Add rosetta export: DB-only dataset directory (B-0022 E1+E2)

CHANGELOG.md

@@ -24,6 +24,8 @@ uses [Semantic Versioning](https://semver.org/).
 
 ### Added
 
+- **New `rosetta export <dir>` command writes a DB-only dataset directory for local audit** (issue #101, the B-0022 export track). Produces `manifest.toml` plus seven flat TSVs — `changelog.tsv`, `callouts.tsv`, `properties.tsv` (joined to pages with a resolved `section_anchor`), `videos.tsv` (metadata + per-video transcript segment/word/byte counts), `commands.tsv`, and the paired `pages.tsv` (one row per section) / `pages_summary.tsv` (per-page rollup) — from the runtime SQLite database alone — dataset generation reads only the resolved DB, with no caches or re-extraction (startup may still download/refresh the DB, as every command does). The serialization contract is a reversible Postgres-COPY-style escape (`\t`/`\n`/`\r`/`\\` backstopped on every value, whole-field `\N` for SQL NULL as distinct from the empty string) with stable ordering so a rebuild on the same DB is byte-identical, and `manifest.toml` carries `db_meta` provenance, per-file row counts, and honest disclosures of what the DB cannot provide (transcript provenance #21, per-version architecture #91). A column the DB cannot produce is omitted and disclosed, never recovered from a source artifact.
+
 - **Every Docusaurus pipe table is now retained as normalized, queryable data in schema v10** (issue #92). `page_tables` keeps raw Markdown plus page order, nearest h1–h6 heading, enclosing h1–h3 `section_id`, width, and raggedness; `page_table_rows`/`page_table_cells` preserve every header/data cell at its actual width with escaped pipes decoded. Table-derived `properties` now point to their exact source row through `source_table_row_id`, while bullet-derived and historical properties remain honestly NULL. Property output is byte-for-byte unchanged; the shared fence-aware parser also corrects the B-0022 census counting error caused by an ordered-list code fence in `zerotier.md` (the rc.98 corpus contains 855 tables / 8,287 data rows, not 852 / 8,258).
 
 - **README.md and DESIGN.md now describe the live Docusaurus corpus as current, not a pending migration** (issue #77). README's intro and "What's Inside" table dropped the "still needs a major migration away from the retired Confluence export" framing and refreshed prose-corpus figures for the actual Docusaurus build (363 pages/~653K words, 4,402 properties, 943 callouts), and gained a "Hardware overlay" row for `hardware_catalog`/`device_aliases` (255 devices incl. legacy/EOL + accessories, ~750 alias mappings) that was previously absent from the feature list entirely. DESIGN.md's "Corpus Snapshot" section is retitled "Historical Corpus Snapshot (legacy Confluence)" to stop reading as current-state to a skimmer; its already-caveated legacy figures are otherwise untouched.

MANUAL.md

@@ -27,6 +27,7 @@ Download a compiled binary from [Releases](https://github.com/tikoci/rosetta/rel
 | `browse` | Interactive terminal browser (REPL) |
 | `browse <cmd> [args]` | Run any TUI command once, then open REPL (e.g. `browse changelog 7.20..7.22`) |
 | `browse --once <cmd>` | Execute any TUI command and exit — no REPL (for piping) |
+| `export <dir>` | Write DB-only dataset directory (TSV + manifest.toml) |
 | `--setup` | Download DB + print MCP config |
 | `--setup --force` | Re-download DB |
 | `--refresh` | Shortcut for `--setup --force` (refresh DB) |
@@ -281,6 +282,37 @@ For a **prerelease** version specifically, `republish_assets: true` has extra ru
 - `package.json` is **not** rewritten in this mode.
 - No `npm dist-tag add` calls happen (npm publish is already fully skipped in `republish_assets` mode).
 
+## Dataset Export
+
+`rosetta export <dir>` writes a deterministic directory of spreadsheet-friendly datasets derived from the runtime database **alone** — dataset generation reads only the resolved SQLite DB, with no caches, no source artifacts, and no re-extraction. (Like every rosetta command, startup first ensures the DB is present and current, which may download or refresh it; that readiness step is the only network activity, and never a source for the exported data.) It is an audit surface for the artifact rosetta actually ships (rationale: `briefings/B-0022-runtime-sqlite-dataset-exports.md`).
+
+```sh
+bunx @tikoci/rosetta export /tmp/rosetta-datasets
+```
+
+The current pass writes one flat TSV per subject:
+
+```text
+rosetta-datasets/
+├── manifest.toml       # provenance (from db_meta) + serialization contract + per-file row counts + disclosures
+├── changelog.tsv       # changelogs: version, released, category, is_breaking, description, sort_order
+├── callouts.tsv        # callouts: page_id, section_id, type, content, sort_order
+├── properties.tsv      # properties joined to pages + resolved section_anchor
+├── videos.tsv          # video metadata + per-video transcript segment/word/byte counts
+├── commands.tsv        # command glossary: path, name, type, parent_path, page_id, description, ros_version
+├── pages.tsv           # one row per section: sizing (word/byte counts) + table counts (the granular view)
+└── pages_summary.tsv   # one row per page: rollup totals (section/table counts, bytes) — GitHub-readable
+```
+
+`pages.tsv` and `pages_summary.tsv` are the same core data for two audiences: the flat per-section rows suit a spreadsheet where you sort and group yourself, while the per-page rollup is small enough to read directly on GitHub. Product specs (`products/**`) and per-fragment table files (`pages/<slug>/**`) are not yet exported.
+
+Serialization contract (stated in full in `manifest.toml`):
+
+- **TSV**, UTF-8, LF line endings, a header row, and stable row/column ordering — a rebuild on the same DB is byte-identical.
+- **Escaping** follows the Postgres COPY text convention: `\\`=backslash, `\t`=tab, `\n`=LF, `\r`=CR, applied as a general backstop to every value. A field exactly equal to `\N` is SQL NULL; an empty field is the empty string. This keeps the one-line-equals-one-record property (parseable by `awk -F '\t'`) while remaining losslessly reversible.
+
+Because the export reads only the DB, a column it cannot produce is omitted and disclosed in `manifest.toml` rather than recovered from a source artifact. Overwriting a previous export is the working assumption; the command only rewrites the files it owns and never deletes anything it did not create.
+
 ## Database (Standalone)
 
 The SQLite database is downloadable on its own from [GitHub Releases](https://github.com/tikoci/rosetta/releases):