merman-analysis

merman-analysis owns the diagnostics-first JSON contract and the richer parser-backed analysis result for Merman lint, validation, document/fence source mapping, binding payloads, and editor projections.

The crate intentionally starts below FFI, UniFFI, WASM, CLI, and render wrappers. It provides stable JSON payload types, AnalysisResult syntax facts, DocumentSource extraction for plain Mermaid / Markdown / MDX, source-position mapping helpers, and the canonical policy for merging parser diagnostics with recovered editor facts.

Diagnostic ownership is intentionally narrow:

• merman-core emits structured parse diagnostics with exact spans, insertion points, or explicit fallback locations.
• merman-analysis maps those parser facts into stable rule ids, metadata, Markdown ranges, and duplicate/recovery policy.
• Editor-core, LSP, and VS Code project analysis payloads without adding semantic deduplication or rewriting recovered-parser messages.

Editor-facing ownership is layered:

• AnalysisResult carries document-level diagnostics plus per-diagram syntax facts.
• AnalysisFactsPayload is the serializable facts contract for bindings. It includes the diagnostics summary, document/fence spans, parser fact provenance, semantic items, outline items, expected syntax, references, and the first typed Flowchart projection.
• FenceTextIndex preserves parser-complete, parser-recovered, or text-scan provenance for semantic facts and expected syntax.
• merman-editor-core owns protocol-neutral completion, hover, symbols, navigation, rename, selection ranges, folding ranges, and semantic-token queries over snapshots projected from analysis facts.
• LSP, WASM, and VS Code convert those protocol-neutral results into host surfaces.

Rust API Migration Notes

DocumentDiagram::text, AnalyzedDiagram::text, and editor FenceSnapshot::text use SharedTextSlice instead of owned String buffers. The slice shares the immutable document text and stores UTF-8 byte bounds, so Markdown/MDX fence snapshots no longer copy every Mermaid body. Consumers that only read the source should use as_str() or AsRef<str>. Consumers that need an owned buffer can call to_owned_text().

See docs/adr/0070-diagnostics-first-analysis-contract.md for the accepted architecture decision and docs/adr/0072-lint-rule-governance.md for rule-origin, profile, and authoring-governance policy.