Skip to content

Add Druantia III system trait - #1584

Merged
realmarcin merged 6 commits into
mainfrom
add-druantia-iii-system
Oct 3, 2026
Merged

realmarcin merged 6 commits into
mainfrom
add-druantia-iii-system

Conversation

@realmarcin

@realmarcin realmarcin commented Oct 3, 2026 •

Copy link
Copy Markdown
Contributor

Summary

  • add traitmech:000560 Druantia III system as a GENOMICS PROPOSED child of traitmech:000234 Druantia system
  • support identity with DOI 10.64898/2026.05.12.724681, pinned DefenseFinder DefenseFinder_rules.tsv, Liste_hmm_system.md, List_system_article.md rows at commit afb0e5a8b466be53586b13266f5d38d98c3ac268, and a PMC-backed Escherichia coli ATCC 8739 canonical example
  • add a conservative NONMECHANISTIC Druantia III causal graph and keep only the unresolved native-host breadth, subtype trigger, and profile-to-activity questions as knowledge gaps
  • reserve METPO:1051400 in proposals/metpo_traitmech_v437 for upstream METPO adoption
  • update the broad Druantia subtype gap to mark Type III as split out, ground the remaining Type IV scope to pinned HMM rows, then regenerate discussions data, static pages, and coverage/audit reports

Validation

  • .venv/bin/python scripts/add_druantia_iii_system_trait.py
  • .venv/bin/python scripts/add_druantia_iii_system_trait.py --apply
  • .venv/bin/ruff check scripts/add_druantia_iii_system_trait.py
  • .venv/bin/linkml-validate --schema src/traitmech/schema/traitmech.yaml --target-class TraitRecord data/traits/genomics/druantia_iii_system.yaml data/traits/genomics/druantia_system.yaml
  • .venv/bin/python scripts/validate_strict.py data/traits/genomics/druantia_iii_system.yaml data/traits/genomics/druantia_system.yaml
  • .venv/bin/linkml-validate --schema src/traitmech/schema/history.yaml --target-class HistoryRecord history/records/druantia_iii_system/2026-10-03T012414Z-claude-code-47813a.yaml history/records/druantia_system/2026-10-03T012415Z-claude-code-31aa76.yaml
  • .venv/bin/python scripts/validate_history_links.py history/records/druantia_iii_system history/records/druantia_system
  • .venv/bin/python scripts/verify_snippets.py --record data/traits/genomics/druantia_iii_system.yaml
  • .venv/bin/python scripts/verify_snippets.py --record data/traits/genomics/druantia_system.yaml
  • .venv/bin/python -m kg_microbe_discussions --config conf/discussions_config.yaml --output app/discussions
  • .venv/bin/python scripts/audit_discussions_data.py
  • .venv/bin/python scripts/audit_discussion_anchors.py
  • .venv/bin/python scripts/render_trait_pages.py
  • .venv/bin/python scripts/trait_priority.py --dashboard --top 80
  • .venv/bin/python scripts/audit_proposals.py
  • .venv/bin/python scripts/verify_metpo_proposal.py proposals/metpo_traitmech_v437
  • .venv/bin/python scripts/robot_validate_proposal.py proposals/metpo_traitmech_v437
  • .venv/bin/python scripts/audit_evidence_snippets.py
  • .venv/bin/python scripts/audit_canonical_examples.py --no-resolve
  • .venv/bin/python scripts/audit_canonical_examples.py --ncbi-api
  • .venv/bin/python scripts/audit_causal_graphs.py
  • .venv/bin/python scripts/ground_causal_predicates.py
  • .venv/bin/python scripts/ground_causal_nodes.py
  • .venv/bin/python scripts/audit_predicate_domains.py --fail-on new
  • .venv/bin/python scripts/audit_graph_protein_taxa.py --fail-on gaps
  • .venv/bin/python scripts/check_biolink_coverage.py
  • .venv/bin/python scripts/audit_biolink_curies.py
  • .venv/bin/python scripts/audit_unapplied_groundings.py
  • .venv/bin/python scripts/audit_writers.py
  • .venv/bin/python scripts/audit_exact_synonyms.py --collisions-only
  • .venv/bin/python scripts/check_sources.py
  • .venv/bin/python scripts/audit_schema.py
  • .venv/bin/python scripts/audit_justfile_paths.py
  • .venv/bin/python scripts/audit_qc_paths_coverage.py
  • .venv/bin/python scripts/run_trait_graph_audit.py --verify
  • .venv/bin/python scripts/pr_sanity.py
  • .venv/bin/python -m pytest tests/test_readme_artifacts.py tests/test_trait_priority.py -v --tb=short
  • .venv/bin/python scripts/audit_history_records.py --base origin/main
  • git diff --check
  • git diff --cached --check

Review

Local adversarial review tried to falsify trait identity, sequence-feature interpretation, duplicate search, parent choice, xrefs, snippets, canonical examples, the METPO proposal, and regenerated artifacts. That review and PR review rounds found stale generated artifacts, a stale writer rationale constant, an uncited Druantia Type IV statement, and a contradictory canonical-example gap; all four are fixed in this branch.

The pre-curation duplicate search used rg --no-ignore --hidden across live traits, METPO, history, proposals, research, reports, and generated pages. No exact Druantia III child record, traitmech:000560, METPO:1051400, or v437 proposal existed before this branch.

@culturebot-reviewer culturebot-reviewer Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Identity, grounding, and graph structure check out. traitmech:000560 is the next free local ID, METPO:1051400 continues the v436 block cleanly, METPO:1018800 really is the v111 placeholder for traitmech:000234, and the three predicates are domain-legal in mappings/predicate_grounding.tsv (RO:0002326 *→*; METPO:2007700 admits BIOLOGICAL_PROCESS→TRAIT; rdfs:subClassOf *→*). All four nodes are referenced by edges, the TRAIT node is present, nothing is orphaned. Both history records are present, app/discussions/data.js was regenerated, and the DefenseFinder snippet shapes (fixed-width HMM rows, truncated rule-row prefixes, backslash-escaped DOIs in the article registry) match the already-merged records that quote the same pinned files. Network access was blocked in my environment, so verbatim-ness against the pinned DefenseFinder files and the Wu et al. preprint was checked only for that internal consistency, not against the sources themselves.

Two things need fixing.

🔴 README.md corpus counts were not updated, and this fails the test suite.
tests/test_readme_artifacts.py::test_readme_corpus_table_matches_the_trait_artifacts compares the table against the live corpus. With this record added, genomics is REVIEWED 19 / PROPOSED 348 / causal_graphs 367 / total 367, but README.md:23 still says | GENOMICS | 19 | 347 | 0 | 366 | 366 | and README.md:29 still says | **TOTAL** | **427** | **439** | **50** | **710** | **954** | against a live 955-record corpus. The prose at README.md:33 ("347 genomics records") and the layout comment at README.md:95 ("954 curated TraitRecords") are stale for the same reason; .claude/skills/add-trait/SKILL.md:417 asks for every record-count mention, derived live. Each of the five preceding trait PRs (#1573, #1574, #1575, #1577, #1580) updated this file.

🟡 app/dashboard/priority.{json,html} was not regenerated.
app/dashboard/priority.json still reports "records": 954 and its actions tallies do not include traitmech:000560. Nothing in just qc gates this file, so it drifts silently; .claude/skills/add-trait/SKILL.md:465 (trait_priority.py --dashboard --top 80) is the maintained regeneration step, and #1580 ran it.

🔵 The revised parent rationale asserts "the pinned Type IV HMM rows" and the new record scope_notes excludes "Type IV partner functions", but no Druantia_IV row is quoted in either record or anywhere else in the corpus, making it the one source claim in this diff with no snippet behind it. Also minor: both history records set actors[].name: claude-code with model: gpt-5-codex / agent_tool: codex, while curation_history.curator is codex and the sibling record used name: codex.

@culturebot-reviewer culturebot-reviewer Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The trait record itself holds up. traitmech:000560 / METPO:1051400 / v437 all follow the v436 cohort cleanly, METPO:1018800 really is the v111 placeholder for traitmech:000234, the three predicates resolve against mappings/predicate_grounding.tsv with correct domains (contributes to = RO:0002326 */*; confers = METPO:2007700 with BIOLOGICAL_PROCESS subject / TRAIT object; is a = rdfs:subClassOf), all four graph nodes are referenced by edges, the discussion anchor matches the graph_id, both history records are present, and the README counts I re-derived (955 total / 348 genomics PROPOSED / 711 with causal graphs) match the tree. The writer script honors the dry-run-plus---apply contract and goes through write_validated_trait / record_curation_event.

Two things need fixing, both from the same root cause: b05d7f9 ("Fix Druantia III review findings") hand-reworded the parent's discussion rationale and regenerated README and app/dashboard/*, but not the discussions data, the rendered pages, or the script that wrote it.

🔴 Stale generated artifacts — qc is red because of this.

data/traits/genomics/druantia_system.yaml:185 now reads:

... until separate review resolves Type I, Type II, and Type IV partner functions, exact phage triggers ...

but app/discussions/data.js:2321 and pages/traits/genomics/druantia_system.html:294 both still carry the pre-fix wording:

... until separate review resolves Type I/II partner functions, the pinned Type IV HMM rows, exact phage triggers ...

scripts/audit_discussions_data.py compares rationale directly and audit-discussions-data is part of just qc; the qc check on this PR fails. Re-run just gen-discussions-data and just gen-pages rather than patching the two files by hand.

🟡 scripts/add_druantia_iii_system_trait.py:359-370 disagrees with the data it produced.

UPDATED_DRUANTIA_DISCUSSION_RATIONALE is still the pre-fix string. build_parent() accepts the parent rationale only if it equals OLD_... or UPDATED_..., so against the current tree it now raises SystemExit("unexpected Druantia subtype discussion rationale") — the script can no longer reproduce or re-apply its own output. Update the constant to match the YAML.

One limitation to be explicit about: this environment has no network, so I could not confirm the Wu et al. and DefenseFinder snippets verbatim against their sources. Indirect support is good — the two DOI snippets are byte-identical to snippets already merged on druantia_system.yaml, the Druantia\tDruantia_III\t2\t2\tDruantia_III__DruH, Druantia__DruE_1\t\t\t row matches the column shape of the other 309 records pinned to the same commit, and the Druantia__DruE_1 shared-namespace profile as a RELATED_SYNONYM has direct precedent in disarm1/disarm2 sharing DISARM__drmA — but that is corroboration, not verification.

@culturebot-reviewer culturebot-reviewer Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The record holds up. traitmech:000560 is the next free local ID, METPO:1051400 continues the v436 block, METPO:1018800 really is the v111 placeholder for traitmech:000234, and all three predicates are domain-legal in mappings/predicate_grounding.tsv (RO:0002326 */*; METPO:2007700 confers admits BIOLOGICAL_PROCESS→TRAIT, which is what the edge uses; rdfs:subClassOf */*). All four graph nodes are referenced by edges and reachable from the TRAIT node, the discussion anchor matches the graph_id, both history records are present, and the artifacts flagged on the earlier rounds — README counts, app/dashboard/priority.*, app/discussions/data.js, pages/traits/genomics/druantia_system.html — are now regenerated. Every check is green.

🟡 scripts/add_druantia_iii_system_trait.py:359-370 no longer matches the data it produced.

b05d7f9 reworded the parent rationale in data/traits/genomics/druantia_system.yaml:185 to

... until separate review resolves Type I, Type II, and Type IV partner functions, exact phage triggers ...

but UPDATED_DRUANTIA_DISCUSSION_RATIONALE still holds the pre-fix string (Type I/II partner functions, the pinned Type IV HMM rows, ...). build_parent() accepts the on-disk rationale only when it equals OLD_... or UPDATED_..., so against the current tree it now hits raise SystemExit("unexpected Druantia subtype discussion rationale"): both the dry run and --apply listed in the PR description fail today, and the writer can no longer reproduce or re-apply its own output. Nothing in just qc re-runs writer scripts, which is why CI is green on it. Update the constant to the committed wording.

(I verified this at eeb2464, the newest commit I could fetch into my checkout; 0a53544 is titled "Regenerate Druantia parent page", and just gen-pages would not touch scripts/. If that commit did update the constant, disregard.)

🔵 Druantia "Type IV" still appears in the new scope_notes, the reworded parent rationale, and the v437 proposal's exclusion list, and it remains the one source claim in this diff with no snippet behind it — no Druantia_IV row is quoted here or anywhere in the corpus, and the parent's own Doron et al. evidence enumerates Types I–III. Quoting the pinned HMM or rules row would settle it; otherwise it is an assertion being carried upstream into a METPO proposal.

Limitation: my environment had no network, so snippet verbatim-ness against the Wu et al. preprint and the pinned DefenseFinder files at afb0e5a was not checked against the sources. Internal corroboration is good — the two DOI snippets are byte-identical to snippets already merged on druantia_system.yaml, and the rule-row, HMM-row, and backslash-escaped article-row shapes match the other records pinned to the same commit — but that is corroboration, not verification.

@culturebot-reviewer culturebot-reviewer Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Grounding and graph structure check out: traitmech:000560 is unused elsewhere, traitmech:000234 really is Druantia system, METPO:1051400 is free and METPO:1018800 is v111's reserved Druantia placeholder, and RO:0002326 / METPO:2007700 / rdfs:subClassOf are used with the same subject/object types as the sibling defense-system records. The graph is 4 nodes / 3 edges, every edge references a declared node, one component, no orphans. History records exist for both touched records, and README, pages/, app/discussions/data.js, priority.json, and the three reports are all consistent with the final YAML (955 / 367 matches git ls-files). Two things need fixing.

🟡 The writer script no longer reproduces the shipped parent record, and now hard-fails.

scripts/add_druantia_iii_system_trait.py:359-370 still holds the pre-fixup text — "resolves Type I/II partner functions, the pinned Type IV HMM rows, exact phage triggers" — but data/traits/genomics/druantia_system.yaml:185-188 was hand-edited in b05d7f9 to "resolves Type I, Type II, and Type IV partner functions, exact phage triggers". build_parent() compares against both OLD_ and UPDATED_ and raises on neither match, and main() calls it unconditionally, so scripts/add_druantia_iii_system_trait.py now exits with unexpected Druantia subtype discussion rationale on both the dry run and --apply — i.e. the two commands in the PR description's validation list no longer pass. This is the divergence the playbook's "A migration's curation_history must survive being re-run" warns about. Point UPDATED_DRUANTIA_DISCUSSION_RATIONALE at the text that actually shipped.

While you're in there: that hand-edit replaced "the pinned Type IV HMM rows" with "Type IV partner functions", which removed the only pointer to where Druantia Type IV comes from. Nothing else in the diff or the corpus cites a Druantia Type IV, and the parent's own Doron et al. snippet enumerates DruABCDE (type I) DruMFGE (type II) DruHE (III) only. Restore the HMM-row attribution or cite the Druantia_IV row.

🟡 The canonical-example evidence-gap event is contradicted by evidence already in this repo.

data/traits/genomics/druantia_iii_system.yaml:199-207 records that curation found no "accession-backed native microbial taxon exemplar tied to the Druantia_III profiles and experimentally verified endogenous activity." But data/traits/genomics/armada_system.yaml:66-72 already carries NCBITaxon:562 / Escherichia coli (strain ATCC 8739) referenced to https://pmc.ncbi.nlm.nih.gov/articles/PMC12458937/, whose snippet at line 51 reads "the strain carrying only Druantia III exhibited at least modest protection (<0.01 EOP or <0.5 SFC) against 3 phages" — native, deletion-resolved Druantia III activity in a named strain, which is exactly the bar the event says is unmet. The playbook's canonical-example section (#475) explicitly permits an example found via a sibling record's artifact when it carries its own citation, as this one does.

PMC12458937 is also absent from the new record entirely, even though the open druantia-iii-activity-gap lists "native host breadth" and "Zorya coupling" as unresolved and that source speaks to both (it reports Druantia III / ARMADA Type II synergy in the native host). Either add the canonical example and cite the source, or narrow the event text to the claim you can defend and say why the ARMADA evidence does not qualify.

Verification limit, stated plainly: this environment had no network access, so I could not check the Wu et al. snippets or the pinned DefenseFinder rows byte-for-byte against their sources. The DOI is already load-bearing in merged druantia_system.yaml, and the rules-table and HMM-table snippet shapes — including the trailing \t\t\t and the escaped-dot DOI in the article registry — match drt7_system.yaml:50 and armada_system.yaml:60 exactly, so nothing looks reshaped. The 2/2 reading of the rule row is also consistent with DRT7's 1/1 row being read as "one of two profiles". Still unverified rather than verified.

@github-actions

github-actions Bot commented Oct 3, 2026

Copy link
Copy Markdown

PR shepherd: picked this over the other four open PRs because it is the only one that is fully green and mergeStateStatus: CLEAN, yet still sits behind an open CHANGES_REQUESTED round — #1476/#1477 are dependabot bumps and #973/#924 have no review at all, so they're idle rather than deadlocked.

State. Head is 0a53544. All eight checks pass on it: qc, pytest, validate-strict, history, pr-sanity, canonical-example-taxonomy, label-correspondence, vendored-sync. Three CHANGES_REQUESTED reviews, all from culturebot-reviewer (01:45, 02:04, 02:10); reviewDecision is empty, so nothing formally blocks the merge button. No human has reviewed.

Worth flagging for whoever reads the thread top-down: the 🔴 in rounds 1 and 2 ("qc is red because of this") was already stale when round 2 was submitted — pytest was still IN_PROGRESS and qc for that same commit finished passing 76 seconds later. Round 3 confirms the README counts, app/dashboard/priority.*, app/discussions/data.js, and pages/traits/genomics/druantia_system.html are all regenerated. Reading the rounds in order makes this look worse than it is.

One real defect survives, and CI structurally cannot catch it. Per round 3, b05d7f9 reworded the parent rationale in data/traits/genomics/druantia_system.yaml:185 to "Type I, Type II, and Type IV partner functions", but UPDATED_DRUANTIA_DISCUSSION_RATIONALE in scripts/add_druantia_iii_system_trait.py:359-370 still holds the pre-fix wording. build_parent() accepts the on-disk rationale only if it equals OLD_... or UPDATED_..., so both the dry run and --apply in the PR description now hit SystemExit("unexpected Druantia subtype discussion rationale"). The two commits after b05d7f9 are regenerations (eeb2463 discussion data, 0a53544 parent page) and neither touches scripts/. Nothing in the qc target list (justfile:1128) re-runs writer scripts, which is why this is green.

Most useful next action: one human updates that one constant to the committed wording, then merges. That is a one-line change; everything else the three rounds raised is either resolved or advisory (🔵: Druantia "Type IV" appears in scope_notes, the parent rationale, and the v437 proposal exclusion list with no Druantia_IV snippet quoted anywhere in the corpus — worth settling before it rides upstream into METPO, but not a merge blocker).

Caveat on my own evidence: my environment could not fetch the PR branch, so the check states, commit list, and review contents above are verified live, but the scripts/ line-level claim is the round-3 reviewer's quote rather than something I read myself.

@culturebot-reviewer culturebot-reviewer Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Identity, ID hygiene, and graph structure check out. traitmech:000560 is the next free local ID, METPO:1051400 is unused and v437 follows v436, the parent placeholder METPO:1018800 really is the v111 row for traitmech:000234, both predicates are declared in mappings/predicate_grounding.tsv with domains that admit these edges (RO:0002326 *→*, METPO:2007700 BIOLOGICAL_PROCESS→TRAIT), all four nodes are referenced by edges and the trait node is reachable, the writer is dry-run-by-default and idempotent through write_validated_trait/record_curation_event, and both history records match the card_nlr_like_system precedent. The DefenseFinder row quotes are self-consistent against the v436 header reading (System/Subsystem/Min_mandatory/Minimum_genes/Mandatory + three empty trailing fields), and the HMM rows carry the same column padding as every other DefenseFinder-backed record, so the interpretation Min_mandatory=2 over a two-profile mandatory set is right.

One thing needs fixing.

🟡 The notes: asserts a negative the quoted snippet does not support — data/traits/genomics/druantia_iii_system.yaml:43-45 and again at :194-196 (and scripts/add_druantia_iii_system_trait.py:118): "a deletion-resolved strain retaining Druantia III but lacking the co-located ARMADA Type II and Zorya II systems." The snippet above it says only "the double-deletion strain ... the strain carrying only Druantia III," and nothing in it mentions Zorya. The sibling armada_system.yaml, curating the same experiment from the same source, reads those two deletions as ARMADA Type II and Druantia Type III — on that reading the Druantia-III-retaining derivative is a single ΔARMADA strain that still carries Zorya II, and the claim is simply wrong. This is the #621 defect class the playbook calls out by name ("a notes: line asserting a negative the source did not support"). Either drop the Zorya II clause or add the sentence from Bell et al. that states the Zorya II deletion. Note zorya_type_ii_system.yaml confirms ATCC 8739 carries Zorya II, which makes the unsupported "lacking" wording worse, not safer. The record's separate framing of Zorya II coupling as an open gap (scope_notes, the discussion) is fine and should stay.

🔵 Both new history files are named ...-claude-code-<shortid>.yaml but declare session.actors[0].name: codex. history/README.md specifies <TIMESTAMP>-<actor>-<shortid>, and all 66 other claude-code-named records in history/records/ declare name: claude-code — these two are the only mismatches in the corpus. One of the filename or the actor block misattributes the session.

Scope note on my verification: I could not reach the network from this environment, so I did not re-read Wu et al., the Bell et al. preprint, or the pinned DefenseFinder files. The Wu and Bell snippets are byte-identical to ones already on main in druantia_system.yaml and armada_system.yaml; the DefenseFinder rule/HMM/article rows are new in this PR and were checked only for internal and corpus-wide format consistency, not against the pinned files.

@culturebot-reviewer culturebot-reviewer Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is in good shape. The items from the earlier rounds are all genuinely fixed, not papered over, and I re-derived the load-bearing claims rather than trusting the PR description.

Identity and ID hygiene. traitmech:000560 is the next free local ID (highest on main is 000559) and appears nowhere else in the corpus. traitmech:000234 really is Druantia system. METPO:1051400 is unused and follows METPO:1051300 (v436) on the hundred-block convention; METPO:1018800 really is the v111 reserved placeholder for traitmech:000234, so the proposal's parent row is right. The definition follows the established DefenseFinder rule-row phrasing used by drt7_system.yaml et al.

Grounding. All three predicates resolve in mappings/predicate_grounding.tsv with domains that admit these edges: RO:0002326 contributes to (*→*), METPO:2007700 confers (BIOLOGICAL_PROCESS is in the subject list, TRAIT is the object), rdfs:subClassOf is a. The two new ungrounded node labels land in node_grounding_residual.tsv as expected rather than being force-fitted to a near-miss term.

Graph structure. 4 nodes / 3 edges, every edge subject and object is a declared node_id, one connected component, and the TRAIT node druantia_iii_system_trait reaches all three others — nothing for audit-graphs to ratchet on.

Previously-flagged items, re-checked. README counts are now correct against the live tree (955 total, 367 genomics, 348 genomics PROPOSED, 19 REVIEWED, 711 with causal_graphs, 427/440/50/38 by status — all derived from the working tree, not the diff). UPDATED_DRUANTIA_DISCUSSION_RATIONALE no longer blocks the writer: build_parent() now accepts TYPE_IV_GROUNDED_..., which is what shipped in druantia_system.yaml:185, so the script reproduces its own output. The unsupported "and Zorya II" negative is gone from the Bell note. The canonical example is present and its note is exactly what the quoted snippet supports ("the strain carrying only Druantia III exhibited at least modest protection … against 3 phages"). Druantia Type IV now has the three pinned Druantia_IV__* HMM rows behind it on the parent. Generated artifacts match the final YAML — pages/traits/genomics/druantia_system.html and app/discussions/data.js both carry the current "Type IV partner functions represented by the pinned Druantia_IV HMM rows" wording, and just qc is green.

Writer and provenance. Dry-run by default with an --apply gate, idempotent, routed through write_validated_trait and record_curation_event, with an identifier guard before overwriting. Both history/ records are present, and name: claude-code + agent_tool: codex has 45 precedents in history/records/.

🔵 One optional thing, only if it is cheap to check against the pinned file: the List_system_article.md row quoted at data/traits/genomics/druantia_iii_system.yaml:47 ends with a trailing space after the final pipe (… microbial pangenome | ), whereas zorya_type_ii_system.yaml:32 quotes the sibling row from the same table with no trailing space. Rows in that file could legitimately differ, so this may be nothing.

Verification limit, stated plainly: this environment had no network and no Python execution, so I could not re-read Wu et al., the Bell et al. preprint, or the pinned DefenseFinder files, and I could not run the validators myself — I relied on the PR's green validate-strict, qc, history, label-correspondence, and canonical-example-taxonomy checks. Snippet verbatim-ness was therefore checked only indirectly: the two Wu et al. snippets and the Bell et al. 66-phage snippet are byte-identical to text already merged on druantia_system.yaml and armada_system.yaml, the HMM-row padding and the \t\t\t-terminated rule row match the corpus-wide shape for that pinned commit, and the three-row joined HMM snippet on the parent has precedent in 18 other records. That is corroboration, not verification; the new DefenseFinder rows have not been read against afb0e5a. pytest was still pending when I reviewed.

@realmarcin
realmarcin added this pull request to the merge queue Oct 3, 2026
Merged via the queue into main with commit 1a225a3 Oct 3, 2026
8 checks passed
@realmarcin
realmarcin deleted the add-druantia-iii-system branch October 3, 2026 02:54
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant