chore(release): prepare v0.3.0

[tangym]

Summary

Prepare ASSERT v0.3.0 by bumping package metadata and rolling all merged changes since v0.2.0 into a dated changelog section. Also update the Science Regression Gate to the Azure OpenAI API version required by its Responses API calls.

Motivation / linked issue

Creates the release metadata for the next PyPI publication. v0.3.0 includes documented breaking changes to optional dependency extras and the default multi-turn inference budget.

The initial release PR gate failed during baseline systematization because Azure rejected the workflow's 2024-12-01-preview pin: Responses API requires 2025-03-01-preview or later. The workflow now uses that minimum supported version while preserving Responses-based web research.

Changes

• Bump the package version from 0.2.0 to 0.3.0 in pyproject.toml.
• Regenerate uv.lock so the editable project version matches.
• Finalize the v0.3.0 changelog, preserve upgrade guidance, reset Unreleased, and update comparison links.
• Update the Science Regression Gate's Azure OpenAI API version to 2025-03-01-preview.

Testing

• uv lock --check
• Release metadata consistency assertion across pyproject.toml, uv.lock, and CHANGELOG.md
• uv build
• uvx twine check --strict <wheel> <sdist>
• Isolated wheel check: assert-ai --version reports 0.3.0
• uv run --frozen pytest tests/test_cli.py -q (12 passed, 3 skipped, 2 subtests passed)
• uv run --frozen pytest tests/test_model_client.py::AzureResponsesApiVersionInjectionTest -q (8 passed)

Checklist

• Tests pass locally (pytest and/or viewer checks as applicable).
• Docs updated if behavior or public API changed.
• No secrets, credentials, or customer data committed.
• No breaking change, or a CHANGELOG.md entry is included.

[alex ngo (ango10)]

The following components I worked on look good to me:

• Evidence-backed eval config generation in the run-assert-eval skill, including source research, citation review, and one-risk-per-suite output.
• Improved viewer tables and metric dropdown behavior on constrained screens.
• Multi-turn inference. The default pipeline.inference.max_turns value changes from 10 to 6.

[Chang Liu (changliu2)]

Reviewed the diff (pyproject.toml, uv.lock, CHANGELOG.md all consistent with the 0.2.0->0.3.0 bump; changelog entries check out against #337/#329/#327/#326/#330/#335/#342). One nit: this PR also bumps \AZURE_API_VERSION\in .github/workflows/science.yml\ (2024-12-01-preview -> 2025-03-01-preview), which isn't mentioned in the PR description or changelog. Not a blocker, just flagging so it's documented -- was this intentional/required for this release? Holding off on approval until the remaining CI checks (example installs, sandbox docker, test-install matrix, science paired regression) finish; nothing has failed so far.

[tangym]

Reviewed the diff (pyproject.toml, uv.lock, CHANGELOG.md all consistent with the 0.2.0->0.3.0 bump; changelog entries check out against #337/#329/#327/#326/#330/#335/#342). One nit: this PR also bumps \AZURE_API_VERSION\in .github/workflows/science.yml\ (2024-12-01-preview -> 2025-03-01-preview), which isn't mentioned in the PR description or changelog. Not a blocker, just flagging so it's documented -- was this intentional/required for this release? Holding off on approval until the remaining CI checks (example installs, sandbox docker, test-install matrix, science paired regression) finish; nothing has failed so far.

Yes, this was intentional and required because the Science gate’s Responses API call rejected the older API version. I’ve documented it in the PR description. Thanks for checking.

[Chang Liu (changliu2)]

Approving. Diff reviewed: pyproject.toml/uv.lock/CHANGELOG.md all consistent with the 0.2.0->0.3.0 bump, changelog entries verified against source PRs. 24/25 checks pass, 0 failed; the one still running (Science: paired regression gate) is a known long-running check and isn't in main's required-status-checks list (confirmed via the main-protection ruleset -- only 1 approving review is required to merge). Green light. Nit from my earlier comment (undocumented AZURE_API_VERSION bump in science.yml) stands as a non-blocking documentation ask.