[miyoungc]

Description

Correct guardrail examples that still consume rail actions as booleans, dictionaries, or scores after the move to RailOutcome. The examples now check is_blocked, read evidence from metadata, and apply masking results through is_transform and transform_text.

Also fix the non-README copy-paste failures identified in the PM triage: missing prompt definitions, invalid tutorial YAML, the complete configuration's streaming setting, and chat requests missing model or using the old config_id placement. Replace the API/actions servers' “secure, isolated” claims with accurate process-separation, network-binding, and authentication guidance.

All changes are in docs/; runtime behavior and defaults are unchanged. README and evaluation-command fixes remain in their existing PRs.

Related Issue(s)

• Closes NGR-1089 — sheet row 7: RailOutcome examples.
• Closes NGR-1092 — sheet row 11: server security wording
• Related to NGR-1090 — sheet row 9: README and non-README examples
• PM triage and documentation scope (Slack)

Verification

• Updated the branch to current develop (537fd774298260261bc4ee15e78d861b9e5814e5).
• uv run --locked pre-commit run --files <changed files> passed, including the repository-wide type check.
• Parsed 28 Colang 1.0 examples and the Clavata Colang 2.x action example. Checked 27 blocking conditions against allow/block outcomes, additional threshold cases, and masking transformations.
• Validated the complete configuration, both actions-server configurations, the tutorial prompt, and two chat requests against the current configuration/request schemas.
• Ran the exact documented self-check input/output flows in four offline conversations using the actual actions, TestChat, and FakeLLMModel; safe verdicts continued and blocking verdicts refused.
• make docs-fern passed with 0 errors and 566 warnings. SDK generation used the locally cached Fern parser image to avoid a stalled registry refresh.
• Reviewed the documentation against the current action contracts and server implementation, including the authentication and input-handling guidance. git diff --check passed.
• No live model or provider calls were made. CI results will be reported separately.

AI Assistance

Codex prepared the documentation changes, performed source review, and ran local verification. Human review is pending.

• No AI tools were used.
• AI tools were used; a human reviewed and can explain every change (tool: Codex).

Checklist

• I've read the CONTRIBUTING guidelines.
• This PR links to a triaged issue assigned to me. Tracking uses the Linear row tickets above and the PM/TPM Slack thread.
• My PR title follows the project commit convention.
• I've updated the documentation if applicable.
• I've added tests if applicable. Documentation examples were validated with a temporary offline harness; no runtime or test-suite files changed.
• I've noted verification beyond CI and any checks I couldn't run.
• I did not update generated changelog files manually.
• I addressed all CodeRabbit, Greptile, and other review comments, or replied with why no change is needed.
• @mentions of the person or team responsible for reviewing proposed changes.

[github-actions]

Staged Fern docs preview: https://nvidia-preview-pr-2585.docs.buildwithfern.com/nemo/guardrails