Add CMMMech and kg-microbe vocabulary comparison

CLAUDE.md

@@ -57,7 +57,7 @@ merge only with the maintainer's go-ahead. Changes merged to `main` deploy.
 - `_config.yml` - site configuration, `header_pages` navigation order, plugins, `exclude:`
 - `Gemfile` - Ruby dependencies
 - `_layouts/default.html`, `_layouts/home.html`, `_layouts/page.html`, `_includes/head.html`, `_includes/footer.html` - Minima overrides; the site header comes from the Minima theme itself
-- `assets/custom.css` - the site palette and styles (`--ink`, `--muted`, `--accent`, `--card` and the rest), which the fleet page also uses; the twelve Mech identity colours (`--mech-<name>` and their `-ink` variants) are defined in `_fleet/fleet_fragment.html`
+- `assets/custom.css` - the site palette and styles (`--ink`, `--muted`, `--accent`, `--card` and the rest), which the fleet page also uses; the thirteen Mech identity colours (`--mech-<name>` and their `-ink` variants) are defined in `_fleet/fleet_fragment.html`
 
 ### Other
 - `.github/workflows/fleet-page.yml` - the only CI; runs the tests and the fleet page checks, and nightly the checks against CLAW's live manifest and the Mech sites
@@ -76,9 +76,9 @@ Content pages use Jekyll front matter with `layout`, `title`, `description` and
 `resources.md` to follow the data flow from recipes to ingredients to communities.
 
 ### The X-Mech Suite page
-The suite has twelve Mechs, all under https://github.com/CultureBotAI, plus the
+The suite has thirteen Mechs, all under https://github.com/CultureBotAI, plus the
 orchestrator `culturebotai-claw`. Three have pages here; the others link to their
-own GitHub Pages sites. DUFMech's card links its family browser and seed
+own GitHub Pages sites, except CMMMech, which links its repository records. DUFMech's card links its family browser and seed
 snapshot, the source of its count. Its native FamilyRecord corpus now has a
 LinkML schema, history and individual pages; the census still measures the
 frozen seed worklist. CLAW membership is read from `_fleet/data/manifest.json`, independently of
@@ -104,17 +104,21 @@ Cells without a published record route link the pinned source file, including
 CommunityMech isolates. Complete cell lists remain separate from graph
 `indexed_voc` and do not change overlap methodology. Rows sort by vocabulary
 count, descending, with ties in biological scale order; DOI and PMID remain
-first, followed by columns in descending Mech coverage.
+first, followed by columns in descending dataset coverage. The heatmap also
+compares kg-microbe's pinned 2025-03-07 core release node table, always last
+with a small gap above it, counted with
+`build_kg_microbe.py`. Its source manifest, census and cell assets are separate
+from Mech membership, scientific-record totals and overlap indexes.
 
 Sources, refreshed differently:
 - **Fleet membership and capabilities:** the membership badges and capability
   table come from a snapshot of culturebotai-claw's fleet manifest,
   `_fleet/data/manifest.json`, refreshed with
   `scripts/fleet/refresh_manifest.py --claw-root <claw checkout>`, separately
   from the census. Its tooltips quote claw's reasons verbatim. The site lists
-  twelve Mechs; the membership count and undeclared-project prose are derived
+  thirteen Mechs; the membership count and undeclared-project prose are derived
   from the snapshot, not maintained as a separate count.
-- **All twelve corpora:** PathwayMech records and DUFMech seed families use the
+- **All thirteen corpora:** PathwayMech records and DUFMech seed families use the
   common census, subset, statistics and site-audit pipeline. DUFMech's frozen
   JSON worklist is validated and projected into individual family documents;
   its numeric card counts seed families, not curated mechanism records.
@@ -123,7 +127,7 @@ Sources, refreshed differently:
   vocabulary and merged-PR tiles, regenerated by the pipeline: record counts and
   vocabularies over snapshots of each Mech pinned at its GitHub `main`, merged-PR
   counts from GitHub search when `mech_stats.py` runs. The headline PR total
-  includes all twelve Mechs plus CLAW and this website; the two supporting
+  includes all thirteen Mechs plus CLAW and this website; the two supporting
   repositories are stored separately from the Mech record statistics. The census,
   the subsets, the cell lists, the stats and `_fleet/data/site_audit.json` record
   the revisions they read, and
@@ -147,7 +151,9 @@ Sources, refreshed differently:
   date and other sites unchanged. Schema-documentation prose reflects the
   website verdict, never an inference about whether a repository has a schema.
 
-Last full data refresh was 2026-10-07, pinned at 19:36 UTC. All twelve Mechs
+The original twelve source revisions remain from the 2026-10-07 refresh, pinned
+at 19:36 UTC. CMMMech was added at its October 8 revision, with a separate
+per-repository pin time. All thirteen Mechs
 participate in the census, overlap assets and repository statistics, including
 PathwayMech's 156 records and DUFMech's 6,532 seed families. TraitMech has 1,059
 records and CellStructureMech 906 at those pins. TaxonMech joined the census in
@@ -161,7 +167,7 @@ neither refresh is permission to redate or recount the other inputs.
 The suite licensing policy is CC BY 4.0 for data and BSD-3-Clause for code.
 Redistributed source material retains its applicable terms.
 
-### The twelve Mechs
+### The thirteen Mechs
 - **HabitatMech** (https://culturebotai.github.io/HabitatMech/) - habitats harmonized from GOLD, BacDive, PREGO and Madin et al. into ontology-grounded records (ENVO, UBERON, FOODON, BTO, PO).
 - **CommunityMech** (`/communitymech/`) - microbial communities, their interactions and cultivation conditions, in LinkML.
 - **TaxonMech** (https://culturebotai.github.io/TaxonMech/) - taxa and strains at species level and below, resolving NCBI Taxonomy, GTDB, LPSN and BacDive onto one NCBI-grounded identity.
@@ -172,11 +178,12 @@ Redistributed source material retains its applicable terms.
 - **DUFMech** (https://culturebotai.github.io/DUFMech/, repository https://github.com/CultureBotAI/DUFMech) - domains and protein families of unknown function, with evidence gathering and triage worklists.
 - **NaturalProductMech** (https://culturebotai.github.io/NaturalProductMech/) - one record per natural product structure with producer, gene cluster and bioactivity; grounded in ChEBI, MIBiG and NCBI Taxonomy. NPAtlas is a cross-reference target only: its CC BY-NC licence (from release 2024_09) bars ingestion into a CC BY 4.0 corpus.
 - **AntibioticMech** (https://culturebotai.github.io/AntibioticMech/) - one record per antimicrobial structure, harmonizing ChEBI and CARD/ARO; record content is CC BY 4.0.
+- **CMMMech** (https://github.com/CultureBotAI/CMMMech) - critical-material microbial extraction, transformation and recovery, with three element-level pilot records, ChEBI identities, taxon references and evidence. No published website; card and record links point to the repository.
 - **MediaIngredientMech** (`/mediaingredientmech/`) - LLM-assisted curation of media-ingredient mappings; its records map mostly to ChEBI, then MeSH, NCIT, MicrO, FOODON and ENVO. Where curation found no ontology term, about 175 mapped ingredients carry kg-microbe registry ids, mostly for preparations and mixtures, and about 85 carry CAS Registry Numbers; about 55 more carry kg-microbe placeholder ids pending curation. None map to PubChem or METPO (#150, #186, #215).
 - **CultureMech** (`/culturemech/`) - versioned, ontology-grounded culture-media recipes from MediaDive, TogoMedium, KOMODO and the major collections.
 
 Wherever the suite is enumerated (index.md, resources.md, each page's Related
-Tools), list all twelve and link to `/mechs/`. Name ontologies consistently across
+Tools), list all thirteen and link to `/mechs/`. Name ontologies consistently across
 pages (ChEBI, NCBITaxon, METPO, ENVO and so on), and describe a Mech's groundings
 by what its records actually use, checked at its pinned revision.
 

_fleet/README.md

@@ -42,13 +42,14 @@
 
 ## Membership and capability updates
 
-The site lists twelve Mechs. Membership counts, badges and prose about
+The site lists thirteen Mechs. Membership counts, badges and prose about
 undeclared projects come from `data/manifest.json`; suite coverage does not
 imply CLAW admission. A project's capability row remains undeclared until the
-snapshot contains its declaration. Tests exercise both an eleven-member
-fixture without DUFMech and a twelve-member fixture with synthetic DUFMech
-declarations, independently of the current snapshot. The October 7 vocabulary
-census, shared-term assets and repository statistics cover all twelve Mechs.
+snapshot contains its declaration. Tests exercise both a twelve-member
+fixture without DUFMech and a thirteen-member fixture with synthetic DUFMech
+declarations, independently of the current snapshot. The October 8 addition
+extends the census, shared-term assets and statistics to CMMMech, retaining
+the original twelve Mech source revisions from October 7.
 DUFMech's contribution to that census is a measured seed worklist, labeled as
 families rather than curated mechanism records. Its schema-documentation prose
 uses the website feature verdict, not membership or an inferred schema absence.
@@ -443,3 +444,64 @@ The DUF nightly card check resolves the latest dated worklist on `main` through
 GitHub directory metadata. Its pinned audit still records the frozen source
 used for this page, so a new worklist cannot leave the live freshness check
 reading the old snapshot indefinitely.
+
+
+## CMMMech and the kg-microbe comparison (October 8, 2026)
+
+CMMMech joins every Mech surface: ladder, card, graph node, vocabulary census,
+shared-term edges, statistics and both support tables. The canonical corpus is
+`data/records/**/*.yaml` and `*.yml`, matching its native CLI and CLAW manifest;
+fixtures, reviews and history sidecars are not additional records. Its three
+records at `905388523b6e9184783989484896e4cc20297832` have no REVIEWED-bearing
+status slot. A null statistic makes no claim about scientific review artifacts.
+There is no published website, so links use the pinned GitHub records and the
+website-feature row explicitly says so. Its card checker counts a complete Git
+tree inventory, rejects truncated API responses, and compares canonical paths
+and blob hashes with `git ls-tree` at the pin. It never treats a README's
+hand-maintained list as the live corpus size.
+
+The original twelve Mech source revisions are retained. All thirteen corpora
+were rescanned for this addition: CELEX and registry aliases observed in the
+kg-microbe node table apply consistently to every row. CMMMech carries its own
+`pinned_at_utc` in the audit and snapshot manifest; the original twelve keep
+their October 7 pin time and nightly growth allowance. Merged-PR activity was
+queried again. The existing website feature audits retain their own dates.
+
+kg-microbe is a **heatmap comparison**, not a fourteenth Mech. It never joins
+`roots.ORDER`, the Mech statistics, cards, capability table or shared-term edge
+index. `_fleet/kg_microbe_source.json` pins the published **2025-03-07 core**
+release asset by SHA-256, not current main or a locally rebuilt graph. Reproduce
+its counts with:
+
+```bash
+python3 scripts/fleet/build_kg_microbe.py /path/to/kg-microbe-core.tar.gz
+python3 scripts/fleet/build_data.py
+python3 scripts/fleet/assemble_page.py
+```
+
+The script verifies the archive hash and reads the nested `merged-kg.tar.gz`
+without extracting archive paths. It records the node-table hash and treats
+one KGX node as one record. Counts scan all node fields (including xrefs and
+prose) with the same prefix regex and normalization registry as Mech records;
+repeat identifiers are repeat occurrences. Matching-node totals count each
+node once per vocabulary. Edge rows, graph-local identifiers, metamodel terms
+and namespaces outside the suite registry are excluded. This is coverage of
+the tracked namespaces, not an exhaustive inventory of graph namespaces or a
+count of distinct ontology terms. The release has 1,379,336 node records;
+headers and blank lines are not nodes.
+
+`data/kg_microbe_census.json` records the counts and source, independently of
+Mech commit pins. `assets/fleet/comparison-cells/` contains up to 300 node links
+per populated cell with total counts and hashes, following the same cell format
+as Mech records. Links open external term pages where available, otherwise the
+source release with the node identifier shown in the label. Mech cell rebuilds
+own only `assets/fleet/cells/` and cannot delete comparison assets.
+
+`build_data.py` merges the comparison into `heat`, `voc` and `cells`, keeping
+`order`, `indexed_voc` and `vocab_edges` exclusively Mech-derived. The embedded
+`comparisons` metadata supplies the row identity and source note. Columns sort
+by dataset coverage after DOI/PMID; Mech rows sort by tracked vocabulary count, with
+biological-order ties. kg-microbe is always the final row, with a small gap
+above it. Tests check
+source hashes, complete cell coverage, matching counts and isolation from
+fleet membership and graph indexes.