Detect one node_id carrying several node_types across records (#356)

[realmarcin]

Detection for #356. Baselined at 294 occurrences across 63 node_ids. No trait data changed — the burn-down is a separate campaign.

The only cross-record check here

Every other check in audit_causal_graphs.py is scoped to one graph. This defect isn't: two records disagree about what a node is, and neither is wrong read alone. That's why nothing caught it.

The issue understates the scale by ~7×

#356 reports proton_motive_force typed four ways across 9 records. Measured:

node_type	records
STATE	18
BIOLOGICAL_PROCESS	13
CHEMICAL	2
CAPACITY	2

35 records, one concept — and 63 node_ids disagree with themselves corpus-wide, membrane_potential also at four types.

Why it stopped being cosmetic

#355 minted powers (METPO:2007900) gated to subject_types = BIOLOGICAL_PROCESS|STATE. Two byte-identical assertions now behave differently purely by subject typing:

physiology/carboxydotrophic.yaml   proton_motive_force (STATE)     -> atp_synthase   grounds
physiology/phototrophic.yaml       proton_motive_force (CAPACITY)  -> atp_synthase   blocked_by_node_type

Typing decides groundings now. A disagreement is no longer just untidy.

Two design decisions worth reviewing

Reported per occurrence, not on a presumed-wrong minority. Nothing here knows which type is right — proton_motive_force splits 18/13, and the gradient genuinely is a state while generating it is a process. The majority is an observation, not a verdict. Per-occurrence rows also mean a family clears together the moment it's normalised.

The detail leads with node_id, deliberately not with the type set. _key takes the leading token as discriminator, so leading with the types would re-key every row of a family each time one member is fixed — un-suppressing rows nobody has reached yet, a burn-down that fights itself. That's FRAGMENTED_GRAPH's documented failure from the other direction, and there's a test pinning it.

Not every hit is a defect

terminal electron acceptor is deliberately both CHEMICAL and MOLECULAR_FUNCTION — mappings/node_grounding.tsv carries a row for each, noting one METPO class covers both senses and the MOLECULAR_FUNCTION typing "surfaces the role-of interpretation". Same two-senses shape as reduces (#330/#333) and the CAPACITY table.

So the playbook gets a decision table: does this id mean one thing? — normalise / decide-and-record / split into two node_ids. Where a family really means two things, the fix is two ids, not one type. It also carries #352's lesson forward, since the burn-down will need it: the test is not "is this type defensible in isolation" but "is it compatible with what the record and its predicates already assert".

just qc green · 534 tests pass (4 new) · ruff clean · history record per #325.

Stacked note: #365 (backlog reconcile) is open and its section 11 describes this work; no file overlap.

[culturebot-reviewer]

Claude encountered an error after 1s —— View job

---

I'll analyze this and get back to you.

[culturebot-reviewer]

Claude encountered an error after 1s —— View job

---

I'll analyze this and get back to you.