Repair the last 16 enables edges whose object cannot satisfy the range (#334)

[realmarcin]

Closes #334. ENABLES_RANGE_VIOLATION 16 → 0.

biolink gives enables (RO:0002327) the range biological process or activity, which of CausalNodeTypeEnum only BIOLOGICAL_PROCESS, PATHWAY and MOLECULAR_FUNCTION satisfy. #315 widened the audit from its original TRAIT-only test to the full range and surfaced 33 edges; #351 took 17. These are the last 16.

There was no single sweep

The issue said as much, and it's right: six idioms are mixed together, and the object type does not classify them — the same lesson reduces recorded when its two senses turned out not to be separable that way. Every decision lives in DECISIONS in the migration script, each with the sentence from the record that justifies it.

n	idiom	fix
1	the process the subject feeds was already in the graph	repoint, keep enables
1	the edge was backwards	reverse onto the existing process node
3	wrong node type, not wrong predicate	retype
3	an existing relation fits better	produces / has output / transports
2	never causal at all	part of / promotes
1	a presentation, not a consequence	manifests as
5	nothing upstream fits	mint powers

A few worth naming:

• cytoplasmic_na pointed at the Mrp complex while the record said "the substrate that sustains alkaliphile antiport activity" — and that antiport node sits three lines above it.
• molecular_oxygen enables terminal_oxidase_cytochrome_bd was backwards, and the node's own description said so: "high-affinity terminal oxidase enabling oxygen respiration at low O2 tensions."
• buoyancy was typed QUALITY here while intracellular_inclusion.yaml already typed the same concept TRAIT. That's a typing inconsistency of exactly The corpus types 'capacity to X' two ways: TRAIT in some records, CAPACITY/STATE in others #352's shape; resolved toward TRAIT.
• ftsW_flippase enables lipid_ii — a flippase flips its substrate. That's transports.

The one new term

METPO:2007900 powers, proposed in proposals/metpo_traitmech_v10 — an ion motive force, a proton motive force, or torque generation, operating a molecular machine.

RO has no relation for it because RO models energetics at the process level while all five objects are protein complexes. That is not a new observation here — it's the same gap METPO:2007804 (exports) and METPO:2007805 (imports) were minted for, whose note already reads "RO models transport at the process level, so no RO relation admits the GENE_OR_PROTEIN subject these edges use."

The label is not a coinage. All five edge descriptions already used the verb:

"Ion motive force powers rotation of the flagellar motor."
"Ion flux through stator complexes powers torque generation."
"Proton motive force can power gliding motility motors."

object_types is gated to GENE_OR_PROTEIN on purpose: an energetic driver feeding a process satisfies enables' range already and must stay on RO:0002327. Leaving the gate open would let this term absorb edges that aren't broken, and the violation it exists to fix would stop being visible.

Two gates widened deliberately, both recorded in the table: GENE_OR_PROTEIN added to transports' subjects (a flippase is a protein, not a pathway), and TRAIT added to confers' (a structural trait conferring a functional disposition — the corpus already carried it and lacked a subject type admitting it).

Structural edits, not textual

All 15 files round-trip byte-identically through emit_trait_yaml, verified before touching them, so they're parsed, modified and re-emitted. Earlier migrations spliced raw lines with regexes, broke YAML three times, and once turned "proper polar growth" into "proper powth".

Two mistakes I made and caught

Both in predicate_grounding.tsv, both found by reading the diff rather than the exit code:

1. csv.writer(escapechar='\\') rewrote the "" quote-doubling in four unrelated rows into \" — corrupting data it had no business touching, which is the failure this file exists to prevent. The tell was a 8-insertion/7-deletion diff for what should have been 2 edits and 1 addition.
2. I filled the new row's columns 4–6 from their names. They are the SSSOM mapping predicate, source ontology and match confidence — not the causal predicate.

Redone with QUOTE_MINIMAL and the columns copied off a real row. The diff is now exactly 2 changed rows + 1 added.

Measured, not predicted

ENABLES_RANGE_VIOLATION        16 -> 0
MICROBE_DOMAIN_ON_NONORGANISM   0 -> 0
audit-graphs                 1533 -> 1533   (218 FRAGMENTED, 1303 UNREACHABLE)

Graph health is unchanged — the repairs repointed edges without stranding a node. The only baseline churn is key moves: two evidence-snippet findings whose edge key changed, and ten gas_vesicle UNREACHABLE details that now name both TRAIT anchors because buoyancy became one. Same node set, same counts, verified by diffing them rather than trusting the totals.

just qc green · 502 tests pass · ruff clean · just verify-proposal metpo_traitmech_v10 PASS.

[culturebot-reviewer]

Claude finished @realmarcin's task in 5m 36s —— View job

---

Review — PR #355

Checked every one of the 16 repairs against the graph it lives in: node types on both ends of each new edge, the gate rows in mappings/predicate_grounding.tsv, the retypes' knock-on edges, and the baseline diffs. The gating all holds — powers (BP|STATE → GENE_OR_PROTEIN) matches all five subjects/objects, transports matches FtsW/lipid II, confers matches both retyped TRAIT objects, has output matches the PATHWAY subject, and the two repointed edges land on nodes that exist and are range-correct. METPO:2007900 is free in data/raw/metpo.owl and doesn't collide with the v8/v9 blocks. CI is green.

Two things to fix.

---

🟡 The reversed cytochrome bd edge kept evidence that no longer supports it

data/traits/environment/facultative_oxygen_preference.yaml:133-143

The description was rewritten to a new claim — "high-affinity terminal oxidase, sustaining aerobic respiration at the low O2 tensions where low-affinity oxidases cannot operate" — but the evidence block was left untouched:

- reference: DOI:10.1128/jb.00389-22
  notes: '"which of these terminal enzymes is active in electron transfer depends
    on the availability of the final electron acceptor: fumarate or oxygen"; supports
    electron-acceptor-dependent branch switching.'

That quote is about which oxidase is active given the available acceptor. It says nothing about affinity or about low O2 tensions, so the assertion the edge now makes is uncited — and the trailing clause "supports electron-acceptor-dependent branch switching" describes the edge that was just deleted. This is the one place in the PR where a claim changed rather than a predicate; everywhere else the description already stated what the new predicate says. The high-affinity reading comes from the node's own description, which is not itself sourced.

Either attach a reference that states the high-affinity/low-O2 property, or keep the reversal and write a description the retained quote actually supports.

Fix this →

---

🟡 powers grounds seven edges, not five

mappings/predicate_grounding.tsv:111 · proposals/metpo_traitmech_v10/metpo_proposal_properties_robot.tsv:3 · proposals/metpo_traitmech_v10/proposal.md

The new row also matches two pre-existing corpus edges that were never part of #334, and your own reports show it: reports/predicate_grounding_residual.tsv goes from powers 6 unmapped to powers 4 blocked_by_node_type, and reports/biolink_coverage.tsv moves powers from residual …6 to applied … METPO:2007900 plus residual …4. The two that got picked up are

• data/traits/environment/ph_optimum_mid2.yaml — proton_motive_force (STATE) → f0f1_atp_synthase (GENE_OR_PROTEIN)
• data/traits/physiology/carboxydotrophic.yaml — proton_motive_force (STATE) → atp_synthase (GENE_OR_PROTEIN)

Both read correctly under the new definition, so this isn't a wrong grounding — but three statements are now understated: traits_addressed says "covers 5 causal-graph edges across 4 trait records", the mapping note says "the five edges", and the round-trip plan says the placeholder is "confined to those two places" and "the five edges' predicate_id". When METPO mints the real ID, whoever follows that plan will miss the two edges a grounding pass will have stamped in the meantime.

Correcting the counts and naming the two extra records is enough; nothing needs regrounding.

---

🔵 The new gate is load-bearing on an inconsistency it didn't create

proton_motive_force is typed four different ways across the corpus — STATE in gliding.yaml/carboxydotrophic.yaml, BIOLOGICAL_PROCESS in electron_transfer.yaml/ph_delta_mid1.yaml, CHEMICAL in metabolism.yaml/oxygenic_photosynthesis.yaml, CAPACITY in phototrophic.yaml. That's why phototrophic.yaml's proton_motive_force powers atp_synthase stays blocked_by_node_type while the byte-identical assertion in carboxydotrophic.yaml grounds. Pre-existing, #352's shape, and out of scope here — but this PR is what makes it decide groundings, so it's worth an issue reference rather than leaving it to be rediscovered.

🔵 hfq -promotes-> rpos still describes itself as enabling translation

data/traits/physiology/stress_response.yaml:148-157. The rationale for the change is that the object is the sigma factor and not the translation — but the description still reads "Hfq enables sRNA-dependent translation of RpoS", which is the assertion the migration argued this edge doesn't make. True sentence, wrong edge.

Advisory

No history/records/ entry for a change touching 15 trait records. CI's curation-history check is warn-only and the inline curation_history blocks are thorough, so this is a note, not a defect.

---

The rest is solid work — in particular, retyping even_pg_distribution and treatment_resistance_persistence rather than rewriting their predicates is the right call and is the harder one to make, and parsing/re-emitting through emit_trait_yaml instead of splicing lines is visibly the correct instinct given what regex migrations did to this corpus before.

Requesting changes on the two 🟡 items.
· fix/334-enables-range

[culturebot-reviewer]

Two 🟡 items. (1) data/traits/environment/facultative_oxygen_preference.yaml:133-143 — the reversed cytochrome bd edge kept evidence (DOI:10.1128/jb.00389-22) that supports acceptor-dependent branch switching, not the rewritten claim about high affinity at low O2 tensions; the notes clause still describes the deleted edge. (2) The new powers row grounds seven edges, not five — reports/predicate_grounding_residual.tsv goes 6 unmapped -> 4 blocked_by_node_type, picking up ph_optimum_mid2.yaml (pmf -> f0f1_atp_synthase) and carboxydotrophic.yaml (pmf -> atp_synthase). Both read correctly, but the proposal's counts and its round-trip plan ('the five edges' predicate_id') would miss them at minting time. Full review in the PR comment.

[realmarcin]

All four correct. The second one I could have caught by reading my own reports.

🟡 The reversed cytochrome bd edge

You're right that this was the one place in the PR where a claim changed rather than a predicate, and I didn't hold it to the standard the other fifteen met — everywhere else the description already stated what the new predicate says, so the evidence carried over intact. Here I wrote a new claim and left the citation behind.

Fixed by writing what the retained quote supports, rather than attaching a citation I don't have:

Cytochrome bd is one of the terminal oxidases through which aerobic respiration proceeds; which terminal enzyme is active depends on the final electron acceptor available.

The evidence note's trailing clause — which still described the deleted edge — now describes this one. The high-affinity reading came from the node's own description, which isn't itself sourced, so it doesn't belong in an edge claim.

🟡 powers grounds seven edges, not five

Verified rather than taken on faith — exactly seven ground, and the two extra are the ones you named:

GROUNDS (7):   ph_optimum_mid2   proton_motive_force(BIOLOGICAL_PROCESS) -> f0f1_atp_synthase
               carboxydotrophic  proton_motive_force(STATE)              -> atp_synthase
               + the 5 from #334
BLOCKED (4):   aerobic, anoxygenic_photosynthesis, photosynthesis  <- process objects, correct
               phototrophic  proton_motive_force(CAPACITY)         <- typing, not design

Both read correctly under the definition, so nothing is regrounded. But you've identified the one that would have cost someone real time: the round-trip plan said the placeholder lives in "the five edges' predicate_id", so whoever swaps in the minted METPO ID would have missed two that a grounding pass had stamped in the meantime. Counts corrected in the mapping note, the ROBOT template and the proposal, with both records named.

Counting this also confirmed the gate does what it was designed to: three of the four blocked edges are blocked correctly — their objects are processes, which satisfy enables' range and belong on RO:0002327. That's the object_types gate earning its keep.

🔵 The gate is load-bearing on an inconsistency it didn't create

Agreed, and filed as #356 with the four typings laid out, referenced from the proposal. Your framing is the right one — a gate is the correct mechanism; what's wrong is that the corpus doesn't agree with itself about what a proton motive force is. I noted there that a check for one node_id carrying multiple node_types across records would catch this whole family and doesn't exist yet.

🔵 hfq

Caught cleanly: the migration's own rationale was that this edge asserts nothing about translation, and then I left the description saying it did. True sentence, wrong edge. Now states what promotes asserts.

Advisory on history/records/

Noted, and it's the open policy question in #325 rather than an oversight — #351, the previous half of this same issue, also added thorough inline curation_history and no history/records/ entry. If #325 lands as "enforce it", this change is one of the ones that should get a retrospective record.

just qc green · 512 tests pass · just verify-proposal metpo_traitmech_v10 PASS.

[culturebot-reviewer]

Claude finished @realmarcin's task in 3m 6s —— View job

---

Re-review — PR #355

Checked commit c8b30ab2 against the four items from the last round. Both 🟡 are resolved; both 🔵 too.

• 🟡 Cytochrome bd — description now matches the retained quote
• 🟡 powers counts corrected to 7 across 6 records
• 🔵 hfq -promotes-> rpos description
• 🔵 proton_motive_force is typed four different ways, and that now decides groundings #356 filed and referenced from the proposal
• Re-verified reports, baselines, rendered pages, CI

---

🟡 Cytochrome bd — resolved

The description is now "Cytochrome bd is one of the terminal oxidases through which aerobic respiration proceeds; which terminal enzyme is active depends on the final electron acceptor available." The second clause is what DOI:10.1128/jb.00389-22 actually says; the first is the node's own definition, not a sourced property. The unsourced high-affinity/low-O2 claim is gone, and the notes clause now describes the surviving edge rather than the deleted one. Writing to the quote instead of hunting a citation for the stronger claim was the right call.

🟡 powers counts — resolved, and I re-counted independently

Eleven powers edges in the corpus; four residual per reports/predicate_grounding_residual.tsv, so seven ground. Records: motility, gliding, flagellated, motile (×2), plus ph_optimum_mid2 and carboxydotrophic — 7 edges, 6 records, matching what the proposal now says. The four blocked are as described: aerobic (atp_synthesis, BIOLOGICAL_PROCESS), photosynthesis (electron_transport, BIOLOGICAL_PROCESS), anoxygenic_photosynthesis (photosynthetic_etc, PATHWAY) — all three correctly blocked, their objects satisfy enables' range — and phototrophic, blocked by its CAPACITY subject, which is #356.

(phototrophic.yaml not appearing in that report row is the example_files cap of 3 in scripts/ground_causal_predicates.py:224, not a miscount.)

The round-trip plan — the part that would actually have cost someone — now reads "seven edges' predicate_id (the five from #334 plus the two named above)".

🔵 Two residuals in the ROBOT template's observations

proposals/metpo_traitmech_v10/metpo_proposal_properties_robot.tsv:3. The traits_addressed column is corrected, but the free-text column beside it wasn't fully brought along:

1. "RO models energetics at the process level while all five corpus objects are protein complexes" — contradicts covers 7 causal-graph edges two columns to its left in the same row. All seven objects are protein complexes, so it's just the stale number.
2. The example quote got detached from its clause: "…every one of the five 33 enables edges point at non-activity objects, violating RO:0002327's range #334 edge descriptions already used the verb 'power'; the gate also picks up two pre-existing proton_motive_force -> ATP synthase edges ('Ion motive force powers rotation of the flagellar motor')." That quote is motile.yaml's, one of the five — it now reads as an example of the two pre-existing edges, which it isn't.

proposal.md got both of these right; only the ROBOT row is stale. It's the artifact that goes upstream, which is the only reason it's worth naming.

Verification

Key moves only in the baselines, confirmed by reading them: molecular_oxygen->terminal_oxidase_cytochrome_bd → terminal_oxidase_cytochrome_bd->aerobic_respiration and cytoplasmic_na->mrp_antiporter_complex → cytoplasmic_na->extreme_alkaliphile_antiport in conf/evidence_snippet_baseline.tsv, no findings added or dropped. Both changed YAMLs have regenerated pages in the same commit. CI green across qc, pytest, validate-strict, pr-sanity, label-correspondence, curation-history.

---

Approving. The 🔵 is a two-word correction in a free-text column, not a reason to hold this. Sixteen edges closed with the object type explicitly rejected as the classifier, every decision carrying the sentence from the record that justifies it, and a new term whose gate is narrow enough that the violation it fixes stays visible — that last part is the thing most easily gotten wrong, and it wasn't.
· fix/334-enables-range

[culturebot-reviewer]

Both 🟡 items resolved.

Cytochrome bd — the description now states what DOI:10.1128/jb.00389-22 supports (acceptor-dependent terminal branches) instead of the unsourced high-affinity/low-O2 claim, and the notes clause describes the surviving edge rather than the deleted one.

powers counts — re-counted independently: 11 powers edges corpus-wide, 4 residual, so 7 ground across 6 records, matching the corrected proposal. Three of the four blocked are blocked correctly (process/pathway objects belong on RO:0002327); the fourth is #356. The round-trip plan now names seven edges. (phototrophic.yaml missing from the residual report row is the example_files cap of 3 in scripts/ground_causal_predicates.py:224, not a miscount.)

🔵 One residual, non-blocking: proposals/metpo_traitmech_v10/metpo_proposal_properties_robot.tsv:3 — the observations column still says 'all five corpus objects are protein complexes' (contradicting covers 7 in the adjacent column), and the example quote 'Ion motive force powers rotation of the flagellar motor' now sits after the clause about the two pre-existing edges, reading as an example of them when it is motile.yaml's. proposal.md got both right; only the upstream-bound ROBOT row is stale.

Baseline churn verified as key moves only, pages regenerated in the same commit, CI green.

[culturebot-reviewer]

Re-review — PR #355 (new commit)

• Read the new commit c43d1a85 (history record + history/README.md)
• Check the history record against the schema / existing records
• Verify the record's factual claims against the diff it describes
• Re-check the outstanding 🔵 on the ROBOT template row
• Confirm nothing else regressed

View job run

[culturebot-reviewer]

Claude finished @realmarcin's task in 3m 3s —— View job

---

Re-review — PR #355 (commits c43d1a85, 57e2321a)

• Read the two new commits (history record; history/README.md revert)
• Check the history record against schema / existing records
• Verify the record's factual claims against the diff it describes
• Confirm history/README.md is byte-identical to main
• Re-check the outstanding 🔵 on the ROBOT template row
• Confirm nothing else regressed

The README revert is clean — git diff origin/main...HEAD -- history/README.md is empty, so the net effect of the two commits is the record and nothing else. The record validates (just validate-history → no issues), its layout matches history/<kind-dir>/<slug>/<ts>-<actor>-<shortid>.yaml, and its shape matches history/records/sulfur_globule/. The advisory presence check counts added history/**/*.yaml, so it's satisfied; Curation history is green along with the rest of CI.

Targeting the script rather than the 15 trait records is the right call, and the commit message makes the argument for it well.

One thing, and it's in the one place that can't be fixed later.

---

🟡 The record's idiom tally sums to 18, for a change it says repaired 16 edges

history/infrastructure/enables-range-migration/2026-08-08T003430Z-claude-code-9da277.yaml:26-28

Six idioms: 4 repointed at a process already in the graph, 1 reversed …, 3 retyped …, 3 moved to produces/has output/transports, 2 to part of/promotes, and 5 onto a new term METPO:2007900 powers.

4+1+3+3+2+5 = 18, two lines below Repaired 16 edges across 15 trait records. Counting the 16 entries in DECISIONS:

n	idiom	edges
1	repoint, keep enables	ph_range_high cytoplasmic_na
1	reversed	facultative_oxygen_preference
3	retyped	human_pathogen, gas_vesicle (both → confers), cell_width_large (keeps enables)
3	produces / has output / transports	piezophilic, chemolithoheterotrophic, sphere_shaped
2	part of / promotes	filament_shaped, stress_response
5	powers	the five
1	manifests as	intracellular_inclusion → METPO:2007400

Two errors, and they nearly cancel: the repoint count is 1 rather than 4, and the manifests as edge is missing from the enumeration entirely. Corrected, it reads 1+1+3+3+2+5+1 = 16.

Worth fixing now rather than shrugging at, because history/README.md says a record is "written once and never edited afterwards" — a correction after merge costs a whole second record. It's one sentence.

(The PR body's table carries the same two errors; that one is editable.)

Fix this →

---

Everything else the record claims checks out

• "16 edges across 15 trait records" — 16 DECISIONS entries over 15 distinct files. ✔
• "Selection rule: every edge reported … as ENABLES_RANGE_VIOLATION" — conf/predicate_domain_audit_baseline.tsv drops exactly 16 rows, all ENABLES_RANGE_VIOLATION, and their file/graph pairs are the DECISIONS keys with motile.yaml appearing twice as it should. ✔
• "audit-graphs unchanged at 1533" — baseline is 1533 lines, 11 changed / 0 net. ✔
• Vocabulary (EDIT/changed, kind: infrastructure, explicit --path for a non-record kind) follows the README's own rules. ✔

🔵 Still open from last round

proposals/metpo_traitmech_v10/metpo_proposal_properties_robot.tsv:3 — observations still reads "all five corpus objects are protein complexes" against covers 7 causal-graph edges two columns left, and the motile.yaml quote still sits after the clause about the two pre-existing edges. Unchanged since I raised it; still non-blocking, still the row that goes upstream.

---

Requesting changes on the 🟡 only because the file it's in is the one this repo declares immutable after write. Nothing else in the two commits needs touching.

· fix/334-enables-range ·

[culturebot-reviewer]

🟡 The new history record's idiom tally sums to 18 for a change it says repaired 16 edges.

history/infrastructure/enables-range-migration/2026-08-08T003430Z-claude-code-9da277.yaml — 4 repointed … 1 reversed … 3 retyped … 3 produces/has output/transports … 2 part of/promotes … 5 powers = 18, two lines below Repaired 16 edges across 15 trait records. Per DECISIONS in the migration script the repoint count is 1, not 4, and the manifests as edge (intracellular_inclusion -> METPO:2007400) is missing from the enumeration entirely. Corrected it reads 1+1+3+3+2+5+1 = 16.

Raising it rather than shrugging because history/README.md says a record is 'written once and never edited afterwards' — a correction after merge costs a second record. It is one sentence. The PR body's table carries the same two errors.

Everything else in the two commits checks out: the README revert is a true no-op against main, the record validates and matches the layout and shape of existing records, the presence check is satisfied, and the record's other claims (16 edges / 15 files, the ENABLES_RANGE_VIOLATION selection rule, audit-graphs unchanged at 1533) verify against the diff. CI green. Targeting the migration script rather than the 15 trait records is the right call.

🔵 Still open, non-blocking: proposals/metpo_traitmech_v10/metpo_proposal_properties_robot.tsv:3 observations still says 'all five corpus objects' against 'covers 7' in the adjacent column, and the motile.yaml quote still reads as an example of the two pre-existing edges.

Full review in the PR comment.

[realmarcin]

Correct, and thank you for catching it in the one artifact where it couldn't be fixed later.

Recounted from DECISIONS, which is the source of truth:

n	idiom	records
1	repoint, keep enables	ph_range_high
1	reversed	facultative_oxygen_preference
3	retyped	human_pathogen, gas_vesicle, cell_width_large
3	produces / has output / transports	piezophilic, chemolithoheterotrophic, sphere_shaped
2	part of / promotes	filament_shaped, stress_response
1	manifests as	intracellular_inclusion
5	powers	the five
16

Your diagnosis was exact — two errors that nearly cancelled: the repoint count was 1, not 4, and the manifests as edge was missing from the enumeration entirely. Fixed in the record now rather than after merge, for the reason you gave: history/README.md says a record is written once and never edited, so a correction afterwards costs a whole second record.

The PR body table is corrected too.

One thing you didn't flag, and it's the worse instance. The same two errors were in docs/CURATION_PLAYBOOK.md, which asserted "four of #334's sixteen needed only repointing" — as guidance for future curators, in the section telling them what to reach for. A wrong count in a record is a wrong record; a wrong count in the playbook actively misdirects. Now reads: one needed only that, and a second was fixed by reversing it onto a node already present.

just qc green.

[culturebot-reviewer]

Claude finished @realmarcin's task in 2m 48s —— View job

---

Re-review — PR #355 (commit ef340806)

• Read the new commit
• Verify the corrected tally against DECISIONS
• Check the playbook correction
• Re-check the outstanding 🔵 on the ROBOT template row
• Confirm just validate-history and nothing else regressed

The 🟡 from last round is fixed. The record now reads "Seven idioms: 1 repointed … 1 reversed … 3 retyped … 3 moved to produces/has output/transports, 2 to part of/promotes, 1 to manifests as, and 5 onto … powers" — 1+1+3+3+2+1+5 = 16, and I re-derived each number from DECISIONS rather than from the commit message: 16 entries over 15 files, one object-only repoint (ph_range_high), one subject rewrite (facultative_oxygen_preference), three retype blocks, five powers. just validate-history → no issues. CI green on all seven runs. No data artifact moved in this commit.

But the playbook paragraph is where I should have looked harder last round, because two more counts in it are wrong — and they are wrong in the direction that matters most, since you correctly identified this file as the one that misdirects rather than merely misrecords.

---

🟡 Two more #334 counts in the playbook, both new in this PR

docs/CURATION_PLAYBOOK.md:177-186

Both sentences were added by aeec2f3a, both sit in the paragraph ef340806 just edited, and both disagree with DECISIONS.

1. "In #334 five of the sixteen edges had the right predicate and the wrong node type" (line 177). Three edges had the wrong node type — treatment_resistance_persistence, buoyancy, even_pg_distribution — and of those exactly one had the right predicate: cell_width_large, which keeps enables. The other two moved to confers as well as being retyped. The sentence's own second example undercuts it: "tolerance to antimicrobial treatment" is treatment_resistance_persistence, whose predicate did change, so it is not an instance of "right predicate, wrong node type."

The distinction is the whole point of the heading — "Check the node type before you change the predicate" — so the number a curator reads there should be the one that supports it: three had the wrong node type, and one of those needed nothing else.

2. "Twice in #334 the edge was simply backwards" (line 185). Once. facultative_oxygen_preference is the only entry in DECISIONS that rewrites subject, and it is the one the sentence then quotes. Your own corrected line 194 says as much nine lines below: "a second was fixed by reversing it onto a node already present" — one reversal, described from the other side. As written the two sentences claim three reversal-or-repoint edges where there are two.

Fix this →

---

🔵 Still open (third round, unchanged)

proposals/metpo_traitmech_v10/metpo_proposal_properties_robot.tsv:3 — observations still reads "all five corpus objects are protein complexes" against covers 7 causal-graph edges across 6 trait records two columns to its left, and motile.yaml's quote still sits after the clause about the two pre-existing edges. proposal.md has both right (lines 46, 76, 133). Only the row that goes upstream is stale.

🔵 "Seven idioms" vs the script's "Six"

The record now says seven; scripts/migrate_enables_range_edges.py:10 still says "Six distinct idioms are mixed together" and its comments number them 1–6, folding manifests as into idiom 6 with the retypes. The record's finer split is the more accurate one and it's the record that's immutable — but the script is what it cites as the source of truth, so one of the two words should move. Genuinely optional.

---

Requesting changes on the 🟡 only. The record — the part that can't be fixed after merge — is correct now, and its tally is verifiable line-by-line against DECISIONS, which is what a provenance record should be.

· fix/334-enables-range ·

[culturebot-reviewer]

🟡 Two more #334 counts in docs/CURATION_PLAYBOOK.md, both added by this PR and both in the paragraph ef340806 just corrected.

Line 177: "five of the sixteen edges had the right predicate and the wrong node type" — per DECISIONS three edges were retyped (treatment_resistance_persistence, buoyancy, even_pg_distribution) and only one of those kept its predicate (cell_width_large, which stays on enables); the other two also moved to confers. The sentence's own second example, "tolerance to antimicrobial treatment", is treatment_resistance_persistence — whose predicate did change — so it is not an instance of the claim it illustrates.

Line 185: "Twice in #334 the edge was simply backwards" — once. facultative_oxygen_preference is the only DECISIONS entry that rewrites subject, and the corrected line 194 nine lines below already describes that same edge from the other side.

This is the file you rightly called the worse place for a wrong count, since it is guidance rather than record.

The 🟡 from the last round is resolved: the record now reads 1+1+3+3+2+1+5 = 16, re-derived independently from DECISIONS (16 entries / 15 files, one object-only repoint, one subject rewrite, three retype blocks, five powers). just validate-history clean, CI green on all seven runs, no data artifact touched by this commit.

🔵 Still open, non-blocking: proposals/metpo_traitmech_v10/metpo_proposal_properties_robot.tsv:3 observations still says 'all five corpus objects' against 'covers 7' in the adjacent column, and the motile.yaml quote still reads as an example of the two pre-existing edges. 🔵 The record now says 'Seven idioms' while the script it targets still says 'Six distinct idioms' and numbers them 1–6.

Full review in the PR comment.

[realmarcin]

Both correct, both mine, and both in the paragraph I'd just edited without re-reading the rest of it.

1. "five of the sixteen edges had the right predicate and the wrong node type." Three had the wrong node type, and of those exactly one also had the right predicate — cell_width_large, which keeps enables. The other two moved to confers as well.

Your sharpest point is that the sentence's own second example undercut it: "tolerance to antimicrobial treatment" is treatment_resistance_persistence, whose predicate did change, so it was never an instance of the thing the sentence claimed. Rewritten so the number supports the heading it sits under.

2. "Twice in #334 the edge was simply backwards." Once. Only facultative_oxygen_preference rewrites a subject, and it's the one the sentence then quotes — and line 194, which I'd corrected nine lines below, already said so from the other side.

This is the third round of counting errors in the same file, so I stopped counting by hand and re-derived every #334 claim the playbook now makes straight from DECISIONS:

three of the sixteen had wrong node type   True
one of those needed only a retype          True
once backwards                             True
one needed only repointing                 True

That check belonged before the first correction, not after the second — and you're right that this file is where it matters, since a wrong number here misdirects a future curator rather than merely misrecording what happened.

just qc green.

[culturebot-reviewer]

Claude finished @realmarcin's task in 2m 8s —— View job

---

Re-review — PR #355 (commit 80d8da92)

• Read the new commit
• Re-derive every #334 count in the playbook from DECISIONS
• Re-check the outstanding 🔵 items
• Confirm nothing else regressed

Both 🟡 from last round are fixed, and I re-derived the corrected numbers from DECISIONS rather than from the commit message:

playbook claim	DECISIONS
three of the sixteen had the wrong node type	3 retype blocks — treatment_resistance_persistence, buoyancy, even_pg_distribution ✔
for one of those the predicate was already right	cell_width_large has no predicate key, keeps RO:0002327 ✔
the other two needed confers as well	both carry METPO:2007700 ✔
once backwards	one entry rewrites subject ✔
one needed only repointing	one entry rewrites object alone ✔

Both illustrative quotes are verbatim against the node descriptions they come from, and the second one is no longer an instance of a claim it contradicts. The high-affinity phrase on line 188 is quoting the node's own description as evidence of direction, which is the right use of it and consistent with how the cytochrome bd edge was resolved. just qc, pytest, validate-strict, history, pr-sanity, vendored-sync all green; no data artifact touched by this commit.

---

🔵 The migration script still carries the description that was removed as uncited

scripts/migrate_enables_range_edges.py:68-70. The DECISIONS entry for the reversed cytochrome bd edge still writes

Cytochrome bd is a high-affinity terminal oxidase, sustaining aerobic respiration at the low O2 tensions where low-affinity oxidases cannot operate.

which is the sentence c8b30ab2 took out of the YAML for being unsupported by the retained quote. The corpus is correct; the script isn't, and re-running it wouldn't reveal that (it would fail MISSING EDGE on the old subject/object).

Worth a line only because this file is target.path in an immutable history record and is what you and I have both been treating as the source of truth for what this migration decided. Same class as the ROBOT row below — stale text in an artifact, not wrong data.

🔵 Still open, unchanged (fourth round)

proposals/metpo_traitmech_v10/metpo_proposal_properties_robot.tsv:3 — observations still says "all five corpus objects are protein complexes" against covers 7 causal-graph edges across 6 trait records two columns to its left, and motile.yaml's quote still sits after the clause about the two pre-existing edges. proposal.md has both right.

🔵 "Seven idioms" vs "Six distinct idioms"

scripts/migrate_enables_range_edges.py:10 still numbers them 1–6 with manifests as folded into idiom 6, while the record now says seven.

---

Approving. The playbook now says what DECISIONS says, which is the thing that mattered — a wrong count there misdirects, and this was the file where it had gone wrong three times. The three remaining items are free-text staleness in artifacts whose data is correct; fix them if you're pushing again, but none is a reason to hold the merge.
· fix/334-enables-range

[culturebot-reviewer]

Both 🟡 from the last round are fixed, and I re-derived the corrected playbook numbers from DECISIONS rather than from the commit message: three retype blocks (treatment_resistance_persistence, buoyancy, even_pg_distribution), exactly one of which keeps enables (cell_width_large) while the other two carry METPO:2007700; one subject rewrite; one object-only repoint. Both illustrative quotes are verbatim against the node descriptions they come from, and the second is no longer an instance of a claim it contradicts. CI green; no data artifact touched by this commit.

🔵 Three non-blocking residuals, all free-text staleness in artifacts whose data is correct:

1. scripts/migrate_enables_range_edges.py:68-70 — the DECISIONS entry for the reversed cytochrome bd edge still writes the 'high-affinity ... low O2 tensions where low-affinity oxidases cannot operate' description that c8b30ab removed from the YAML as unsupported by the retained quote. The corpus is right; the script disagrees with it, and re-running would fail MISSING EDGE rather than surface that. Worth a line only because this file is target.path in an immutable history record and is the cited source of truth for what the migration decided.
2. proposals/metpo_traitmech_v10/metpo_proposal_properties_robot.tsv:3 — observations still says 'all five corpus objects are protein complexes' against 'covers 7 causal-graph edges across 6 trait records' two columns left, and motile.yaml's quote still reads as an example of the two pre-existing edges. proposal.md has both right.
3. scripts/migrate_enables_range_edges.py:10 still says 'Six distinct idioms' and numbers them 1-6 while the record now says seven.

Approving. The playbook now says what DECISIONS says, which is what mattered — a wrong count there misdirects a curator rather than merely misrecording history, and that is the file it had gone wrong in three times.