v0.7.0

Language Simplification (Feature 28)

• All pipe steps are now NL. The grammar no longer distinguishes structural
  vs NL pipe steps — every step in a pipe chain is natural language. The
  structural and mixed transform classifications are removed from the grammar,
  core, CLI, LSP, viz-model, and viz component.
• metric keyword replaced with schema(metric) metadata. Metrics are now
  regular schemas with a metric metadata tag, making them valid both as sources
  and targets. The dedicated metric keyword and its grammar rules are removed.
• Pipeline tokens (to_decimal, count, equals, etc.) are reframed as
  vocabulary conventions, not grammar-level constructs.
• All canonical examples updated to the simplified syntax.

Standalone Viz Harness (Feature 29)

• New @satsuma/viz-backend package extracts VizModel production logic from
  the LSP into a shared package consumable by both the LSP and the harness.
• New satsuma-viz-harness provides a fixture-driven browser app that renders
  source code alongside the <satsuma-viz> web component. Includes syntax
  highlighting, fixture picker, and lineage/single mode toggle.
• 8 Playwright browser tests covering overview rendering, detail view, hover
  events, cross-file lineage, navigation intents, and layout stability.
• LSP refactored to delegate viz model assembly to the shared backend.

VS Code Visualization

• @ref highlighting in mapping detail and edge layer — @ref tokens in NL
  transform text now render bold in a darker green for visual emphasis.
• Full transitive lineage display with file-scope filter dropdown.
• Field notes, arrow notes, and left-aligned transforms in the mapping detail.
• Schema card labels capped at 400px width; minimap anchored to visible viewport.
• Removed ClassificationFilter and structural CSS from field-lineage panel.
• Removed pipeline/mixed rendering branches and orange edge colour.

LSP

• Symbol-level import reachability — diagnostics now enforce that referenced
  symbols are actually imported, not just file-level reachability.
• Core semantic diagnostics via a new SemanticIndex adapter.
• Migrated field extraction, spread resolution, classification, metadata
  extraction, and CST text helpers from LSP into @satsuma/core.

CLI

• File-based entry points (ADR-022) — the CLI now requires .stm file
  arguments and rejects directory arguments. Import reachability determines the
  workspace scope.
• Fixed 3 lineage/graph JSON output bugs, 4 nl/arrows bugs, 6 validate/lint
  bugs, and 4 diff/dispatch bugs.
• Refactored graph.ts into graph-builder.ts and graph-format.ts.
• Migrated all test files from JavaScript to TypeScript.
• Added test coverage measurement with c8 and coverage summary in CI.
• Standardized JSON output format across 7 CLI commands.

Core

• Added source positions to FieldDecl and startColumn to Extracted* types.
• Added escape handling to stringText().
• Consolidated extraction tests and fixed sourceRefNameNs regression.

Grammar

• Simplified pipe_text — all steps are NL.
• Restored parenthesized NL tokens in pipe_text.
• Rejected empty backtick identifiers as parse errors.
• Replaced metric keyword with schema(metric) metadata decoration.

Infrastructure

• Extracted satsuma-lsp package from vscode-satsuma/server (ADR-021).
• Core extraction consolidation (ADR-020) completed.
• Added satsuma-viz-harness to root install:all, ci:all, and clean:all.
• Added build-artifacts.sh for unified release artifact builds.
• CI dependency bumps: actions/checkout v6, actions/setup-node v6,
  actions/upload-artifact v7, dorny/test-reporter v3.
• Retrospective ADRs 010–023 documenting implicit architectural decisions.

Documentation

• Added Satsuma Diaries section to the website.
• Archived 7 completed features (22–25, 29).
• Reorganised docs/ into developer/ and product-owner/ subfolders.
• Added adr-draft skill for assessing and drafting ADRs.