> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.nvidia.com/docsfab/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/docsfab/_mcp/server. # Baseline Documentation Health in DocsFab > Run the multi-lens review-documentation job as a documentation-set baseline. As a documentation leader, you can use DocsFab to baseline health for an existing documentation set. The catalog job is `review-documentation`. It collects audit, lint, technical, editorial, audience, and quality evidence as separate lenses. There is no separate baseline job, and the result is not a single readiness score. For more information, refer to [About DocsFab Workflows](/docsfab/workflows). Audience `insufficient-evidence` still blocks a success claim. Treat that limitation as a gap in the baseline, not as a pass. Source updates have a separate outcome. Confirm an update from its own run and the resulting repository revision; this review's audience or quality limitations do not establish whether a separate update changed the source. ## Prerequisites Complete the steps in [Install DocsFab](/docsfab/get-started/installation). Use a Git repository that contains the documentation set you want to assess. Choose the commit that defines the baseline, and keep the run state directory available for inspection. Run the offline readiness check and inspect the catalog: ```bash cd /absolute/path/to/input-repository docsfab doctor docsfab list docsfab plan review-documentation export DOCSFAB_STATE_DIR='/absolute/operator/state' ``` Expected result: Doctor reports readiness for the current environment. `docsfab list` includes `review-documentation`. It may also list `readiness`, which checks agent readiness and is a separate workflow. `docsfab plan` prints the fixed `review-documentation` step graph. It does not read the target repository, run a model, or grant write authority. The `DOCSFAB_STATE_DIR` setting keeps the run record and report in one known place. This job is read-only. It does not edit files or open a merge request. ## Steps Run the workflow against one selected commit, then inspect the evidence saved for that run. The result keeps execution status, findings, and evidence limits separate so you can choose a follow-up without treating a partial lens as a pass. From the input repository, capture the full commit ID that defines the baseline. Replace `HEAD` with the intended commit or ref if you want to assess another revision. ```bash cd /absolute/path/to/input-repository export BASELINE_REVISION="$(git rev-parse --verify 'HEAD^{commit}')" ``` Expected result: `BASELINE_REVISION` identifies the commit that the run will inspect. DocsFab exports tracked files from that commit, so uncommitted changes are not part of the baseline. The run record stores the resolved commit for later comparison. Run the comprehensive review with no named page targets so the scope stays the full documentation set. Pass the recorded commit explicitly. ```bash docsfab run review-documentation \ --repo /absolute/path/to/input-repository \ --revision "$BASELINE_REVISION" ``` Expected result: The workflow runs its read-only review lenses against the selected revision. Its result includes findings, gaps, and limitations without a write grant. Observed source evidence remains useful when editorial, quality, or audience evidence is unavailable; it does not turn those limitations into a successful review. Replace `` with the ID printed by the review. Inspect the layered evidence instead of collapsing it into one score. ```bash docsfab runs show --state-dir "$DOCSFAB_STATE_DIR" docsfab runs show --json --state-dir "$DOCSFAB_STATE_DIR" ``` Expected result: Execution status, product disposition, review findings, quality limitations, and audience evidence remain separate fields. Inspect `evaluation.json` for the quality vector and findings, and `quality-report.md` for the human report, validated summary, and artifact locators. `quality-score.json` is optional and applies only to a safely selected content-type rubric. A missing score does not invalidate a separately validated layered evaluation. These are runtime-retained artifacts addressed by logical locators, not necessarily files written into the repository under review. Schema and cross-artifact checks establish consistency and provenance, not technical truth. Refer to [Evaluate Document Quality Skill in DocsFab](/docsfab/reference/quality/skills/assess-and-evaluate/evaluate-document-quality) and [Reading a Result](/docsfab/about/core-concepts/documentation-quality#reading-a-result). If the run summary says an HTML report is available, serve it from the same state directory when you need to read the evaluator Markdown as a page. ```bash docsfab reports serve --state-dir "$DOCSFAB_STATE_DIR" --port 8787 ``` Expected result: An available report loads at `/reports//`. Serving the report does not publish documentation or grant merge authority. Decide the next bounded job from the lenses, not from a composite score. Use [Update Documentation Beside a Change in DocsFab](/docsfab/workflows/update-documentation-beside-a-change) for a documented defect. Use [Gather Product Knowledge](/docsfab/workflows#gather-product-knowledge) when load-bearing product evidence is missing. Expected result: The follow-up request names one outcome and keeps write authority explicit. Use `docsfab list` to confirm the current workflow catalog before selecting a separate restructuring workflow. ## Audience Evidence Audience understanding is one lens inside `review-documentation`. It is not a sixth catalog job. The audience step tests the frozen candidate only when the run supplies an immutable target identity, an authoritative reference model, a versioned scenario corpus and oracle, a representative participant, independence, a retention boundary, and run authority. If a prerequisite is missing, no participant attempt runs; the step returns `insufficient-evidence`. The participant does not receive author or reviewer context, expected answers, the scoring oracle, or prior evaluation results. Without those inputs, the step returns `insufficient-evidence` and names the missing prerequisites. That result remains visible on the baseline. It does not become editorial approval, release readiness, or a passing audience check. Human or agent audience outcomes emitted by the model remain `model-claimed-only` even when their envelope passed host validation. Create and update outcomes also remain `insufficient-evidence` until a separately host-validated governed campaign supplies the required semantic validation receipt. A structurally complete audience record does not clear this gap. An authoring workflow can record an editorial update while audience evidence remains insufficient. Read those outcomes separately. Inspect the authoring run's change evidence to confirm what it produced. With `--patch-only`, DocsFab retains a patch without applying it to the source checkout. A retained patch does not prove that files changed in the checkout. Missing audience evidence on this baseline is the same kind of gap. Record it. Do not treat the rest of the review as a complete health claim. ## Troubleshooting If doctor reports blocked readiness, follow the printed next command. Refer to [Doctor Reports Blocked Readiness](/docsfab/troubleshooting#doctor-reports-blocked-readiness). If the HTML report is unreachable, keep `docsfab reports serve` running from the same state directory used for the run. Confirm that `DOCSFAB_STATE_DIR` or the `--state-dir` value points to that directory. Refer to [The Review Report Page Does Not Load](/docsfab/troubleshooting#the-review-report-page-does-not-load). If you are looking for a `baseline` catalog job, no workflow has that name. Use `review-documentation` for a documentation-set baseline. `readiness` is a separate catalog workflow for checking agent readiness. Run `docsfab list` to confirm the current catalog. If audience evidence is missing, do not rerun the unchanged review as a repair. Supply the required corpus, oracle, independence, and authority, or keep the `insufficient-evidence` limitation. ## Related Use these references to interpret the review lenses, choose a follow-up, and troubleshoot the run. [Review Documentation Contract](/docsfab/reference/workflow-contracts/review-documentation) [Review Merge Request Impact in DocsFab](/docsfab/workflows/review-merge-request-impact) [Test Audience Understanding Skill in DocsFab](/docsfab/reference/quality/skills/assess-and-evaluate/test-audience-understanding) [Evidence Before Drafting](/docsfab/about/core-concepts/documentation-quality#evidence-before-drafting) [Use Cases](/docsfab/about/use-cases) [Run a Documentation Workflow in DocsFab](/docsfab/workflows/run-a-documentation-workflow) [Troubleshooting DocsFab](/docsfab/troubleshooting)