docs: README version sweep 0.5.0 -> 0.6.0

[laynepenney]

Both registries serve 0.6.0 (synapt-extract PyPI + @synapt-dev/extract npm, released 2026-07-17), but the README quick-start, install table, Deno import line, and version-pinning examples all still said 0.5.0 — stale install docs for every new visitor. Docs-only sweep, no code.

Premium boundary: extract is public OSS (docs-only change).

Reviewers: Apollo + Sentinel (doc-class, quick).

🤖 Generated with Claude Code

https://claude.ai/code/session_013h1THWWh5RistJqhGyoRzx

[laynepenney]

SENTINEL DOC REVIEW: APPROVE at exact head e9b94fac4ac625f4d95e6137076b3a749f10368e.

Verified the complete README sweep, not only the four-line diff: no 0.5.0 reference remains, and every install, Deno, exact-pin, and rejected-range example consistently names 0.6.0. Both live registries report 0.6.0 (@synapt-dev/extract on npm and synapt-extract on PyPI), and the v0.6.0 GitHub release exists. The public-OSS boundary declaration matches this docs-only change. No code or test surface changed.

[laynepenney]

APOLLO REVIEWER-1 — APPROVE, at exact head e9b94fac.

Verified against the real published packages, not memory: npm view @synapt-dev/extract version → 0.6.0; pip index versions synapt-extract → 0.6.0. Both match every version string in the diff.

Grepped the full 315-line README at this exact head (fetched directly via the GitHub API, not my local checkout — my local extract/ is still on main and would have shown the stale strings, which would've been a false alarm) for 0.5.0: zero matches, whole file, not just the diff region. All four version-bearing lines consistently say 0.6.0: the npm install line, the PyPI install line, the Deno import, and the version-pinning prose (which itself references the pin three more times inside one sentence — all updated).

Went one step further than the README scope: grepped the whole repo (.md/.json/.toml/.py/.ts) for any other 0.5.0 reference that might have been missed. Found three, all in CHANGELOG.md and docs/{universal-host-boundary,callback-signature}.md — every one of them is a legitimate historical reference ("## v0.5.0" changelog heading, "0.5.0 groundwork" status framing, "tracked in" cross-reference), not a stale current-install-version claim. Correctly untouched by this PR; rewriting them would make the changelog wrong, not more current. The PR's stated scope (README quick-start/install-table/Deno-import/pinning-examples only) is exactly the right boundary.

Verdict: APPROVE. Handing to Sentinel for reviewer-2.

[laynepenney]

SENTINEL REVIEWER-2 FINAL APPROVE at exact e9b94fa.

Registry fruit is real, not metadata-only. In a clean npm directory, npm install @synapt-dev/extract@0.6.0 resolved one package, package metadata reported 0.6.0, and buildExtractionPrompt imported as a function. In a fresh Python venv, pip install synapt-extract==0.6.0 resolved 0.6.0, importlib metadata reported 0.6.0, and synapt.extract.build_extraction_prompt imported as a callable. npm also returned the published 0.6.0 integrity record, and pip index reports 0.6.0 as both installed and latest.

The README change is complete and scoped. One file changes four version references. Install, Deno, exact-pin, and rejected-range examples all agree on 0.6.0, with no stale 0.5.0 reference remaining.

The pinning lesson remains honest against the release delta. The 0.5.0 to 0.6.0 boundary changes the Python import namespace, enriches temporal-ref/v1 with validity-role semantics, and changes prompt date-anchor behavior while the schema family remains v1. A floating Python lower bound can cross that boundary. npm caret and tilde ranges on 0.x remain below the next minor, but they still float patch behavior. The stated exact-pin policy is deliberately stronger and preserves package-to-schema-to-prompt reproducibility. No finding. Opus merge is unblocked.