4. Pathotype is carried structurally, not by an identifier — see rung 3. Leave subtype_term absent for a pathotype with no honest NCBITaxon node rather than binding the parent species. This is the dismech-terms rule that no term beats a bad one.
5. Add a pathotype: slot to InfectiousAgent (free text or a small controlled enum) so ETEC/EPEC/EAEC/EHEC and MRSA/MSSA are machine-queryable without a false ontology binding. InfectiousAgent currently carries only name, infectious_agent_term, food_source, evidence, description, has_subtypes — there is no strain, serovar, serotype, or pathotype slot anywhere in the schema.

---

3. Assessment

What is already right and should be preserved.

• The organism-defined core (cholera, measles, rabies, leprosy, tetanus, Buruli ulcer, the hepatitides A–E) is at the level the field actually uses, and is stable.
• Infection mechanism is correctly handled as modules, not taxa — intracellular_pathogen_persistence, granuloma_formation, innate_antiviral_interferon_response, molecular_mimicry_autoimmunity, the antibiotic/antiviral target modules. This mirrors the hallmark-module pattern §3a praised for cancer.
• Post-infectious sequelae are already separate entries with distinct mechanisms (Rheumatic_Heart_Disease, Postinfectious_Vasculitis, MIS-C, Postpoliomyelitis_Syndrome, Acute_Disseminated_Encephalomyelitis). That is the right call — the mechanism is host-immune, not microbial.
• Treponematoses is a model artifact. It states in its grouping_rationale exactly which axes justified splitting (transmission route, vertical transmission, organ systems, second-line drug susceptibility) and why the membership criterion is "a pathogenic treponeme" rather than "a T. pallidum subspecies". Every contested cluster in §2 needs a paragraph like that.

The three systematic deviations (deliberately parallel to §3a's three):

1. Syndrome-level entries used as taxa. Organ syndromes spanning unrelated organisms (Bacterial_meningitis, Infective_Endocarditis, Endophthalmitis, Choroiditis, Folliculitis, Chorioamnionitis, Travelers_Diarrhea, Viral_Encephalitis, Viral_Hemorrhagic_Fever) are curated as single diseases, most with no per-pathogen subtypes.
2. Grouping-shaped entries curated as Disease. Spotted_Fever_Rickettsiosis and Viral_Hemorrhagic_Fever list named diseases as subtypes, two of which are simultaneously standalone entries with the same MONDO IDs.
3. No recorded rule, hence: 131 parent strings for 125 entries, 0% use of hosts/vectors, 29% subtype binding, 17 MONDO mappings total, and 40 UNDECIDED stubs.

---

4. Proposed decision ladder (draft for design-decisions.md §3b)

Default level for a new microbial entry is the named clinical entity — the pathogen–syndrome pair that infectious-disease practice names, diagnoses and treats as a unit (cholera, Legionnaires' disease, Pontiac fever, Buruli ulcer). Neither the organism alone nor the organ syndrome alone is the default.

The ladder has five rungs. Rung 2 is the default; you move off it only by meeting that rung's stated test.

Rung 0 — Out of scope

An ontology class too abstract to carry a mechanism: MONDO:0005550 infectious disease, viral infectious disease, bacterial infectious disease, parasitic infectious disease.

Requirement: record as a stub with entry_type: OUT_OF_SCOPE and a notes: line giving the reason, so the concept is not re-nominated. Never a Disease entry.

Rung 1 — Grouping (kb/groupings/)

A curated union of Disease entries sharing an organism clade, a transmission mode, or a clinical/programmatic category. Treponematoses is the reference implementation.

Requirement	Level
Explicit members — no ontology-closure substitute	required
grouping_basis	required
grouping_rationale stating the axes on which the members were split, not only what they share	required
mappings.mondo_mappings with an explicit mapping_predicate (closeMatch where the MONDO class is defined differently — see Treponematoses)	required
membership_criteria	recommended
Its own pathophysiology	prohibited — a grouping is a union, not a disease

A grouping is also the correct home for an organ syndrome spanning unrelated organisms (bacterial meningitis, infective endocarditis, endophthalmitis) when the syndrome has no organism-independent mechanism of its own. Where it does have one, prefer a kb/modules/ module plus conforms_to on the member entries.

Rung 2 — Disease entry: the named clinical entity (the default)

Requirement	Level
disease_term bound to MONDO, at the entry's own scope	required
— where no exact class exists, anchor to the nearest and record mappings.mondo_mappings with skos:narrowMatch; never reuse a broader parent term bare	required
infectious_agent — at least one, each with an NCBITaxon-bound infectious_agent_term	required
transmission — at least one route	required
agent_life_cycle with hosts (each carrying a role), and vectors where an arthropod vector exists	required whenever a non-human reservoir or vector exists
life_cycle_stages (OPL-bound)	required for helminth and protozoan agents; recommended otherwise
At least one pathophysiology node specific to this entity	required
parents drawn from the controlled list (see §5 item 7)	required
classifications	recommended

Test for being on this rung rather than rung 1: the entity has a name the field uses, a diagnostic pathway, and a treatment; and its pathophysiology nodes are true of every case it covers. If a node has to be written vaguely enough to cover two organisms with different mechanisms, the entry is a rung-1 grouping wearing a Disease costume.

Rung 3 — has_subtypes within a Disease entry (the organism axis)

Strata below the entry go here: species → subspecies → serovar → serotype → pathotype / strain. A stratum earns a subtype when it is documented to differ from its siblings in any one of:

• clinical presentation or organ involvement
• diagnosis or diagnostic pathway
• first-line treatment or drug susceptibility
• prognosis or natural history
• transmission route, vector, or reservoir
• geography or at-risk population

One documented axis is enough for a subtype. (Two are needed to promote to a separate entry — rung 4.)

The threshold is documented difference, not nomenclatural availability, and the default is to lump. A taxonomy offering a name for a stratum is not a reason to create a subtype for it. M. tuberculosis lineages and HCV genotypes earn subtypes because the literature reports distinct clinical behaviour; the ~2,000 Salmonella serovars do not, and enumerating them would bury the handful (Typhi, Paratyphi A, Typhimurium) that genuinely behave differently. Where the differentiating evidence does not yet exist, keep the strata together in one entry and record in notes: that they are deliberately lumped.

Splitting is progressive and additive. A subtype added later, as pathogenicity, symptom or treatment evidence accrues, is the expected lifecycle of an entry — not a correction of an earlier mistake. This is why the rung-3 threshold is one axis rather than two: a subtype is cheap to add and cheap to revise, so the KB can track the literature as it develops rather than waiting for a stratum to be fully characterised before acknowledging it. The same logic runs upward — a subtype that accumulates a second differentiating axis becomes a rung-4 promotion candidate.

Requirement per subtype	Level
name — short and slug-friendly (ETEC, RMSF, Type 1); it is the foreign-key target for subtype: fields elsewhere in the file	required
display_name where the short name is too terse	recommended
description stating what differs, not just what the stratum is	required
subtype_term bound where an honest term exists — MONDO for a named disease variant, NCBITaxon for a species / subspecies / serovar / serotype	required if a term exists
evidence for the differentiating claim	required
subtype_frequency, geography, genes	recommended where known

Only 29% of existing infectious subtypes carry a subtype_term; that is the gap this rung's requirements close.

Rung 3a — Pathotypes and other non-phylogenetic strata

A pathotype is a stratum defined by virulence gene content rather than phylogeny: ETEC / EPEC / EAEC / EHEC / EIEC / ExPEC / UPEC within E. coli, MRSA / MSSA within S. aureus, toxigenic vs non-toxigenic C. difficile or C. diphtheriae. Pathotypes are the concrete case of "subtype by variant where it is clinically relevant" — and they usually are relevant, because they determine the toxin, the syndrome, and often the therapy.

No taxonomy carries them. NCBITaxon has no ETEC or EPEC class; GTDB has none either and would additionally collapse Shigella into E. coli (§2.7). So pathotype is carried structurally, not by an identifier:

Requirement	Level
A has_subtypes entry with name = the pathotype acronym	required
description naming the defining virulence determinants (LT/ST enterotoxin; bundle-forming pilus and the attaching-effacing lesion; Shiga toxin; mecA/SCCmec)	required
subtype_term bound only where a genuine node exists at serotype rank (E. coli O157:H7 → NCBITaxon:83334)	conditional
subtype_term omitted where the only candidate is the parent species — binding E. coli to ETEC is a false grounding	required
The defining determinant modelled as a pathophysiology node where the mechanism is curated	recommended
A genome or assembly accession (GCA_/GCF_) as the stratum's identity	prohibited — see §2.7
A pathotype: slot on InfectiousAgent, once added, populated in parallel	required after schema change

The same shape covers other non-phylogenetic strata: biotype (V. cholerae classical vs El Tor), lineage (M. tuberculosis L1–L9), viral genotype (HCV 1–6, HBV A–H) and clade (mpox I vs II). Bind where NCBITaxon has a node; otherwise carry it structurally and say so.

The rung-3 evidence gate applies here too, and this is where it bites hardest. A pathotype earns a subtype when its clinical features, pathogenicity or treatment are documented to differ — which for ETEC, EPEC, EAEC and EHEC they clearly are, since each maps to a different toxin, a different syndrome, and in the EHEC case a different management rule (antibiotics are avoided because they raise the risk of haemolytic uraemic syndrome). A serotype carrying a name but no reported clinical distinction stays lumped. The asymmetry is deliberate: pathotype is the stratum where sub-species variation most often carries clinical weight, and serovar is the stratum where it least often does.

Rung 4 — Promotion of a stratum to its own Disease entry

The Treponematoses test. Promote only when the stratum differs from its siblings in at least two of: transmission route or vector; reservoir host; natural history or tempo; organ systems ultimately involved; first-line therapy.

Requirement	Level
Name the two-or-more satisfied axes explicitly, in the entry or in the covering grouping's grouping_rationale	required
Own disease_term; where no exact MONDO class exists, skos:narrowMatch to the parent	required
A rung-1 Grouping covering the promoted siblings	required
The parent entry drops the promoted stratum from has_subtypes	required — no double-modelling
Everything in rung 2	required

The last row is the rule Spotted_Fever_Rickettsiosis currently violates: RMSF and boutonneuse fever are full entries and subtypes, carrying the same MONDO IDs in both places.

Orthogonal axes — never a rung

These are real distinctions that are modelled somewhere other than the taxonomy, and reaching for a new entry for any of them is the error:

Distinction	Belongs in
Phase of one infection (acute vs chronic; primary / latent / reactivated; colonisation vs invasive disease)	progression: on the single entry — which should list them, see below
Severity or stage	stages:, severity qualifiers
Organ syndrome across unrelated organisms	rung-1 Grouping, or a kb/modules/ module
Post-infectious immune sequela (rheumatic heart disease, MIS-C, post-infectious vasculitis, ADEM)	its own Disease entry — the mechanism is host-immune, not microbial — cross-linked to the index infection via parents
Infection-attributed neoplasm	§3a cancer ladder; the pathogen is an infectious_agent annotation, never a granularity axis
Mechanism shared across diseases	kb/modules/ + conforms_to
Antimicrobial resistance, where it does not change the syndrome	treatments / notes — but see rung 3a where it defines a pathotype (MRSA)

Phases are listed, not split off

The rule has two halves and the second is easy to lose: a phase is never its own entry, and the entry should enumerate its phases. Dropping the phase distinction is as wrong as promoting it to an entry.

progression: (ProgressionInfo) is the home. It already carries phase, subtype — so a phase can be scoped to one subtype — age_range, incubation_days / incubation_years, duration_days / duration, notes and evidence. That is enough to say everything an acute-vs-chronic split was being used to say.

Requirement	Level
progression entries for the phases the field recognises, where the disease has them	required
phase naming the phase as the field names it (Acute infection, Chronic infection, Latency, Reactivation, Active TB)	required
evidence for a phase's existence or duration where the claim is substantive	required
incubation_* / duration_* where reported	recommended
subtype where the phase applies to only one stratum	conditional

The mechanism is already in use and works: 48 of the 125 flagged entries carry a progression block, 44 of them with two or more phases. The gap is the other 77.

Where MONDO has a separate class for a phase, it is recorded as a mappings.mondo_mappings entry with skos:narrowMatch on the single entry — which retires the concept from the curation queue under the existing MONDO coverage rule, so nothing is lost by not making it an entry.

Decision procedure

1. Is the concept too abstract to carry a mechanism? → rung 0.
2. Is it a union of diseases the field already names separately? → rung 1.
3. Is it a named clinical entity with its own diagnosis and treatment? → rung 2. (default)
4. Is it an organism stratum below such an entity? → rung 3; check rung 3a if it is defined by gene content rather than phylogeny.
5. Does that stratum satisfy ≥2 Treponematoses axes? → rung 4, and add the covering grouping.
6. Is it a phase, stage, sequela, neoplasm, or mechanism? → not a rung; see the table above.

---

5. Recommended follow-up actions

Ordered by ratio of value to effort. Marked [M] where the work is microbial-specific and [G] where the same defect exists across the whole KB and should be tracked as a separate general issue.

#	Action	Scope
1	[M] Ratify the rung ladder in §4 as design decisions §3b	1 PR
2	[M] Backfill infectious_agent: (NCBITaxon-bound) on the 21+ microbial entries lacking it, starting with COVID-19, Lyme_Disease, Legionnaires_Disease and the nine respiratory-infection tranche entries	21 entries
3	[M] Bind the 7 free-text pathogen names — all seven exist in NCBITaxon (§2.7) — and add disease_term to Adenovirus_Respiratory_Infection and Lone_Star_Virus_Infection	9 entries
4	[M] Convert Spotted_Fever_Rickettsiosis and Viral_Hemorrhagic_Fever to Grouping records on the Treponematoses model (rung 1); keep RMSF and boutonneuse fever as the Disease entries they already are, and drop them from the parent's has_subtypes per rung 4	2 groupings, 2 entries edited
5	[M] Add rung-3 bound subtypes to Bacterial_meningitis, Infective_Endocarditis, Endophthalmitis, Chorioamnionitis, Travelers_Diarrhea, Choroiditis, Folliculitis — or convert them to rung-1 groupings	7 entries
6	[M] Add a pathotype: slot to InfectiousAgent, then populate the rung-3a strata: ETEC / EAEC / EPEC in Travelers_Diarrhea, MRSA/MSSA where S. aureus is the agent, toxigenic vs non-toxigenic C. difficile	1 schema PR + ~6 entries
7	[M] Re-anchor the 10 over-broad disease_term bindings, adding skos:narrowMatch mappings where no exact MONDO class exists	10 entries
8	[M] Add agent_life_cycle (with hosts, and vectors where an arthropod vector exists) to the ~24 zoonotic and vector-borne entries lacking it, since rung-4 promotion turns on reservoir and vector	~24 entries
9	[M] Decide entry_type on the 40 infectious stubs, and seed the coverage gaps in §2.5 (typhoid fever, herpes zoster, anthrax, diphtheria, sepsis, UTI and the rest)	queue work
10	[M] Retire Infectious_Disease (MONDO:0005550) under rung 0, recording the reasoning so the concept is not re-nominated	1 entry
11	[M] Add rung-1 groupings: Rickettsioses, Viral hepatitides, Soil-transmitted helminthiases, Enteric fevers, Arboviral haemorrhagic fevers	~5 groupings
11a	[M] Merge Acute_Hepatitis_C_Virus_Infection into Hepatitis_C as a progression phase, keeping MONDO:0100371 as a skos:narrowMatch mapping (decision 1)	2 entries
11b	[M] Add progression phases to the 77 flagged entries with no progression block, prioritising those with a clinically recognised phase structure — syphilis, TB, HIV, the hepatitides, Lyme disease, Chagas disease (decision 1)	~77 entries, triaged
11c	[M] Review the ten unstructured lumps in §1.3 individually against the rung-3 evidence gate; add subtypes only where the literature already reports a clinical difference, and record deliberate lumping in notes: where it does not (decision 2)	10 entries
12	[G] Normalise parents: to a controlled vocabulary and decide whether it should be ontology-bound rather than range: string. The microbial set alone uses 131 distinct strings for 125 entries, but 2,366 entries KB-wide use the slot	KB-wide
13	[G] Raise subtype_term binding rates KB-wide. 29% in the microbial set; measure and target the rest	KB-wide
14	[G] Audit for grouping-shaped Disease entries outside infection — the Spotted_Fever_Rickettsiosis pattern is the same one §3a found in cancer organ-umbrella entries	KB-wide

Items 2, 3 and 7 are mechanical and independently verifiable; item 1 should land first so items 4, 5, 6 and 9 have a rule to cite. Items 12–14 belong in a separate general issue so the microbial work is not blocked behind KB-wide refactors.

Item 11c is the one to resist rushing. Under decision 2 an undifferentiated lump is not a defect by definition — it is a defect only where the literature already carries the distinction. The output of 11c is as likely to be a notes: line recording a deliberate lump as it is a new subtype.

Two decisions taken

Both questions this review opened have been settled by the maintainers.

1. The ladder wins over MONDO on phase, and the entry lists its phases. A phase of one infection is never a separate Disease entry, but the entry it belongs to should enumerate its phases in progression: rather than leave the distinction unrecorded. Consequence for the KB: Acute_Hepatitis_C_Virus_Infection merges into Hepatitis_C as a progression phase, and MONDO:0100371 is retained on Hepatitis_C as a skos:narrowMatch mapping. The same treatment applies to any future acute/chronic, latent/reactivated, or colonisation/invasive pair that MONDO happens to class separately.

2. Rung 3 is gated on documented clinical difference, and lumping is the default. Individual serovars do not earn subtypes. Pathotypes with distinct clinical features do — ETEC, EPEC, EAEC and EHEC each map to a different toxin and syndrome, and EHEC to a different management rule. Everything else stays grouped in one entry until pathogenicity, symptom or treatment evidence accrues, at which point it is split. A subtype added later is the normal lifecycle of an entry, not a repair.

The second decision changes how the KB should read an undifferentiated lump. Babesiosis with five taxa and no subtypes is not automatically a defect — it is a defect only if the literature already distinguishes B. microti from B. divergens clinically, which it does (geography, severity, splenectomy risk). Rhinovirus_Infection with three species and no subtypes is probably correct as it stands. Each of the ten unstructured lumps in §1.3 needs that question asked of it individually rather than being treated as a uniform backlog.

Tracking

• #10115 — the microbial ladder and its work items (§4, §5 items 1–11).
• #10116 — the cross-KB defects this review surfaced (§5 items 12–14).
• #10113 — shared disease_term values without a skos:narrowMatch anchor; the same defect as §2.7's over-broad anchoring, reached from the mapping side.

---

Provenance

Generated by Claude Code on 2026-08-29 against main at 2,476 disorder entries, 99 groupings, 164 modules, 1,385 stubs. The microbial set was identified by the presence of an infectious_agent: block (125 entries) plus a keyword sweep of entry headers reviewed by hand (≥21 further entries). All counts in §1.3 were computed from the YAML, not estimated; the scripts are one-off and are not committed. No KB content was modified.

NCBITaxon term existence in §2.7 was checked live against the EBI OLS4 API; the GTDB Shigella placement was checked live against gtdb-api.ecogenomic.org. The vector-borne/zoonotic classification behind the “≥24 entries” figure in §2.6 was made by hand from the entry descriptions, not by a rule.

A first revision of this report stated that hosts:, vectors: and life_cycle_stages: were unused KB-wide. That was measured at the wrong nesting level — all three are slots of AgentLifeCycle, not of Disease, and are well populated on the 15 entries that carry an agent_life_cycle block. §1.3, §2.2 and §2.6 carry the corrected finding, which is one of partial coverage rather than disuse.