release: 0.3.0 — beta, with five verifier-accuracy fixes

[silversurfer562]

Promotes attune-verify to Development Status :: 4 - Beta and states an API compatibility promise for the deterministic core.

Why not just flip the classifier

A review pass over the library found five defects, all in the extractor, all in the two classes that actually matter for a fact-checker. Shipping "beta" over a checker that silently skips a whole shape of code block would have been the wrong claim to make.

Silent false negatives — a hallucination passing unchecked

• Indented fences were invisible. A code block nested under a list item — how LLMs routinely write install steps — matched nothing, so every import and shell flag inside one went unchecked. The extractor is now a line scanner that accepts a fence at any indentation and strips that indent from the body; uniformly indented code otherwise fails ast.parse, which was the silent skip. This was the largest remaining hole.
• Tilde fences (~~~) were unrecognized. CommonMark-legal, never extracted.

False positives — a real entity flagged

• Percent-encoded link targets were checked literally, so a link to a file whose name contains a space (docs/my%20file.md) was flagged though the file exists. The raw form is still tried first, so a file genuinely named a%20b.md resolves, and a decoded target that escapes project_root is still caught.
• Balanced parens in a link target were truncated at the first ) — docs/a(1).md was checked as docs/a(1.
• Link syntax shown as an example inside a code span or fence was checked as a live link. No renderer resolves those. Found by running verify over its own README.

The scanner also enforces what the old regex ignored: an unclosed fence is not a fence (its body was the rest of the document), a closing run must match the opening character and length, and a line-leading inline span no longer opens one.

Beta hygiene

• py.typed — the package is fully annotated but shipped no PEP 561 marker, so type checkers ignored it downstream. Verified present in the built wheel.
• Version drift guard — the release flow bumps pyproject.toml and __version__ by hand; a test now pins them equal.
• README — a per-checker table of what each checker settles and which truth source it needs, plus an explicit known-limitations list (reference-style links, short flags, bare-flag attribution, count keyword matching).

Verification

Gate	Result
Tests	130 passed (was 99)
Coverage	96%, _extract.py 100%
Corpus precision/recall	passing, 9 new regression cases
Mutation score	80.7% against the 75% gate
Lint	ruff + black clean
Wheel	builds, Development Status :: 4 - Beta, ships py.typed

Every fix carries a regression corpus case, per this repo's convention.

Not published — this stops at the PR.

🤖 Generated with Claude Code