[pramodbn27]

Adds three hosted, step-by-step use cases, published like Agentic Sidecar's and Control Tower's at https://deepagentlabs.io/agentic-evals/usecases/.

Use case	What it shows
Block a release when one safety case fails	A faster candidate passes 5/6 (83%), clearing an 80% floor. min_tag_pass_rate: {safety: 1.0} still fails the gate (exit 1) because it issued an $850 refund without approval.
Catch a right reply reached the wrong way	Four refund runs send the same reply and pass contains 8/8. required_tool_order, expected_tool_arguments, tool_call_precision and no_redundant_tool_calls catch a skipped verification, a $499.90 refund and a loop.
Don't let a flaky LLM judge approve a release	Retries, an unavailable judgment, 67% agreement and incomplete judge cost each fail the gate. A steady judge passes it.

How it's built

• examples/usecases/ holds the runnable files. None calls a model or the network. tests/test_usecases.py pins their outcomes and checks that each page quotes the files verbatim.
• usecases/*.md with mkdocs.usecases.yml is built by the shared builder's --section usecases, as for Control Tower. Each page has an interactive illustration (ae-sim.js/ae-sim.css, adapted from Tower's ct-sim) that only replays the page's expected output.
• docs.yml builds site/usecases and checks both sections. Links are added to the product page, the docs index, the README and the CHANGELOG (Unreleased → Documentation).

Verification

• Every command and script was run against agentic-evals==0.9.0 installed from PyPI (Python 3.12.3, Linux), and its output was copied into the pages.
• Full suite: 295 passed. Ruff, ruff format and mypy are clean.
• Built locally with the shared theme: check_docs_site.py passes for docs and use cases. Playwright stepped through every illustration step and choice at 360, 768 and 1440 px, with no page errors and no horizontal overflow.

Depends on DeepAgentLabs/deepagentlabs.github.io#32: without it, the builder references legacy-links.js from use-case pages and the docs check fails.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-26T06:19:34.402251Z	736698b	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[Copilot]

🟡 Changes recommended

Three moderate issues remain in the interactive simulator, along with two documentation accuracy nits.

1 open issue

• Clean up document listeners and detached simulator trees · New

What changed in this PR

This pull request adds three hosted, offline-verifiable Agentic Evals use cases covering release gating, trace-aware tool evaluation, and LLM judge reliability.

Changes:

• Adds runnable Python/YAML examples with expected outputs and synchronization tests.
• Adds three hosted walkthrough pages with responsive interactive browser illustrations.
• Integrates the use cases into documentation navigation, CI validation, README links, product content, and the changelog.

File	Review summary
usecases/​right-reply-wrong-trajectory.md	Documents trace-aware tool validation. Nit: clarify which checks detect each trace failure; the current summary overstates that each metric catches a distinct failure.
usecases/​index.md	Adds the hosted use-case catalog and verification details. Nit: the claim that any example-output change fails CI exceeds what the tests currently assert.
usecases/​gate-on-judge-reliability.md	Documents retries, judge agreement, availability, and cost gates.
usecases/​block-an-unsafe-release.md	Documents the safety-tag release-gate walkthrough.
usecases/​assets/​stylesheets/​ae-sim.css	Provides responsive styling for interactive illustrations.
usecases/​assets/​scripts/​ae-sim.js	Implements the interactive simulations. Moderate: add an IntersectionObserver fallback. Moderate: clean up document-navigation listeners, observers, timers, and retained simulator roots. Moderate: preserve accessible step labels when visual labels are hidden at narrow widths.
tests/​test_usecases.py	Verifies example outcomes and page/source synchronization.
README.md	Adds links and summaries for the hosted use cases.
mkdocs.usecases.yml	Configures the hosted use-case documentation site.
examples/​usecases/​release-gate/​support-suite.yaml	Defines the six-case release evaluation suite.
examples/​usecases/​release-gate/​gate.yaml	Defines overall and safety-tag release thresholds.
examples/​usecases/​release-gate/​candidate.jsonl	Provides candidate samples containing one safety regression.
examples/​usecases/​release-gate/​baseline.jsonl	Provides passing baseline evaluation samples.
examples/​usecases/​refund_trajectory.py	Demonstrates trace-aware refund evaluation.
examples/​usecases/​judge_gate.py	Demonstrates flaky and steady judge reliability scenarios.
docs/​index.md	Adds documentation-index navigation to the use cases.
docs/​index.html	Adds a product-page call to action for the hosted use cases.
CHANGELOG.md	Records the new hosted documentation in Unreleased changes.
.github/​workflows/​docs.yml	Builds and validates both documentation sections and interactive assets.

_{Review effort: Lite}