[fas89]

What

Documents forge-cli #865, a Unity Catalog metric view from each Databricks expose's semantics block, and forge-cli #873, which clears what a product made outside OpenTofu in a schema the contract stops using, so the plan can destroy that schema. Both shipped in forge-cli 0.30.0.

docs/providers/databricks.md:

• a "Metric views" section: an example and the metric view YAML forge-cli writes for it, what maps to what, who can read the view, the hooks around tofu apply, dry-run, fluid diff --out, fluid generate iac, the [databricks] extra, the Genie space, a failed statement, and what fluid validate refuses;
• "A schema the contract stops using", under "What fluid apply provisions": the drops before tofu apply, the destroy that waits for the tables, the drops and the destroy-only second plan after it, what is dropped and what is left, and the dry-run output;
• one line each in the overview, the masks' "What the functions are", "What validate refuses", Limitations and "How it is tested".

docs/upgrading.md gets a "Changes in 0.30.0" section with one Databricks bullet for each.

Status

0.30.0 is released (tag v0.30.0 on fd1b78d5, 2026-11-13). By ancestry it contains #865 (224454b1), #873 (84da9dde) and #863 (97811add, the Genie space one bullet mentions), so every marker this PR adds is the release form, *([forge-cli #865](<url>), since 0.30.0)*.

The metric views text comes from a draft written against forge-cli 5cea5688, before #865 merged. It is ported onto main after #208 and #210. The one conflict was two sections added at the same place; "Metric views" now follows "Data quality monitoring".

What #873 changed in the draft

Three of the draft's claims were out of date after #873:

• Limitations said a metric view left in a schema the contract stops using is not found, and that the schema's destroy fails until the view is dropped by hand.
• The "Before tofu apply" row of the metric view hooks and the "another environment" bullet said views are dropped only in the schemas the contract's tables are in.

All three now say what 0.30.0 does, and link the new part.

Verification

• Ground truth: the source at tag v0.30.0, and data-product-forge[databricks]==0.30.0 in a scratch venv (its databricks*.py files are byte-identical to the tag).
• A verifier agent (sonnet, high effort) checked about 80 claims in two passes. It compiled the page's example with the 0.30.0 package: the metric view YAML it emits is byte-identical to the page's block, and fluid validate prints the page's derived-metric error word for word. It drove prepare_destroys in dry-run against forge-cli's stand-in, and the printed lines match the ones the page quotes. It found 10 false or imprecise claims, all fixed here: the dry-run lines for grants, revokes and drops; which fields merge; the source name refusal; the bare-name refusal; the refusal test count (39 refusals and 3 near misses, not 42 refusals); the missing-warehouse cases; the conditions for the second apply; the line printed for a drop in a destroyed schema; and a view over a table that no longer exists.
• The earlier review's correction stands, re-run on 0.30.0: fluid plan does not list the metric view; only the report fluid diff --out writes does.
• scripts/check_cli_docs.py (full, --version-only, and --markers-only against forge-cli main's CHANGELOG), scripts/check_providers.py and scripts/gen_contract_reference.py --check, with the pinned CLI 0.29.0: all pass.
• npm ci && npm run docs:build (239 pages) and node scripts/check-dist-links.mjs: pass. Every #anchor in the two pages resolves to a heading id in the built HTML. lychee offline over the source, with CI's route mirror: 0 errors.
• Not re-run: the live Free Edition runs the pages cite. They come from forge-cli's HONESTLY_TESTED.md at v0.30.0.

Left alone

• The markers for #861, #864 and #871 elsewhere on these pages still say "not yet in a release", although 0.30.0 contains all three. They are outside this PR.
• A Lakeflow pipeline's table gets no metric view yet (0.30.0's CHANGELOG). The page does not cover Lakeflow builds, so this PR does not either.
• The CLI pin stays at 0.29.0.