Reconcile NEXT_TASKS.md (2026-08-08)

[realmarcin]

The file said "everything merged through #272"; main is at #363. 41 PRs merged since the 2026-08-05 reconcile.

Reconciled

• Closed since last reconcile: Cross-Mech: vendored-sync paths filter omits chem_formula.py and the three id_label tests in every repo #198, No workflow-authoring conventions page, so concurrency lessons keep being relearned per workflow #217, An absolute machine path is baked into the front matter of 342 committed reports #248 (→Record the template as a repo-relative path (#248) #311), Nothing checks that qc.yaml's paths filter covers what the qc chain reads — third recurrence #252 (→Gate the qc paths filter against what qc actually reads (#252) #285), Snippet baseline keys on an array index, so improving the corpus can fail qc #270 (→Key the snippet baseline by occurrence count, not by array index (#270) #290), docs/WORKFLOW_CONVENTIONS.md has no inbound link, and duplicates the comments it restates #275 (→Link every workflow to the conventions page, and gate it (#275) #308), NEXT_TASKS.md reconciles headers without its section bodies, so the file asserts both halves of a contradiction #283.
• Newly filed: audit-qc-paths does not follow a chain recipe's own dependencies when it also has a body #289, Editing a REUSED_SNIPPET's shared text reads as a new finding #292, proton_motive_force is typed four different ways, and that now decides groundings #356, Vendored history.yaml still states the pre-#325 enforcement policy #358, METPO has no generic salt-tolerance or low-pH-tolerance disposition #364.
• Header rewritten to say what those 41 PRs actually were — one thread, not a list: the corpus disagreeing with itself about types and predicates, run as detect → propose → migrate → burn down, three times over.

Measured rather than relayed

Section 5's numbers were from 2026-07-30 and predate #294/#300/#351/#360. Re-measured: 218 graphs still split, 1296 UNREACHABLE_FROM_TRAIT — barely moved in five weeks. Added the number that should replace it now that #363 exists: 353 graphs, 861 components over 4129 wired nodes, 69.7% attached.

#356 understates itself by 7×. The issue reports proton_motive_force typed four ways across 9 records. Measured today it is 35 records (STATE 18, BIOLOGICAL_PROCESS 13, CHEMICAL 2, CAPACITY 2) — and 63 node_ids carry more than one node_type corpus-wide, including membrane_potential at four. New section 11 records this, the shape of the fix, and the trap: the test is not "is this type defensible in isolation" but "is it compatible with what the record and its predicates already assert" — which #360 took three review rounds to learn.

Section 7 records #358's ordering constraint where someone will actually find it: ../culturebotai-claw is checked out locally so the canonical fix is actionable today, but adding history.yaml to check_vendored_sync.sh still waits on #209.

Recommendation carried in the file

Next: #356. Same shape as #352/#353, machinery fresh from three consecutive successes at exactly this, and #356 itself names the missing check. Runner-up: #358. Not actionable as "next": #183's backfill (wants its own campaign, and now has a metric that can't be gamed), #364 (upstream METPO), #191/#197/#209 (cross-Mech, want the hub).

Docs-only, so path-filtered workflows may report no checks — that's MERGEABLE/CLEAN, not a failure.

[culturebot-reviewer]

Claude finished @realmarcin's task in 4m 27s —— View job

---

Review — approved

Every measured claim spot-checked against the tree at c9032b3b. All exact.

claim	verified
63 node_ids with >1 node_type	63 ✓
proton_motive_force 35 records, STATE 18 / BIOLOGICAL_PROCESS 13 / CHEMICAL 2 / CAPACITY 2	✓ all four
membrane_potential 4 types, membrane_lipid_composition 3, superoxide_dismutase GENE_OR_PROTEIN/MOLECULAR_FUNCTION	✓
1296 UNREACHABLE_FROM_TRAIT, 218 split graphs	conf/causal_graph_audit_baseline.tsv: 1296 / 218 ✓
DISPOSITION_MISTYPED and DUPLICATE_GROUNDING at 0, predicate domain/range at 0	0 / 0, and predicate_domain_audit_baseline.tsv is header-only ✓
353 graphs, 861 components, 4129 wired, 69.7%	353 rows, 135 single-component → 218 split; totals match #363's own regeneration note ✓
#358's premise — history.yaml states the pre-#325 policy	lines 9 and 20-23, verbatim ✓
the powers consequence	mapping row gates subject_types=BIOLOGICAL_PROCESS|STATE, object_types=GENE_OR_PROTEIN; 11 powers edges, 4 blocked; phototrophic's CAPACITY-subject edge is byte-identical to carboxydotrophic's STATE-subject one modulo typing ✓

No 🔴, no 🟡. The re-measurement was the right call — section 5's numbers really were five weeks stale, and #356 really is 7× its filed size.

🔵 Three suggestions

1. reports/predicate_grounding_residual.tsv won't show what section 11 points at. That report aggregates per predicate label — row 42 is powers / 4 / blocked_by_node_type, and its example_files column lists aerobic.yaml|anoxygenic_photosynthesis.yaml|photosynthesis.yaml, with no node_id anywhere in the file. phototrophic.yaml is the unlisted 4th. A reader following the pointer finds neither half of the pair; naming the two files directly would be more useful than the pointer.

2. Header omits v10. The domain/range bullet credits "METPO proposals v8 (METPO proposal v8: trait-realization and electron-role predicates (#302, #303) #320) and v9 (METPO proposal v9: causal-graph counterparts for the microbe-domain family (#301) #326)", but METPO:2007900 — the term section 11's argument rests on — is proposals/metpo_traitmech_v10, via Repair the last 16 enables edges whose object cannot satisfy the range (#334) #355. Header and section 11 currently disagree about where that term came from.

3. The same node_ids disagree on label, not just type. data/traits/environment/ph_phenotype_with_numerical_limits.yaml:52 labels proton_motive_force as "proton motive force across cell envelope" where the other 34 records say "proton motive force"; membrane_potential carries three distinct labels across four records ("membrane potential (Δψ)", "transmembrane potential (delta-psi)", "membrane potential"). Section 11 step 1 scopes INCONSISTENT_NODE_TYPE to node_type only, so this axis stays invisible — cheap to fold in while the cross-record index is being built.

Not verified

Issue open/closed states (#198, #217, #248, #252, #270, #275, #283 closed; #289, #292, #356, #358, #364 open) — gh issue list / gh api are not permitted in this sandbox. The table's own arithmetic is self-consistent: 17 − 7 + 5 = 15, and the table has 15 rows. Likewise "../culturebotai-claw is checked out locally" is a statement about the author's working copy, not checkable from CI.

Docs-only, no data/traits/** change, so no history/ record is expected.
· branch chore/reconcile-next-tasks-2026-08-08

[culturebot-reviewer]

Spot-checked every measured claim against the tree at c9032b3; all exact.

• 63 node_ids carry >1 node_type (verified corpus-wide), proton_motive_force 35 records STATE 18 / BIOLOGICAL_PROCESS 13 / CHEMICAL 2 / CAPACITY 2, membrane_potential 4, membrane_lipid_composition 3, superoxide_dismutase GENE_OR_PROTEIN/MOLECULAR_FUNCTION.
• Baseline counts: UNREACHABLE_FROM_TRAIT 1296, FRAGMENTED_GRAPH 218, DISPOSITION_MISTYPED 0, DUPLICATE_GROUNDING 0, predicate-domain baseline header-only. Connectivity report: 353 rows, 135 single-component -> 218 split; the 861/4129/69.7% figures match #363's own regeneration note.
• history.yaml lines 9 and 20-23 do state the pre-#325 policy verbatim, so #358's premise holds.
• The powers/METPO:2007900 story checks out: mapping row gates subject_types=BIOLOGICAL_PROCESS|STATE, object_types=GENE_OR_PROTEIN; 11 powers edges exist, 4 blocked, and phototrophic's CAPACITY-subject edge is byte-identical to carboxydotrophic's STATE-subject one modulo typing.

Two suggestions, neither blocking:

1. Section 11 points at reports/predicate_grounding_residual.tsv for the carboxydotrophic/phototrophic pair, but that report aggregates per predicate label — its example_files column lists aerobic/anoxygenic_photosynthesis/photosynthesis and no node_id at all, so a reader following the pointer will not find the pair. Worth naming the two files directly instead.

2. The header bullet credits proposals v8 (#320) and v9 (#326), but METPO:2007900 — the term section 11 leans on — was proposed in metpo_traitmech_v10 via #355. Adding v10 to that list would make the header and section 11 agree.

3. While measuring: the same node_id also disagrees on label. ph_phenotype_with_numerical_limits.yaml:52 calls proton_motive_force 'proton motive force across cell envelope' where 34 other records say 'proton motive force', and membrane_potential carries three distinct labels across its four records. That axis is invisible to the INCONSISTENT_NODE_TYPE check as step 1 describes it; cheap to fold in while the cross-record index is being built.