Artifacts

Directory layout

A typical run produces:

model-explorer/
├── manifest.json
├── standardized-output.json
├── sync-navigation.json
├── entity-node-mappings.json
├── models/
│   ├── <model>-source.json
│   └── <model>-hierarchy-<entity-kinds>.json
├── node-data/
│   └── <graph-collection>/
│       └── <overlay>.json
└── bundle/
    └── index.html

The exact number and names of hierarchy and overlay files depend on the MLIA results. Filenames are sanitized, length-bounded, and collision-disambiguated for portable output.

ETRecord runs can additionally include compact chain-mapping, ETDump backend trace and companion PTE collections. They use this same directory layout, manifest and graph-qualified mapping contract. See ETRecord profiling for their views and metric meaning.

File roles

Artifact	Purpose
standardized-output.json	Preserves the MLIA result supplied to visualization.
models/*-source.json	Contains the adapter-produced source graph collection, normalized to the Source Model label.
models/*-hierarchy-*.json	Contains generated graphs for result-local entity-kind paths.
node-data/**/*.json	Provides graph-indexed placement, status, metric, check, or advice overlays for named graph collections.
sync-navigation.json	Contains graph-qualified mappings used to synchronize panes.
entity-node-mappings.json	Relates MLIA entity IDs to collection, graph, and node-qualified references.
manifest.json	Records run metadata, artifact paths, overlay routing, and logical overlay display names.
bundle/index.html	Embeds all usable artifacts, host configuration, and the browser runtime in one file.

A format integration may add presentation.preferred_right_pane to the manifest, containing a collectionLabel and graphId. The bundler uses that reference when it identifies an available generated graph, including a later graph in its collection. Without a valid preference, the existing node-data coverage rule selects the initial right pane. Ordinary inputs omit this optional metadata.

Graph and node identity

Graph collection labels scope graph IDs, and graph IDs scope node IDs. Mapping records therefore use all three values. Consumers should not join data using a bare node ID, because separate subgraphs or generated collections may legitimately reuse it.

Model Explorer also creates artificial group rows from namespace prefixes. The plugin normalizes escaped slashes and alternative namespaces using the same rules as the browser runtime so overlays target the visible group rather than an unrelated node with similar text.

Overlay behavior

Each node-data file is a graph-indexed JSON object and is routed only to graph collections containing matching node or namespace keys. Logical overlay identity includes result index and kind, category, name, unit, and qualifiers; display names are disambiguated when those properties would otherwise appear identical.

Qualitative facts may be repeated on the source nodes participating in a higher level entity. Metrics use more conservative projection rules so a group total is not presented as though it were measured independently for every child.

Contract versioning

manifest.json currently carries its own schema version, and the embedded page carries a host contract version. These versions allow the plugin and vendored runtime to evolve together. The files are useful for inspection and integration, but their complete field layouts are not declared as a frozen public interchange schema.