X-Mech Suite: fourteen autonomous knowledge factories, one shared standard

Baseline audit

Each Mech's live GitHub Pages site as checked on October 10, 2026 (US Pacific time), independently of the repository pins used by the census. AI auditors reassessed all 24 catalogue features across the twelve published sites using headless Chrome: identifier and label searches, filters and pagination, URL restoration, page-to-page theme persistence, 390 px layouts, and deliberately aborted data requests where applicable. Evidence was reviewed independently; changed ratings were independently re-tested, and columns were compared across sites against the same catalogue criteria. Notes identify sampled routes rather than claiming an exhaustive check of every record. Each deployed_revision identifies a deployment whose latest status was verified as successful for that site's final browser check; later deployments may differ. CMMMech and AIMech had no published website when checked; both began publishing on October 12, 2026, and their sites were judged later in site-only rechecks.

Features

Search

A labeled text search over the whole published corpus, tested with a known identifier and a known label. Present: it finds records by identifier and by label, shows how many match, and every match can be opened. Partial: results are capped with no route to the rest, no match count is shown, identifiers are not searchable, or search works only within one category.

Filters

Filter controls on record properties, with a way to clear them. Present: at least one filter on a record property in the main record listing, with a reset. Partial: filters only on a map or plot page, or a single selector without a reset. A free-text box is a search, not a filter.

Result counts

A count of matching records that updates with search and filters, distinct from the total. Partial: only a static total.

Sorting

The reader can change the sort order of a listing. Present: a sort control or sortable column headings.

Linkable search

Search and filter state is kept in the URL, so a filtered view can be bookmarked and restores on load. Partial: some state, such as a facet or a selected record, restores from the URL but the main search does not. Does not apply: the site has no search or filters.

Complete catalog

Every record the site counts is reachable from it, by listing, pagination, shards or lazy loading. Partial: some records the site counts cannot be reached from its pages.

Record pages

Each record has its own stable, linkable URL.

Term links

Ontology terms show a readable label beside the CURIE, and the CURIE links to a resolver. Present: the record's own grounding and the terms in its main sections are labeled and linked; a few minor prefixes may be plain text. Partial: a whole main section of terms is unlinked, unlabeled or broken, or the record's own identifier does not resolve for many records.

Evidence links

References (PMID, DOI or URL) shown on record pages are linked. Partial: references are plain text, or only a minority of the records that cite something link it.

Curation status

Each record shows its curation or review status.

Curation history

The record view shows its curation events: what changed, when and by whom.

Source link

The record view links to the record's own source file in the GitHub repository. Partial: the file is named as plain text, or only the repository root is linked.

Domain visual

A view suited to the domain, such as a structure or embedding map, a pathway or causal diagram, an interaction network or a lineage. Partial: the visual exists but is broken for readers.

Cross-Mech links

Record views link to specific records in sibling Mechs or in kg-microbe. Partial: related records are named as plain text, or the links are broken.

Corpus statistics

The landing page shows generated corpus counts.

Category browsing

Category or class pages lead into subsets of the corpus. Partial: categories are reachable only as a filter, or their index is not linked from the landing page.

Suite link

A visible link to the X-Mech Suite directory at culturebotai.github.io/mechs/.

Downloads

A bulk machine-readable export of all records is linked from the site. Partial: only the repository, per-record files, partial exports, or a bulk file that no page links. Missing: no data or repository link at all.

Schema docs

LinkML schema documentation is published on or linked from the site. Partial: the site says a schema exists but links only the repository root, or a documentation page exists that nothing links. Does not apply: there is no schema.

License

The site states the data and code licenses, consistent with the repository's license files. Partial: stated on only some pages, or inconsistent with the repository.

Build or version

The site states when it was built, which commit it shows, or a dated version of its own records. Partial: only the dates or releases of upstream sources.

Theme toggle

A light and dark toggle whose choice carries across pages. Partial: follows the operating system's setting without a toggle, or some linked pages ignore the choice.

Narrow screens

No horizontal page overflow at 390 pixels on pages linked from the main navigation, the browse page or record pages; wide tables scroll in their own region. Partial: some of those pages overflow.

Load errors

When a data request fails, the page shows an error rather than an empty or misleading result. Partial: some failures are silent or misleading. Does not apply: the pages fetch no data.

HabitatMech

Site-specific check on October 11, 2026, deployed revision 853fa61. All 24 catalogue features were reassessed in Chrome on the recorded successful deployment and independently checked against the same criteria. Checks covered 16 linked listing pages with 3,208 distinct record URLs, identifier and label searches in the Aquatic listing and semantic map, the map result cap, reset and URL restoration, two habitat records and a causal graph, the current map manifest, both OS color schemes, fresh and resized 390 px layouts, and a failed category-index request.

• Search partial: The semantic map searches all 3,208 records by identifier and label: ENVO:03000033 returns one match and sediment returns 138. It exposes only the first 20 links without paging; category search is limited to one category. Evidence
• Filters missing: Category pages and the semantic map offer free-text search, with Clear filter on category listings. They do not expose property-filter controls. Evidence
• Result counts present: Aquatic search updates to Showing 1–1 of 1 matching records (468 in category) for ENVO:03000033, and 89 matching records for sediment. The semantic map reports its matching count separately. Evidence
• Sorting missing: Category listings and the semantic map offer no sort control. Clicking the first Aquatic table heading leaves the first rows in the same order. Evidence
• Linkable search missing: Category and semantic-map searches leave their URLs unchanged. Reloading the map or opening either page with ?q=sediment restores an empty search. Evidence
• Complete catalog present: Following all eight category indexes and pagination discovers sixteen HTTP-200 listing pages with exactly 3,208 unique record links, matching the landing and text-map totals. Evidence
• Record pages present: Habitat records have stable static URLs. Both category and map searches link marine sediment at habitats/marine-sediment-envo-03000033.html, which opens successfully. Evidence
• Term links partial: Marine sediment has a labeled grounding CURIE with a resolver link and linked broader habitats. Its Associated taxa section still displays readable taxon names beside unlinked NCBITaxon identifiers. Evidence
• Evidence links partial: Marine sediment displays PMID:31388058 and DOI:10.1016/j.epsl.2022.117841 as plain text in mechanism evidence and curation notes; those citations have no PubMed or DOI anchors. Evidence
• Curation status present: Record headers show SEEDED or REVIEWED. The landing now reports 2,510 SEEDED and 698 REVIEWED records and explicitly says workflow status does not establish human review. Evidence
• Curation history present: Marine sediment shows attributed, dated Curation events, including ADD_CAUSAL_GRAPH on 2026-09-04 by codex and SOURCE_SYNONYM_SCOPED on 2026-10-07 by seed_from_sources, with their change summaries. Evidence
• Source link present: Marine sediment’s Provenance section links View the record directly to data/habitats/aquatic/marine_sediment.yaml in its GitHub repository. Evidence
• Domain visual present: The semantic map contains all 3,208 habitat descriptions and links matching records. Marine sediment renders an SVG causal graph and offers a Show at full size control. Evidence
• Cross-Mech links missing: The sampled habitat records do not expose specific sibling-Mech or kg-microbe record links. Associated taxa remain plain NCBITaxon CURIEs, and the footer links the kg-microbe repository. Evidence
• Corpus statistics present: The landing shows 3,208 habitat records, 2,154 ontology-grounded, 176 corroborated by two or more sources, and 698 with REVIEWED status, alongside category and grounding summaries. Evidence
• Category browsing present: Landing and Browse link eight named categories with counts, including 468 Aquatic and 1,649 Host-associated records; category listings provide pagination and category-wide text search. Evidence
• Suite link present: All Mech projects is visible in the shared header and semantic-map wrapper and links to the X-Mech Suite directory. Evidence
• Downloads partial: Reviewed navigation links the repository, individual record YAML and the map’s provenance manifest. It does not link a bulk export of complete habitat records. Evidence
• Schema docs missing: The reviewed landing, category, record and map navigation links the source repository but does not expose LinkML schema documentation. Evidence
• License partial: The sampled templated pages state project data CC BY 4.0, code BSD-3-Clause and third-party terms. The linked standalone semantic-map page omits the project license statement. Evidence
• Build or version partial: Main-page footers date the upstream kg-microbe extraction to 2026-08-16. The linked semantic-map manifest dates that projection to 2026-10-11T05:20:48.381814+00:00, rather than identifying a common build of all site pages. Evidence
• Theme toggle partial: There is no user-operated theme toggle. Templated pages follow the operating-system light/dark preference; the linked standalone semantic map keeps its light appearance. Evidence
• Narrow screens present: The landing, Browse, Aquatic and Host-associated category entries, two sampled records and both semantic-map pages fit fresh and resized 390 px viewports. This recheck samples these routes rather than every category or record. Evidence
• Load errors present: Aborting the Host-associated category JSON request makes search explicitly report filter unavailable while retaining the static unfiltered rows. The checked record pages and semantic map render without data fetches. Evidence

CommunityMech

• Search partial: Name search ANME returns 2 openable records, but CommunityMech:000134 and NCBITaxon:2157 return 0. The search does not index identifiers. Evidence
• Filters present: Category and ecological-state checkboxes combine with counts and Clear All: SYNTROPHY shows 32, adding STABLE shows 20, and clearing restores 458. Evidence
• Result counts present: The matching count updates independently from the 458 total: ANME shows 2, an unknown query 0, SYNTROPHY 32 and SYNTROPHY plus STABLE 20. Evidence
• Sorting missing: The browser has no sort selector or sortable headings; cards retain source order while filtering. Evidence
• Linkable search missing: Search and filter changes leave the URL unchanged. Loading browser.html?q=ANME#category=SYNTROPHY restores neither and shows all 458 records. Evidence
• Complete catalog present: The browser exposes 458 distinct community links, matching 458 pinned kb/communities records. Four sampled records open; four repository isolate records remain outside the published total. Evidence
• Record pages present: Browser cards link stable individual community pages. ANME, Aalborg, Copper Biomining and the new Alseth phage community all open successfully. Evidence
• Term links present: Sampled records display ontology labels beside NCBITaxon, ChEBI and ENVO resolver links. Curated environmental descriptions are distinguished from ontology labels; retired MIM ingredient identifiers remain plain text. Evidence
• Evidence links present: Sampled records link PMID and DOI evidence with snippets and support/source-type labels, including ANME PMID:37747940 and phage PMID:38648198. Evidence
• Curation status missing: Sampled record headers show ecological state rather than record curation status; evidence SUPPORT and source-type labels do not state a record review status. Evidence
• Curation history missing: The four sampled record pages show no curation-history section, including Aalborg and the new phage record, although the repository keeps history provenance. Evidence
• Source link partial: The View on GitHub footer links the repository root rather than the displayed community YAML file. Evidence
• Domain visual present: Sampled records render connected SVG interaction networks. The PaCMAP and graph-layout pages show 322 and 292 communities; search and color/size selectors work on these separately generated maps. Evidence
• Cross-Mech links present: Copper Biomining links its Standard 9K Medium to a specific CultureMech normalized recipe, pages/normalized/003277.html; the destination returned 200. Evidence
• Corpus statistics present: The live landing reports 458 communities and 16 categories, matching the pinned community record inventory and schema category enum. Evidence
• Category browsing partial: Categories are filters within the browser rather than standalone subset pages. The landing category tile links the category facet, which is accessible through Filters at 390 px. Evidence
• Suite link present: Landing, browser, both maps and four sampled record headers visibly link All Mech projects to culturebotai.github.io/mechs/. Evidence
• Downloads partial: The inspected landing, browser, maps and records link the repository but no bulk community export; the repository release list is empty. Evidence
• Schema docs partial: The landing GitHub card mentions the schema but links only the repository root; inspected navigation has no schema documentation or schema-file link. Evidence
• License missing: Landing, browser, both maps and four sampled records state no data or code license. The pinned repository licenses data under CC BY 4.0 and code under BSD-3-Clause. Evidence
• Build or version missing: The inspected pages show no build date, commit or dated corpus version; record footers say Generated by CommunityMech and embedding version v3 is undated. Evidence
• Theme toggle present: Choosing dark on the landing stored mech-theme=dark and persisted across the browser, both maps and four sampled records, with consistent dark backgrounds. Evidence
• Narrow screens present: Landing, browser, both maps and four sampled records fit a 390 px viewport without page overflow. Filters reveals the mobile browser controls. Evidence
• Load errors not applicable: The inspected landing, browser, records and both maps use inline data; their observed requests include no fetch or XHR data index to abort. Evidence

TaxonMech

• Search present: The corpus search finds NCBITaxon:2173 (32 substring matches) and Methanobrevibacter smithii (26); NCBITaxon:562 gives 259 matches across two result pages. Evidence
• Filters missing: Browse offers a text search but no property filter controls. Domain and type-strain subsets are separate linked pages. Evidence
• Result counts present: Search shows the matching count separately from the 625,960-record corpus, for example 26 of 625960 records match for Methanobrevibacter smithii. Evidence
• Sorting missing: Browse and subset tables retain a fixed order. Their headings and controls do not provide a choice of sort order. Evidence
• Linkable search missing: Typing and submitting searches leaves the URL unchanged. Opening ?q=NCBITaxon%3A2173 restores the unfiltered 625,960-record listing. Evidence
• Complete catalog present: The browser exposes 3,130 pages for 625,960 records; the last page opens at page 3130 of 3130. Domain and type-strain listings provide additional routes into the same corpus. Evidence
• Record pages present: Stable taxon.html?id=NCBITaxon:N routes load individual records, including Methanobrevibacter smithii (2173), with paged strain details. Evidence
• Term links present: Taxon identifiers and named lineage terms link to NCBI. The main taxonomy sections now link LPSN identifiers and GTDB names through working resolver routes; strain and assembly identifiers are separate from ontology grounding. Evidence
• Evidence links present: The Methanobrevibacter smithii record links its LPSN DOIs and PMID:390357, and strain details link source references. Evidence
• Curation status present: Record headers and browse rows show their curation status, including SEEDED on Methanobrevibacter smithii; the unchanged pinned corpus also retains its deprecated records. Evidence
• Curation history present: Records display timestamp, curator, action and change text. Methanobrevibacter smithii shows the 2026-09-13 seed_from_sources event and its attesting sources. Evidence
• Source link present: Record views link their own repository YAML, such as data/taxa/archaea/methanobrevibacter_smithii.yaml. Evidence
• Domain visual present: Each checked record displays a labeled, linked taxonomy lineage from root to its parent taxon, alongside taxonomy mappings and strain details. Evidence
• Cross-Mech links partial: Record details name kgmicrobe.strain identifiers as plain text; the sibling-project footer does not link these to specific sibling or kg-microbe records. Evidence
• Corpus statistics present: The landing page reports 625,960 taxa, 100,745 listed strains and 20,648 taxa with a type strain, with per-domain counts. Evidence
• Category browsing present: Landing links open domain listings and the type-strain subset. The checked Archaea page reports 14,013 records; the type-strain page reports 20,648. Evidence
• Suite link present: Landing, browse, source, domain, type-strain and record pages carry an All Mech projects link to the suite directory. Evidence
• Downloads partial: Sources links upstream catalogue snapshots and their manifest; record pages link individual YAML files. A bulk download of the harmonized taxon records is not linked. Evidence
• Schema docs partial: The landing page describes LinkML validation but links the repository root rather than published schema documentation or the schema file. Evidence
• License present: The checked page templates state project-authored data CC BY 4.0 and code BSD-3-Clause, with links to the corresponding license files and a reminder that upstream terms apply. Evidence
• Build or version partial: Sources gives upstream snapshot dates and versions. The footer names its generator, but the site does not identify its own build date, commit or corpus version. Evidence
• Theme toggle present: The toggle saved mech-theme=dark; landing, sources, type-strain, browse and taxon pages opened in dark mode in the same browser context. Evidence
• Narrow screens present: Landing, browse, Sources, type-strain, Archaea and taxon record pages fit a 390-pixel viewport. The former Sources-table overflow is fixed. Evidence
• Load errors present: Aborting pages/index.json.gz preserves static pagination and says Search is unavailable. Aborting a taxon-details shard displays an explicit load error with Browse and Try again controls. Evidence

TraitMech

• Search present: Browse searches all 1,089 live records: METPO:1004002 and traitmech:000616 each return one record; bipolar mating system returns two linked matches. The sampled result pages open successfully. Evidence
• Filters present: Browse provides Category and Kind selectors plus Clear filters. Pathogen with ECOLOGY and CLASS returns six records; changing Kind to OBJECT_PROPERTY gives an explicit zero count, and reset restores all 1,089. Evidence
• Result counts present: The live count updates with text and property filters: identifiers give 1 of 1089 records shown and pathogen plus ECOLOGY and CLASS gives 6 of 1089. Evidence
• Sorting missing: Browse has no sort selector or sortable heading behavior. Clicking the first heading leaves the first rows in their original order. Evidence
• Linkable search present: Search, category and kind are written to the URL. Reloading ?q=pathogen&category=ECOLOGY&kind=CLASS restores all controls and six matching records. Evidence
• Complete catalog present: Browse exposes 1,089 linked record rows, matching the landing total. Ten category cards lead to subsets of that same listing. Evidence
• Record pages present: Individual traits have stable static pages linked from Browse and related-trait lists, including traits/ecology/animal_pathogen.html and traits/physiology/bipolar_mating_system.html. Evidence
• Term links partial: Record identifiers have readable labels and METPO resolver links, but aerobic respiration’s Cross-references section still presents GO:0009060 as plain text; graph groundings are not consistently linked. Evidence
• Evidence links present: Animal pathogen and aerobic respiration link their trait and causal-edge DOI citations, with supporting text displayed alongside the evidence. Evidence
• Curation status present: Record headers and Identifier cards display workflow status, including REVIEWED on animal pathogen. Browse retains the complete corpus across REVIEWED, PROPOSED, SEEDED and DEPRECATED records. Evidence
• Curation history present: The checked trait pages display dated curation events, attributed curators and change summaries. Some records additionally link recorded curation-session sidecars. Evidence
• Source link present: Record Source cards link View YAML on GitHub, Raw YAML and Edit on GitHub for the record’s own data/traits file. Evidence
• Domain visual present: Record pages render causal SVG diagrams. The PaCMAP and graph-layout views render the 477 embedded traits, with searchable points and record links. Evidence
• Cross-Mech links present: Wood–Ljungdahl’s Pathway context links directly to PathwayMech WikiPathways:WP5019, whose Acetogenesis page opens successfully. The source page states organism scope and its pinned PathwayMech revision. Evidence
• Corpus statistics present: The landing reports 1,089 trait records, 10 populated categories and 43.8% embedded; Browse gives the same corpus total and per-category counts. Evidence
• Category browsing present: Browse links ten category pages with counts, including ECOLOGY with 29 records and QUANTITATIVE_PROPERTY with seven. Evidence
• Suite link present: All Mech projects visibly links the landing, Browse, sampled categories and records, both embedding views and the linked app/discussions index to the X-Mech Suite directory. Evidence
• Downloads partial: The reviewed site links the repository and per-record Raw YAML. The map CSV contains only the 477 embedded points, rather than a bulk export of all 1,089 complete trait records. Evidence
• Schema docs partial: The landing describes source code, data and schema but points to the repository root. Reviewed navigation does not expose LinkML schema documentation. Evidence
• License missing: The reviewed landing, Browse, categories, records and map pages do not state the project’s data and code licenses. The pinned repository licenses project data CC BY 4.0 and code BSD-3-Clause. Evidence
• Build or version present: Browse and map footers state Corpus as of 2026-10-11 02:46 UTC, the METPO baseline, 1,089 TraitRecords and embedding coverage. Record pages also carry record-specific dates. Evidence
• Theme toggle present: Selecting dark on the landing persists through Browse, both sampled categories, three records, PaCMAP, graph-layout and the linked app/discussions index in the same browser context. Evidence
• Narrow screens partial: Fresh 390px loads fit the reviewed landing, Browse, categories, records, discussions index and both maps. Resizing either map from 1280px to 390px leaves the document 1,133px wide. Evidence
• Load errors present: Aborting each map’s JSON request produces explicit Unavailable and Map data could not be loaded text with Retry and Browse alternatives. Aborting app/discussions/data.js gives Discussion counts unavailable, Discussions could not be loaded and Retry loading discussions. Evidence

PathwayMech

Site-specific check on October 11, 2026, deployed revision 720ebcd. All 24 catalogue features were reassessed in Chrome on the recorded successful deployment and independently checked against the same criteria. Checks covered all 198 browse links, identifier and label queries, counts/reset and cold query restoration, five record-source families, SVG zoom/node selection/evidence navigation, all three KGX downloads, both OS color schemes, 390 px layouts, and verification that browse/record data is pre-rendered.

• Search present: Search over all 198 records finds MetaCyc:PWY-6543 (1 result) and trehalose (2); matching record links open stable pages. Evidence
• Filters missing: The main listing offers a labeled search box and Reset, but no filter on pathway properties. Evidence
• Result counts present: The matching count changes with search: MetaCyc:PWY-6543 shows 1 of 198, trehalose 2 of 198, and an unmatched query 0 with a no-match message. Evidence
• Sorting missing: The browse page has no sort selector or sortable headings; records retain the generated order. Evidence
• Linkable search present: The q query parameter changes with search; reloading browse.html?q=glycolysis restores the input and its 1-of-198 result view. Reset clears the query. Evidence
• Complete catalog present: The browser exposes 198 distinct pathway links, matching the landing total. Five sampled records from MetaCyc, WikiPathways, GO-CAM, GO and MIBiG open successfully. Evidence
• Record pages present: The listing links stable records/<identifier>.html URLs; five sampled source families open as individual pathway pages. Evidence
• Term links partial: Organism-scope taxa have labels and NCBI resolver links, but the participant and mechanistic-edge tables still render labeled chemical, protein and reaction CURIEs without resolver links. Evidence
• Evidence links present: Sampled pathway pages link evidence references and source locations, distinguishing structured source assertions from quotations. The MetaCyc record links Rhea and PubMed evidence. Evidence
• Curation status missing: The five sampled record pages show pathway type and editing history, but no record-level curation or review status. Evidence
• Curation history present: The five sampled records show dated history with curator, action, summaries and links to retained curation sessions. Evidence
• Source link present: Read the source YAML links each sampled record to its own GitHub file under data/pathways/. Evidence
• Domain visual present: Pathway pages render directed SVG networks with typed nodes. Zoom changes the view from 47% to 65%; the node finder highlights a selected node and its relationship. Clicking an arrow opens its matching #mechanism-edge-1 evidence row. Evidence
• Cross-Mech links missing: The five sampled record pages offer suite navigation, but no links to specific sibling-Mech or kg-microbe records. Evidence
• Corpus statistics present: The generated landing reports 198 pathway records, 20 taxa, 8,977 mechanistic edges and 2,138 distinct references. The downloadable manifest also declares 198 pathway records and 8,977 graph edges. Evidence
• Category browsing missing: The landing leads to one flat pathway browser; no category, source-database or taxon subset index is offered. Evidence
• Suite link present: Landing, browse and sampled record navigation visibly link All Mech projects to culturebotai.github.io/mechs/. Evidence
• Downloads present: The landing links full KGX node and edge TSV exports and a manifest of version, row counts and checksums. All three links returned 200 during the check. Evidence
• Schema docs missing: The landing and sampled navigation provide no LinkML schema documentation link. The linked curation guide describes validation commands but does not link a schema. Evidence
• License partial: The landing download section states data CC BY 4.0 with upstream terms, but omits the code license; browse and sampled record pages have no license statement. Evidence
• Build or version partial: The landing presents an undated graph-content hash and manifest; record citations carry upstream dates, but no site build date, displayed commit or dated version of its own pathway corpus is stated. Evidence
• Theme toggle partial: There is no theme toggle. Switching the browser OS color preference changes the page background between light and dark. Evidence
• Narrow screens present: Landing, browse and five sampled record pages fit a 390 px viewport without page overflow; record tables and diagrams stay within their own regions. Evidence
• Load errors not applicable: Observed landing, browse and sampled record requests are HTML, CSS and the diagram script; all record data is pre-rendered and no data fetch or XHR is made. Evidence

CellStructureMech

Site-specific check on October 11, 2026, deployed revision d9cea09. All 24 catalogue features were reassessed in Chrome on the recorded successful deployment and independently checked against the same criteria. Checks covered the 909-row browser, identifier and label searches, combined category/kind filters, reset and URL restoration, four structure pages, the text map and neighbors, saved theme across nine routes, fresh and resized 390 px layouts, and both map-artifact failure paths.

• Search present: Browse searches all 909 linked records: GO:0097589, MitoStore and cellstructuremech:insoluble_protein_deposit each return one match, while an unmatched query returns zero. Evidence
• Filters present: Browse offers Category and Kind with Clear filters. MICROCOMPARTMENT plus PROTEIN_SHELLED_ORGANELLE returns five records; reset restores all 909. Evidence
• Result counts present: Search and property filters update the matching count separately from the total, including 1 of 909 records shown and 5 of 909 for the combined microcompartment filters. Evidence
• Sorting missing: Browse exposes no sort control, and clicking the first table heading leaves the first rows in their existing order. Evidence
• Linkable search missing: Search and property filters leave the URL unchanged. Reloading clears them, and opening a ?q= identifier URL restores an empty search and all 909 records. Evidence
• Complete catalog present: Browse has 909 linked record rows, matching the landing total. Thirteen category pages and the GO-grounded listing provide additional routes to the same static records. Evidence
• Record pages present: Static record URLs open successfully, including structures/appendage/archaeal_type_flagellum.html, attachment_organelle_membrane.html, envelope/s_layer.html and cytoskeleton/stentor_myoneme_system.html. Evidence
• Term links partial: Own GO identifiers, labeled part-of terms and UniProt locations are linked, but attachment organelle membrane’s mechanism Subject and Object cells omit the source grounding CURIEs and resolver links. Evidence
• Evidence links present: The reviewed records link DOI, PubMed and ontology evidence from components, functions, mechanism edges and record-level Evidence sections. Evidence
• Curation status present: Record headers and the Browse Status column display Proposed. The sampled archaeal-type flagellum, attachment organelle membrane and S-layer pages expose their workflow state alongside separate curation histories. Evidence
• Curation history present: Records show dated curator actions and change summaries, including archaeal-type flagellum’s creation, expansion and protein-example events and subsequent attributed edits. Evidence
• Source link present: The Source link at the end of each sampled record opens its own data/structures YAML file in the GitHub repository. Evidence
• Domain visual present: The PCA text-embedding map renders structure points, a category selector and full-vector neighbor details; selecting GO:0097589 displays the archaeal-type flagellum record and its semantic neighbors. Evidence
• Cross-Mech links present: The S-layer record’s Associated traits section links traitmech:000064 directly to TraitMech’s S-layer record. Evidence
• Corpus statistics present: The landing reports 909 structures, 13 categories and 760 GO-grounded records; its category table includes 214 membrane organelles. Evidence
• Category browsing present: Thirteen labeled category buttons and the landing category table lead to subset pages. The checked Microcompartment category has six records and search/filter controls. Evidence
• Suite link present: All Mech projects visibly links the landing, Browse, categories, records and text map to the X-Mech Suite directory. Evidence
• Downloads partial: Reviewed navigation links the repository and individual record YAML files without offering a bulk export of all complete structure records. Evidence
• Schema docs partial: The Source and schema card describes LinkML-validated YAML but links only the repository root rather than schema documentation. Evidence
• License present: The sampled landing, Browse, category, grounded listing, four records and map state project data CC BY 4.0, code BSD-3-Clause and third-party terms. Evidence
• Build or version missing: The shared footer names scripts/render_pages.py as its generator but does not state a site build date, commit or corpus version. Evidence
• Theme toggle present: Dark mode selected on the landing persists through Browse, GO-grounded, the Microcompartment category, four records and the text map; each has the toggle. Evidence
• Narrow screens present: Landing, Browse, GO-grounded, Microcompartment, four sampled records including Stentor myoneme, and the text map fit both fresh and resized 390px viewports. Evidence
• Load errors present: Separately aborting the text map’s structure and neighbor JSON requests produces Could not load embedding artifacts: Failed to fetch. Browse and record pages are pre-rendered. Evidence

ProteinTraitsMech

Site-specific check on October 11, 2026, deployed revision cc99114. All 24 catalogue features were reassessed in Chrome on the recorded successful deployment and independently checked against the same criteria. Checks covered identifier and label searches, all five axes and 429,293 records, combined category/query filters, reset and cold hash restoration, three cold record routes, all three map modes, schema and SLC10 pages, saved dark mode, fresh and resized 390 px layouts including expanded categories, and a failed facet-index request.

• Search present: Searching the full corpus finds cd05137 (one record) and homogentisate 1,2-dioxygenase (16 records), with results loaded from the published shards. Evidence
• Filters present: Axis, Source, Category and Status checkboxes filter records. STRUCT_ACTIVE_SITE selects 1,004 records; adding racemase gives nine. Clear all resets facets while retaining the search text; the search box is cleared separately. Evidence
• Result counts present: The results count updates independently of the fixed 429,293-record badge. STRUCT_ACTIVE_SITE plus racemase shows nine matches, an unmatched query shows zero, and facet counts explicitly state whether they include search text. Evidence
• Sorting missing: Browse provides result paging but no control to change result order; the checked record and map interfaces do not add a listing sort. Evidence
• Linkable search present: Search and facet changes update the URL. A cold load of #cat=STRUCT_ACTIVE_SITE&q=racemase restores both controls and nine matching records; #cat=STRUCT_ACTIVE_SITE&sta=REVIEWED restores 1,003 reviewed active-site records. Evidence
• Complete catalog present: Selecting all five axes exposes 429,293 records across 7,155 result pages, matching the site total; hash routes open individual records. Evidence
• Record pages present: Cold browse.html#record=CURIE links open individual records. MCSA:1, METPO:1000639 and ECOD:F.3755.3.1.364 render their own detail views. Evidence
• Term links partial: METPO:1000639 has a labeled OLS source link, and parent traits have internal record routes. MCSA:1 and ECOD:F.3755.3.1.364 explicitly say their source entry link is unavailable, so own identifiers do not all resolve. Evidence
• Evidence links present: Reference-bearing records expose linked PMIDs and DOIs; MCSA:1 includes linked M-CSA mechanism citations. Source links remain separate from scientific references. Evidence
• Curation status present: Record cards and detail views show SEEDED, PROPOSED or REVIEWED status, and the Status facet filters the corpus. Evidence
• Curation history present: Record details render Curation history with timestamp, curator, action and LLM-assisted attribution; MCSA:1 shows its edison-causal-graphs SEEDED-to-REVIEWED event. Evidence
• Source link present: Record details link the repository YAML source, including the MCSA glutamate-racemase record, separately from the external database entry. Evidence
• Domain visual present: MCSA:1 displays a 254-residue sequence with 40 stored features and causal graphs. Corpus, protein and sequence embedding maps provide complementary views with coverage information. Evidence
• Cross-Mech links missing: Checked record views link databases and other ProteinTraitsMech records, but not specific sibling-Mech or kg-microbe records. METPO:1000639 opens OLS rather than a TraitMech record. Evidence
• Corpus statistics present: The landing page reports 429,293 records, five trait axes, 56 categories and 34 source labels. The corpus map states that 344,797 current records are plotted; the protein and sequence maps show 68,657 and 12,705 proteins. Evidence
• Category browsing partial: The landing axis, category and source links select hash-based filters in the generic record browser. A cold active-site and REVIEWED route restores 1,003 records; these filtered views meet the partial category-browsing criterion. Evidence
• Suite link present: The landing, browser, maps and sampled record pages visibly link All Mech projects to the suite directory. The linked SLC10 pilot has separate navigation without that link. Evidence
• Downloads partial: Browse links the facet/shard manifest, and maps offer CSVs of plotted points; no single bulk export of all full trait records is linked. Evidence
• Schema docs present: The schema guide describes ProteinTraitRecord, enums and causal graphs and links the LinkML schema in the repository; related guides explain worked examples and biophysical records. Evidence
• License partial: Core pages and maps state data CC BY 4.0 and code BSD-3-Clause. The linked SLC10 pilot mentions source licenses but omits the project data/code license statement. Evidence
• Build or version partial: Pages identify upstream releases and explain retained embedding coverage, but do not provide a site build date, repository commit or dated complete corpus version. Evidence
• Theme toggle partial: The shared toggle carries dark mode across core browse, record, map and documentation pages. The linked SLC10 pilot ignores the saved setting and has no theme toggle. Evidence
• Narrow screens partial: Core pages, three cold record routes, the three checked map modes and the SLC10 pilot fit 390 px. Expanding CATEGORY with Show all (56) widens the browser to 400 px, including after selecting STRUCT_ACTIVE_SITE. Evidence
• Load errors present: Aborting the browser facet index displays Failed to load index: Failed to fetch rather than an empty corpus. Sampled record views separately label absent source entries or supplemental information. Evidence

DUFMech

• Search present: The labeled search covers all 8,296 families. PF04149, its label Domain of unknown function (DUF397), IPR007278 and the new PF18141 each give one openable result. Evidence
• Filters present: Seed-status and characterization filters have a reset. EX_DUF gives 1,796, historical DUFs 373 and unknown candidates 6,127; Reset restores all 8,296. Evidence
• Result counts present: The matching count updates separately from the total: tested identifiers and labels give 1 of 8,296, while an unmatched query gives 0 and an empty state. Evidence
• Sorting present: All five sort options work: protein count ascending/descending, Pfam accession, family name and structure count. The first displayed rows change consistently with the chosen order. Evidence
• Linkable search present: Search, seed status, characterization, sort and page are encoded in the URL. A second-page DUF search with UNKNOWN_CANDIDATE, UNSCORED and Pfam ascending restores every control and its page on reload. Evidence
• Complete catalog present: The complete served index contains 8,296 families, matching 8,296 committed family HTML routes and 166 static listing pages. Sampled family pages open; the last static page works without JavaScript and contains 46 rows. Evidence
• Record pages present: Every measured Pfam identifier has a committed families/PFxxxxx.html route, linked from the catalogue. PF04149, PF02698 and the newly added PF18141 opened successfully. Evidence
• Term links present: Sampled family grounding shows readable labels with linked Pfam and integrated InterPro identifiers, including PF04149/IPR007278 and PF18141/IPR040812. Evidence
• Evidence links present: Family views link frozen source descriptions and references. PF02698 links UniProt accessions and unresolved InterPro PUB citations to their source entry; those citations are not presented as invented PMIDs or new functional validation. Evidence
• Curation status present: The served index marks all 8,296 families SEEDED; sampled pages separately display seed classification and UNSCORED characterization. EX_DUF naming history is distinct from record review status. Evidence
• Curation history present: PF04149 retains its dated Codex REVIEW event, no_change outcome and scoped non-scientific reports. PF18141 explicitly says no retained events or review were supplied. Evidence
• Source link present: Sampled family pages link their own YAML on GitHub at corpus source 0e1bbcbc62ff9b47e98d6100856f8aaa343179f9, plus a build-local YAML and JSON download. This source pin is distinct from the deployment commit. Evidence
• Domain visual present: PF04149 shows two frozen protein-domain tracks: A0A010YFW0 residues 9-60 of 67 and A0A010YYK3 residues 5-57 of 65. These depict sampled sequence coordinates, not a whole-family structure or functional assignment. Evidence
• Cross-Mech links present: PF04149 links specific ProteinTraitsMech Pfam and InterPro records at the retained source revision. The new PF18141 page explicitly marks it outside the historical cross-Mech scan. Evidence
• Corpus statistics present: The generated landing shows 8,296 families, 8,238 with InterPro IDs, 1,906 with structures and 8,242 with AlphaFold models. Seed categories are 6,127 unknown candidates, 373 historical DUFs and 1,796 EX_DUF families; upstream counters do not imply local structures or validated function. Evidence
• Category browsing present: The linked Categories index opens separate paginated EX_DUF, historical-DUF and unknown-candidate subsets with totals 1,796, 373 and 6,127. All three subset routes were opened. Evidence
• Suite link present: Visible All Mech projects links on landing, category, record, sources, schema and cross-Mech pages target culturebotai.github.io/mechs/. Evidence
• Downloads present: Sources & schema links the full index.json export, which returned 200 and contained all 8,296 families and provenance. Per-family YAML/JSON and the sampled-member files supplement the full export. Evidence
• Schema docs present: Sources & schema links readable LinkML documentation for FamilyRecord, FamilyCuration, evidence/history fields and enums, with a downloadable schema. Evidence
• License present: Shared page footers state project data CC BY 4.0 and code BSD 3-Clause, link their license files and retain third-party terms, consistent with the repository license policy. Evidence
• Build or version present: Footers state site build 95de50b9c2ba5da2, corpus source 0e1bbcbc62ff and dated frozen input IDs. These differ from the Pages deployment revision 9d7a87f74786. Evidence
• Theme toggle present: System/Light/Dark controls are available. Choosing Dark persisted through landing, category index, all three subsets, sources, schema, cross-Mech and three sampled family pages. Evidence
• Narrow screens present: Landing, main navigation pages and three sampled families fit 390 px in the initial theme; the same pages plus all three category subsets also fit in Dark. Wide tables stay within their scrolling regions. Evidence
• Load errors present: Aborting catalogue.json produces an explicit search-unavailable message stating URL parameters were not applied, while retaining 50 static rows and numbered pages. The last static page also works without JavaScript. Evidence

NaturalProductMech

Site-specific check on October 11, 2026, deployed revision 63e28c6. All 24 catalogue features were reassessed in Chrome on the recorded successful deployment, including its changed rabelomycin page, and independently checked against the same criteria. Checks covered all 3,115 browse records, identifier and label searches, map filters/reset and selected-record URL restoration, five compounds, the Terpenoids index, both maps, a sibling-record link, both OS color schemes, fresh and resized 390 px layouts, and separate map/search-artifact failures.

• Search present: Browse searches the complete 3,115-record listing by identifier and label: CHEBI:338412 and marneral each return one linked match, while an unmatched query returns zero. Reset restores all compounds. Evidence
• Filters partial: Browse offers text search and Reset only. Property filters are confined to the Structure map: TERPENOIDS gives 213 of 3,115 compounds, and Reset restores the full map. Evidence
• Result counts present: Browse updates its count to 1 of 3115 compounds match for both tested identifier and label searches. The Structure map also reports search and pathway-filter counts. Evidence
• Sorting missing: Browse retains fixed pathway groups and alphabetical lists. Its controls expose search and reset without a sort selector or sortable headings. Evidence
• Linkable search partial: The Structure map restores the selected compound from ?id=CHEBI:338412. Browse search does not update the URL and opening ?q=marneral leaves an empty search and the complete listing. Evidence
• Complete catalog present: Browse exposes 3,115 record links across eight filing pathways, matching the landing and Structure map total. The separate text map explicitly covers 3,114 compounds. Evidence
• Record pages present: Compounds have stable static URLs reached from the complete listing and map record selections. Anisomycin, rebeccamycin, staurosporine, dehydrohistidyltryptophanyldiketopiperazine and the changed rabelomycin page each open successfully. Evidence
• Term links partial: Record identities and terms are displayed, but the checked records leave CHEBI, NCBITaxon, MIBiG and GenBank grounding references unlinked to their resolvers. Evidence
• Evidence links partial: Anisomycin’s PMID:28373542 and occurrence DOIs remain plain text, with no PubMed or DOI anchors. The sampled biosynthesis records also display unlinked reference lists. Evidence
• Curation status present: The sampled compound headers display grounding and SEEDED curation status with a What these status tags mean explanation and links to record history. Evidence
• Curation history missing: Sampled compound pages do not display dated, attributed curation events. Their status explanation links GitHub record history, but the record view itself does not show the event history; rabelomycin’s new Open questions section does not add one. Evidence
• Source link present: Each sampled Provenance section links directly to the compound’s data/natural_products YAML, including alkaloids/anisomycin.yaml. Evidence
• Domain visual present: The Structure map renders chiral-fingerprint UMAP points with pathway filters, compound selection and structural neighbors. A separate text-embedding Corpus map renders 3,114 points. Evidence
• Cross-Mech links present: Anisomycin’s Elsewhere in the fleet section links CHEBI:338412 directly to AntibioticMech’s anisomycin record; the target opens successfully. The relationship is labeled same structure, same inchikey. Evidence
• Corpus statistics present: The landing reports 3,115 structures from 10 adopted sources, 3,407 producer claims, 808 supported producer claims and 2,342 occurrence-bearing records, plus pathway and field-coverage tables. Its current causal_graphs coverage is 265 records. Evidence
• Category browsing present: Eight filing-pathway cards give counts and open pathway indexes; the checked Terpenoids index covers 213 records. Evidence
• Suite link present: All Mech projects visibly links landing, Browse, pathway, record and map pages to the X-Mech Suite directory. Evidence
• Downloads partial: Reviewed pages link the source repository and individual compound YAML files but do not offer a bulk export of all complete natural-product records. Evidence
• Schema docs missing: The reviewed landing, Browse, pathway, record and map navigation does not mention or link LinkML schema documentation, though the repository contains a schema. Evidence
• License present: The sampled page footers state project-authored records CC BY 4.0 and code BSD-3-Clause, link both license files, and preserve upstream attribution terms. Evidence
• Build or version partial: The landing dates upstream source retrieval to 2026-10-07. Reviewed pages say they were generated from the committed corpus but provide no own build date, commit or dated corpus version. Evidence
• Theme toggle partial: The site responds to operating-system light and dark preferences, but the reviewed pages expose no user-operated theme toggle. Evidence
• Narrow screens present: Landing, Browse, Terpenoids, five sampled records including the long dehydrohistidyltryptophanyldiketopiperazine heading and current rabelomycin page, and both maps fit fresh and resized 390 px viewports. Evidence
• Load errors present: Aborting the Structure map’s data request produces Chemical map could not be loaded: Failed to fetch. The separately tested search-index failure is reported explicitly. Evidence

AntibioticMech

Site-specific check on October 11, 2026, deployed revision 0611c4e. All 24 catalogue features were reassessed in Chrome on the recorded successful deployment and independently checked against the same criteria. Checks covered 13 linked listing pages with 2,939 distinct record URLs, identifier and label searches, all 457 broad-query results, class/map filters and reset, selected-record and query URL restoration, four compound pages, both maps, saved theme across routes, fresh and resized 390 px layouts, and deliberate map/list data failures.

• Search present: The Chemical map searches all 2,939 compounds by identifier and label: CHEBI:2637 finds amikacin, and amikacin gives two matches. The mycin search reports 457 matches; 22 Show more clicks expose all 457 selectable results. Evidence
• Filters partial: Property filters are map-only at corpus level: ANTIFUNGAL selects 594 of 2,939 records, and Reset restores all 2,939. Class listings expose text search without record-property facets. Evidence
• Result counts present: Counts respond to searches and filters: amikacin gives 2 of 1,043 antibacterial records, CHEBI:2637 gives 1, and an unmatched query gives 0. Map search separately reports 457 matching mycin compounds and how many are shown. Evidence
• Sorting missing: The reviewed class listing and maps expose no sort control. Clicking the Compound table heading leaves the first row and row order unchanged. Evidence
• Linkable search partial: The Chemical map restores selected CHEBI:100241 as ciprofloxacin from ?id=. Its text search clears on reload, and ?q=amikacin opens an unfiltered antibacterial listing with an empty input. Evidence
• Complete catalog present: Traversing all category and pagination links discovers 13 HTTP-200 listing pages with exactly 2,939 distinct record URLs. This matches the landing and map total; a 457-match map query exposes every result through Show more. Evidence
• Record pages present: Static compound pages are linked from listings and map selections. Amikacin, avibactam, D-cycloserine and enfuvirtide opened successfully at their own record URLs. Evidence
• Term links partial: Own compound identities and activity roles have labeled external links. Avibactam still shows Label unavailable beside each of four broader ChEBI identifiers, so grounding labels remain incomplete. Evidence
• Evidence links present: Avibactam links PMID:25406838 from its molecular-target and record evidence. D-cycloserine links PMID:4555420 and retains assay and organism limits alongside the pathway correspondence. Evidence
• Curation status present: Record headers and class tables expose workflow status: avibactam and D-cycloserine are REVIEWED, and amikacin is SEEDED. The landing reports 80 REVIEWED and explicitly distinguishes workflow state from human review. Evidence
• Curation history present: Avibactam retains dated source seeding and reseeding events plus its 2026-09-02 codex REVIEWED event, marked LLM-assisted and accompanied by the specific work described. Evidence
• Source link present: The Provenance section links View the record directly to the corresponding data/antibiotics YAML on GitHub. These source links follow main; deployment provenance is recorded separately in this audit. Evidence
• Domain visual present: The Chemical map renders the chiral-fingerprint UMAP and a selected compound with nearest neighbors. The separate Corpus map renders a canvas placing 2,939 compounds by record-text embeddings. Evidence
• Cross-Mech links present: D-cycloserine links directly to PathwayMech WikiPathways:WP5060 and retains its source revision. The record explains target/process correspondence, cites PMID:4555420 and states that it does not establish strain identity. Evidence
• Corpus statistics present: The landing reports 2,939 compounds, 2,669 ontology-grounded, 281 with both ChEBI and ARO source concepts, and 80 REVIEWED. It reports 454 records with mode of action, 282 with molecular targets, 308 with resistance mechanisms and 16 with causal graphs. Evidence
• Category browsing present: Browse and the landing class table link all seven antimicrobial classes to paginated listings. The antibacterial listing has 1,043 directly filed records; the landing includes its 78 antimycobacterial subclass records in the 1,121 antibacterial aggregate. Evidence
• Suite link present: The shared footer visibly links All Mech projects to culturebotai.github.io/mechs/, alongside sibling websites. Evidence
• Downloads partial: Record pages link their YAML and, where available, Complete record JSON.gz and activity-page data. Reviewed navigation exposes per-record resources without a linked full-corpus export. Evidence
• Schema docs partial: The landing Source and schema card explains LinkML-validated YAML, but its link opens the repository root rather than readable schema documentation. Evidence
• License present: Shared footers state project-authored data CC BY 4.0 and code BSD-3-Clause, link LICENSE-DATA and LICENSE-CODE, and retain upstream terms and attribution. Evidence
• Build or version partial: The footer names the upstream ChEBI/CARD input date 2026-08-30. It does not display an own site build date or deployment commit. Evidence
• Theme toggle present: A user-selected dark theme persists from the landing through Browse, the antibacterial listing, both maps and four sampled compound pages; every sampled route retains its toggle. Evidence
• Narrow screens present: The landing, Browse, antibacterial listing, both maps and four sampled compound pages, including enfuvirtide, fit fresh and resized 390 px viewports. Evidence
• Load errors present: Aborting Chemical map JSON displays Chemical map could not be loaded: Failed to fetch. Aborting antibacterial listing JSON displays filter unavailable while preserving fallback rows. Evidence

CMMMech

Site-specific check on October 14, 2026, deployed revision 51adf4b. All 24 catalogue features were judged in Chrome by AI auditors for the site's first entry in the table: each verdict by an auditor and an independent skeptic, with an arbiter for disagreements and a comparison against the other sites' verdicts, then a final browser check on the recorded successful deployment that confirmed every status. Checks covered all 403 record pages, identifier and label searches, kind, process and criticality filters with their clear control, URL restoration attempts, the six kind and seven process listings, ChEBI and organism links, saved theme across routes, 390 px layouts in both themes, and aborted data requests. A later review regraded search from present to partial, since the site's own record identifiers are not searchable.

• Search partial: Browse searches all 403 records by label and by the ChEBI, IMA and NCBITaxon identifiers it shows (CHEBI:30145 and IMA:2003-042 one match each, NCBITaxon:920 35), but not by the site's own cmmmech: record identifiers: cmmmech:calcite, shown on vaterite's page, returns 0 of 403, and cmmmech:lithium, shown on Browse, returns only the two records that cite it. Evidence
• Filters present: Browse offers Material Kind, Mechanism Process and Criticality selectors with Clear filters: Element gives 65, adding bioleaching 15, and Clear filters restores 403. Process options omit recycling. Evidence
• Result counts present: The live status line changes from 403 of 403 materials shown to 1 for CHEBI:30145, 65 for Element, 15 with bioleaching added, and 0 with a no-match message for an unmatched query. Evidence
• Sorting missing: Browse, kind and process listings offer no sort control. Clicking the Material, Kind and Criticality headings leaves the alphabetical row order unchanged. Evidence
• Linkable search missing: Search and filter changes leave the URL unchanged and reloading clears them. Opening browse.html?q=Cadmoindite, #q=Cadmoindite or ?kind=element shows all 403 records with empty controls. Evidence
• Complete catalog present: Browse links 403 record rows, matching the landing total and the 403 deployed record pages, all of which returned 200. The six kind listings partition the same 403 records. Evidence
• Record pages present: Each material has a static materials/<slug>.html page linked from Browse and the kind and process listings; cobalt, akaganeite, acanthite, uranium and fluorescent-lamp-phosphor opened at their own URLs. Evidence
• Term links present: Element records link labeled ChEBI groundings, and all 317 organism entries show labels with NCBITaxon links to NCBI. The 14 IMA mineral numbers are labeled plain text; 324 records show no own grounding. Evidence
• Evidence links present: All 1,542 reference entries across the 403 records link a DOI or source URL, and evidence points jump to them in-page. Some secondary PMIDs and DOIs in reference titles remain plain text. Evidence
• Curation status missing: Record headers show only kind and criticality badges. Evidence labels such as Support and Context_only and history actions do not state a record curation or review status. Evidence
• Curation history present: All 403 records show a Curation History table with timestamp, curator, action and summary; cobalt lists Codex's 2026-10-08 provenance adoption and 2026-10-09 criticality addition. Evidence
• Source link present: Each record's Record source line links its own data/records YAML on GitHub, such as cobalt.yaml; all 403 linked files exist on main. The links follow main rather than a pinned commit. Evidence
• Domain visual missing: No landing, Browse, kind, process or record page renders a diagram, map or network. The deployed HTML contains no SVG, canvas or image elements, and its two scripts only filter rows and set the theme. Evidence
• Cross-Mech links missing: Record pages link external resolvers and sources but no specific sibling-Mech or kg-microbe records. The footer links only sibling site roots and the kg-microbe repository. Evidence
• Corpus statistics present: The generated landing reports 403 materials, 404 mechanisms, 335 materials with mechanisms and 87 designated critical, plus kind and process counts. Its Biological Mechanisms tile includes 54 abiotic ones. Evidence
• Category browsing present: Landing cards open six kind and seven process listings with counts, search and filters; the kind pages partition all 403 records. Two recycling-only records have no process card or page. Evidence
• Suite link present: All Mech projects visibly links to culturebotai.github.io/mechs/ in the footer of the landing, Browse, every kind and process page and all 403 record pages. Evidence
• Downloads partial: Pages link the GitHub repository and each record's YAML. The deployed pages/index.json summarizes all 403 records without their full content and is not linked; no complete bulk export is offered. Evidence
• Schema docs missing: No deployed page mentions or links LinkML schema documentation; the only schema matches are the word schematic in evidence text. The repository holds src/cmmmech/schema/cmmmech.yaml. Evidence
• License present: Every page footer states data CC BY 4.0 and code BSD-3-Clause, linking the CC license and LICENSE-CODE, consistent with the repository license files. It does not mention third-party terms. Evidence
• Build or version partial: Footers name scripts/render_pages.py but give no build date, commit or corpus version. Besides curation timestamps, only upstream dates appear: each reference has an access date, such as cobalt's ChEBI entry, 2026-10-06. Evidence
• Theme toggle present: Dark chosen on the landing saved mech-theme=dark and persisted through Browse, kind, process and record pages via link clicks; every deployed page then opened dark with the toggle. Light persisted too. Evidence
• Narrow screens partial: Landing, Browse and all kind and process pages fit 390 px in both themes, but 364 of 403 record pages overflow: 87 criticality tables widen pages to 597 px and history tables add 3 to 69 px. Evidence
• Load errors not applicable: No page makes a fetch or XHR request; listings, search data and records are pre-rendered HTML. Aborting JSON, fetch and XHR matched nothing, and blocking record-browser.js leaves all static rows. Evidence

MediaIngredientMech

• Search present: Identifier and label searches work across the 2953-record browser: CHEBI:26710 gives 11 matches, table salt and UNMAPPED_0196 each give 1, and the previously unsearchable mapped term CHEBI:63895 now gives 2. A matched ingredient opens its record. Evidence
• Filters present: Status, source and quality menus now include the full observed status/source values, including REJECTED, MICRO, CAS and kgmicrobe registries. MAPPED + CHEBI + EXACT_MATCH gives 1357; REJECTED gives 81; Clear search and filters restores 2953. Evidence
• Result counts present: Showing N of 2953 ingredients updates after search and filter changes, alongside Currently Showing and a status breakdown. Tested counts include 11, 1, 1357 and 81. Evidence
• Sorting missing: The ingredient browser offers status, source and quality filters but no sort control. Results retain catalogue order, including the 11 CHEBI:26710 matches. Evidence
• Linkable search partial: The landing #status=MAPPED link restores 2611 mapped records. Search and other filter choices are not written to the URL; ?q=table+salt and #q=table%20salt load an empty search and all 2953 records. Evidence
• Complete catalog present: The live browser renders 2953 record links and the ingredient index lists the same 2953 records, matching the pinned catalogue and YAML files. Sampled mapped, rejected and unmapped result links open their stable pages. Evidence
• Record pages present: Ingredients have stable individual pages; UNMAPPED_0196 opens records/ingredient/unmapped/2-Carene-3-One.html, while NaCl and GYPS show mappings, evidence, chemistry and history. Evidence
• Term links present: Sampled ingredient and component terms have labels and resolver links, including CHEBI, ENVO, FOODON and MICRO; CAS numbers link to Common Chemistry. MeSH and BTO now have resolvers too. Local kgmicrobe registry identities remain plain text. Evidence
• Evidence links present: PMID, DOI and URL citations now link from record sections: NaCl links PMID:40820329, GYPS links https://doi.org/10.1099/00207713-52-5-1621, and NaCl’s role evidence links its source URL. The renderer also links DOI and URL references in discussions. Evidence
• Curation status present: Record headers show mapping status and quality. The live catalogue and pinned records agree on 2611 MAPPED, 260 UNMAPPED, 81 REJECTED and 1 AMBIGUOUS; Glucose_2 explicitly displays REJECTED. Evidence
• Curation history present: Sampled mapped, rejected and complex ingredient pages have Curation history tables with timestamp, curator, action and changes, including attributed LLM-assisted events. Evidence
• Source link partial: Record footers name the source YAML, for example data/ingredients/mapped/Nacl.yaml, as plain text without linking that file on GitHub. Evidence
• Domain visual present: PaCMAP and graph-layout maps display 2339 ingredients with color, size and facet controls. Map points now support keyboard inspection: Enter on the first point opens an inspector linking its individual ingredient record. Evidence
• Cross-Mech links present: GYPS now links CultureMech:002799 to its specific normalized recipe page, both from Related CultureMech recipe and its comparison evidence. The linked recipe returns 200. kg-microbe node identifiers on ingredient records remain plain text. Evidence
• Corpus statistics present: The landing computes 2953 ingredients, 2611 mapped and 88% coverage from data/ingredients.json, matching the pinned catalogue status counts. Evidence
• Category browsing partial: The ingredient index now has working identifier-prefix jump links, such as CHEBI (1848), but the landing still links only the full browser and MAPPED subset; its record breadcrumbs expose the separate index. Evidence
• Suite link present: The landing, browser, ingredient index, schema index and sampled record breadcrumbs visibly link All Mech projects to the suite directory. Evidence
• Downloads partial: The sampled site navigation links the repository and individual records but no bulk machine-readable export. The existing all_ingredients JSON/CSV assets remain unlinked from the landing/browser interface. Evidence
• Schema docs present: The browser footer now links a rendered HTML Schema reference, including classes, slots, enums and types; clicking IngredientRecord opens its class page. The raw Markdown index remains available as a secondary link. Evidence
• License missing: No data or code license is stated on the sampled landing, browser, maps, schema and record templates. The repository itself specifies CC BY 4.0 data and BSD-3-Clause code. Evidence
• Build or version present: Record footers and the ingredient index state build timestamps; NaCl shows Generated 2026-10-09T05:29:48+00:00 and GYPS shows 2026-10-09T05:29:44+00:00. Evidence
• Theme toggle present: Dark mode set on the landing persisted through the browser, ingredient index, both maps, sampled record pages and the HTML schema index/class page via mech-theme=dark. Evidence
• Narrow screens present: Landing, browser, ingredient index, both maps, three sampled records and the linked HTML schema pages fit 390 px without page overflow. Evidence
• Load errors present: Aborting data/ingredients.json produces Unavailable count tiles and Ingredients could not be loaded, with a Retry loading ingredients button. Evidence

CultureMech

• Search present: Searching CultureMech:000252 gives one recipe linking to halophilic_medium; the spaced label Halophilic medium gives four results. The identifier result opens a stable recipe page. Evidence
• Filters present: Checkbox facets on recipe properties combine with search and have Clear All. Category archaea gives 746 recipes; adding halo narrows to 148; clearing restores 15910. Evidence
• Result counts present: The browser shows the live matching count, including 15910 recipes, 746 recipes and 148 recipes during facet tests, and 1 recipe for CultureMech:000252. Evidence
• Sorting missing: No sort control is offered by the browser, either record index or the map data tables; sampled listings use their fixed catalogue order. Evidence
• Linkable search missing: Search and facet changes leave the URL unchanged; ?q=halophilic and #q=halophilic&category=archaea both open an empty search with all 15910 recipes. Evidence
• Complete catalog partial: The browser lists the 15910 normalized records. The landing page also counts 6320 merged records, but their pages/index.html listing is still not linked from the landing or browser navigation. Evidence
• Record pages present: Normalized and merged recipes have stable individual pages. Browser result 000252 and merged-index result media/000038 both open with HTTP 200 and show overview, composition, target organisms and provenance. Evidence
• Term links partial: Ingredient ChEBI terms and organism NCBITaxon terms show labels and OLS links on sampled records. However the source grounding mediadive.medium:1399 on 000252 is still unlinked, now explicitly marked source link unavailable instead of linking to #. Evidence
• Evidence links partial: References now render as links: 2asw shows both DOIs and its CCAP PDF, including growth-evidence citations. Source URLs in Notes remain plain text, such as the DSMZ Medium1399 PDF on 000252. Evidence
• Curation status missing: Sampled normalized and merged record pages and browser cards still show no curation or review status. The separate growth-evidence review contains review states but does not supply a status on each record. Evidence
• Curation history present: Sampled records show Provenance tables with timestamp, curator or script, action and notes. Record 000252 has seven events; 000038 includes the October 5 ingredient correction. Evidence
• Source link partial: The footer still names the source file as plain text, such as data/normalized_yaml/archaea/DSMZ_1399_HALOPHILIC_MEDIUM.yaml on 000252, without a link to that file. Evidence
• Domain visual present: The retained media PaCMAP and sfdp maps render and keyboard activation of 2asw opens normalized/000038.html with HTTP 200. Historical points without a verified current target show an explicit notice and a browser link. The ingredient map renders 653 points and a data table; Top 100 updates its visible count to 100. The old broken record Mermaid diagram is no longer rendered on 000252. Evidence
• Cross-Mech links present: Imported recipe 015438 now links CommunityMech:000064 to its specific pinned Soybean_N_Fixation_sfSynCom.yaml source record on GitHub; the linked target returns HTTP 200. Evidence
• Corpus statistics present: The landing page shows generated counts of 15910 normalized records, 6320 merged records and 5 populated categories, matching the pinned README and corpus inventory. Evidence
• Category browsing partial: Categories remain available through the browser facet and five headed sections on the long indexes, without category jump links or landing-page category cards. Evidence
• Suite link present: All Mech projects links to the suite directory from the landing footer, browser header, both indexes and sampled record breadcrumbs. Evidence
• Downloads partial: The landing page links the GitHub repository but offers no bulk export download. The browser loads all normalized records from app/data.js, which is not presented as a download. Evidence
• Schema docs partial: The landing GitHub card says the repository holds the schema but links only the repository root. No schema-documentation link appears in the sampled landing, browser, index or record navigation. Evidence
• License missing: No code or data license is stated on the sampled landing, browser, map, index or record templates. The pinned repository licenses project data under CC BY 4.0 and code under BSD-3-Clause. Evidence
• Build or version missing: The sampled landing, browser and record templates give no build date or deployment revision. Dates in record provenance and the separate growth-evidence report describe curation or that report, not the site build. Evidence
• Theme toggle partial: The dark preference persists from the landing through the browser, all three maps, both indexes and sampled records. The growth-evidence review linked from the merged index still has no toggle and remains light. Evidence
• Narrow screens partial: At 390 px the landing, browser, all three maps, merged index and sampled records fit. The normalized index still overflows to 747 px and the growth-evidence review to 520 px; the two media maps no longer overflow. Evidence
• Load errors present: Aborting app/data.js shows Recipes could not be loaded, explains that the catalog is unavailable and provides Retry loading recipes. Evidence

AIMech

Site-specific check on October 14, 2026, deployed revision 95b2869. All 24 catalogue features were judged in Chrome by AI auditors for the site's first entry in the table: each verdict by an auditor and an independent skeptic, with an arbiter for disagreements and a comparison against the other sites' verdicts, then a final browser check on the recorded successful deployment that confirmed every status. Checks covered all 1,159 record pages, identifier and label searches, origin, kind and branch filters with reset and cold URL restoration, the 17 kind and branch listings, saved theme across routes, 390 px layouts in both themes, and aborted data requests. The site redeployed during the final check, adding two records; the notes give the counts at the recorded revision.

• Search present: Browse searches all 1,159 records: AIMech:KernelPCA and AIO:HumanReportingBias each give 1 of 1159, and the label Kernel Principal Component Analysis gives 2. Matching rows open their record pages. Evidence
• Filters present: Origin, Concept Kind and AIO Branch selectors filter Browse, with Reset: Architectures gives 67, AIO plus Layer 168, curated architectures matching attention 16, and Reset restores all 1,159. Evidence
• Result counts present: Showing N of 1159 updates with search and filters, separately from the total: AIMech:KernelPCA gives 1, Adam 24, Architectures 67 and an unmatched query 0. Evidence
• Sorting missing: Browse, kind and branch listings have no sort control, and clicking each of the five column headings leaves the row order unchanged. Evidence
• Linkable search present: Search, origin, kind and branch are written to the URL. Reloading ?q=attention&origin=curated&kind=architecture restores all three controls and 16 matches; a cold ?origin=aio&branch=Bias load gives 61. Evidence
• Complete catalog present: Browse links all 1,159 records the landing counts, 717 curated and 442 AIO. All 1,159 record pages returned HTTP 200, and the nine kind and eight branch pages list the same records. Evidence
• Record pages present: Each record has a stable static URL, curated/<name>.html or aio/<name>.html, linked from Browse. Adam, KernelPCA, AStarSearch, AbstractRNNCell and TrainingStrategy opened, as did all 1,159. Evidence
• Term links present: All 837 AIO mappings show a labeled CURIE linked to its class page, which links its w3id IRI; parents are labeled links. Rejected candidates on 111 no-match records lack labels; subsets and owl:Thing are plain text. Evidence
• Evidence links present: All 1,105 sources on the 717 curated records link their documents, and claims link their source rows; DOIs, arXiv IDs and URLs in the ID column are plain text. One source and 11 AIO definition links are dead. Evidence
• Curation status present: Every record header shows its workflow status: all 717 curated concepts are DRAFT and all 442 imported AIO classes NOT_REVIEWED, as on Adam and AbstractRNNCell. Evidence
• Curation history partial: Curated records show curator Codex, method, model and created and updated times, but no event list or description of what changed. AIO class pages show only import provenance. Evidence
• Source link present: Every record page links its own YAML on GitHub, data/curated/<name>.yaml or data/aio/<name>.yaml. All 1,159 targets exist at the deployed revision, and the sampled links opened. Evidence
• Domain visual missing: No diagram, map, network or lineage view is offered. AIO pages list only direct parent classes and branch as text links, and the deployed pages contain no svg, canvas or image elements. Evidence
• Cross-Mech links missing: No record body links a sibling-Mech or kg-microbe record. The shared footer links only sibling site home pages and the kg-microbe repository. Evidence
• Corpus statistics present: The landing reports 1,159 concepts, 717 curated, 442 AIO classes, 2,069 claims and 1,105 evidence citations, matching the 2,069 claim cards and 1,105 sources on the record pages. Evidence
• Category browsing present: The landing links nine curated-kind pages and eight AIO-branch pages with counts, each a filterable listing; Architectures lists 67 and Biases 61, matching their cards. Evidence
• Suite link present: Every page footer visibly links All Mech projects to culturebotai.github.io/mechs/, which opened the X-Mech Suite page; checked on landing, Browse, kind, branch and record pages. Evidence
• Downloads partial: The header links the GitHub repository and each record links its YAML there, but no bulk export is linked. The served pages/index.json holds 1,159 summary entries and no page links it. Evidence
• Schema docs missing: No deployed page mentions a schema or LinkML. The repository's LinkML schemas under src/aimech/schema are neither linked nor published on the site. Evidence
• License partial: Footers say data CC BY 4.0 / BSD-3-Clause, code BSD-3-Clause, but the data BSD link opens LICENSE-DATA, which grants CC BY 4.0 to AIO-derived records only. The repository states no license for the 717 curated concepts. Evidence
• Build or version partial: Footers name only the generator script. AIO class pages give the upstream AIO commit 397851c547 and import time 2026-10-10T07:34:10Z; no site build date, deployed commit or corpus version appears. Evidence
• Theme toggle present: Dark chosen on the landing persisted through Browse, Curated, AIO Classes, kind, branch, curated and AIO record pages; switching back to light on a record page also persisted. Evidence
• Narrow screens partial: Landing, Browse, its Curated and AIO views, and all kind and branch pages fit 390 px in both themes, but 545 of 1,159 record pages overflow, up to 774 px, from unbroken IDs, IRIs and URLs. Evidence
• Load errors not applicable: Landing, Browse, kind, branch and record pages request only HTML, style.css and one or two scripts. Record data is pre-rendered; no fetch or XHR occurs, and aborting JSON and XHR requests changes nothing. Evidence