docs(eval-author): add work-in-progress design overview

[mstaats-nvidia]

Summary

Work in progress: This draft introduces a design overview for Eval Author so the evolving command surface, strategy extension model, and Insight-driven reference workflow can be shared and reviewed before implementation. The document is intentionally a discussion draft and records open questions rather than asserting a finalized architecture.

Changes

• Define the purpose and mission of Eval Author around evaluation-set measurement and generation.
• Describe the proposed discover, audit, propose, run, and doctor command surfaces.
• Establish deliberately minimal AuditStrategy and ProposalStrategy extension points with Python entry-point registration.
• Illustrate how the existing Insight-driven generation path and a new Insight audit could be adapted behind those strategy boundaries.
• Record an incremental implementation path and unresolved design questions for review.

Type of Change

• Code change (feature, bug fix, or refactor)
• Code change with documentation updates
• Documentation only
• Contributor tooling or automation
• CI, build, or test infrastructure

Quality Gates

• Tests added or updated for changed behavior
• Existing tests cover changed behavior — justification:
• Tests not applicable — justification: This draft adds a design document and changes no executable behavior.
• Documentation updated for user-visible behavior
• Documentation not applicable — justification:

Verification

• Pull request title follows the repository's Conventional Commit format
• Every commit includes an appropriate Signed-off-by: trailer
• uv run pre-commit run -a passes, or any blocked checks are identified below
• Targeted tests pass, or tests are marked not applicable above
• No secrets, API keys, or credentials are included

Targeted validation:

• docker run --rm -v "$PWD:/workdir" -w /workdir ghcr.io/igorshubovych/markdownlint-cli:v0.47.0 --config .markdownlint.json plugins/nemo-eval-author/DESIGN_OVERVIEW.md passed.
• git diff --check origin/main...HEAD passed.
• The DCO audit passed for every commit in origin/main..HEAD.
• uv run pre-commit run -a was run. Ruff, Ruff format, ty, config-reference validation, lock-drift validation, copyright, UI, plugin-import, merge-conflict, and Flox-lock checks passed. The environment-wide Helm and toolchain-consistency hooks are blocked because this host lacks helm-docs and yq, and its uv 0.11.18 does not match the repository-pinned 0.9.14.

[github-actions]

Suite	Lines Covered	Line Rate	Branch Rate
Unit Tests	33113/41824	79.2%	63.9%
Integration Tests	N/A	N/A	N/A