Release checklist

Every item traces to a finding in the audit report. The ordering is by severity: clear the blockers, then the should-fixes, then the polish. Items are written so they can be checked off literally.

🚀 Slice D — the pre-flip / pre-tag punch list (maintainer executes)¶

Added 2026-07-11 at the close of the pre-public improvement program (Slices A–C). These are the concrete actions between “repo goes public” and “tag v0.1.0”, in order. Every claim below was re-verified against the repo on 2026-07-11.

Before or at the public flip:

• Re-enable the 3 GitHub Actions workflows. All three are disabled_manually in the repo settings (the YAML itself is current): tests.yml (PR gate), physics-validation.yml (slow lane), docs.yml (docs gate). Re-enable in Settings → Actions, then confirm one green run on a real PR before trusting them.

• Fix the stale Python-3.10 matrix leg first. pyproject.toml declares requires-python = ">=3.11", but .github/workflows/physics-validation.yml line 62 still runs a "3.10" matrix leg (and the comment above it, line 54, still says >=3.10). Change the matrix to ["3.11", "3.13"] and fix the comment — otherwise the first CI run after re-enable fails on an unsupported interpreter, not on physics.

• Add CONTRIBUTING.md. Minimum useful content: uv-based setup (uv pip install -e ".[dev]" with the side-by-side jaxstro checkout), the two-tier local gate (FAST -m "not slow" / FULL), the JAX-native ground rules (no numpy/scipy on core paths, differentiability non-negotiable), and that PRs must pass make -C docs/website gate when they touch docs.

At the v0.1.0 tag:

• Add CITATION.cff + mint the Zenodo DOI. File does not exist yet. Enable the GitHub–Zenodo integration before pushing the tag (Zenodo archives on tag-push), then backfill the minted DOI into CITATION.cff and the README. Template details in release strategy.

• Add an sdist include/exclude. The build backend is hatchling and only a [tool.hatch.build.targets.wheel] table exists — the sdist currently bundles ~20 MB of docs/ (built figures included) and internal files. Add a [tool.hatch.build.targets.sdist] table with exclude (at minimum docs/, laboratory/, audits/, papers/, .brain-drafts/, tests/experimental/) or an explicit include allowlist, then verify with uv build --sdist + tar -tzf dist/*.tar.gz | head -50.

• Fix the hardcoded path in scripts/check.sh (line 77). The clean-venv wheel smoke installs /Users/anna/projects/jaxstro-dev/jaxstro verbatim. Replace with a repo-relative resolution, e.g. "$(cd "$(dirname "$0")/../../jaxstro" && pwd)", so the gate script survives on any checkout (CI included).

After the tag — get it on PyPI:

• Publish progenax to PyPI (added to scope 2026-07-11). This is gated on the dependency story: the wheel depends on the sibling jaxstro, which is also unpublished, so jaxstro must reach PyPI first (or the strategy in release strategy — namespace package / vendor / git-URL — must be decided instead). Concrete steps once decided:

  1. Publish jaxstro (its own gate + tag).

  2. Replace the [tool.uv.sources] local-path dependency with a versioned PyPI pin (jaxstro>=X.Y).

  3. uv build → twine check dist/* → upload to TestPyPI and verify pip install progenax in a clean venv resolves end-to-end.

  4. Upload to PyPI (twine upload or a trusted-publisher GitHub Action on the tag). A GitHub source release does NOT wait on this — flip/tag first, PyPI after.

❌ Blockers — must resolve to claim “released”¶

• Resolve the jaxstro dependency story (R2). An outsider cannot pip install progenax while the runtime depends on the unpublished jaxstro. Decision deferred to the maintainer (PyPI-first / namespace package / vendor / git-URL) — see release strategy. Not a blocker for a GitHub source tag; a hard blocker for PyPI.

• Re-enable CI (D8). Three GitHub Actions workflows are disabled_manually. Re-enable tests.yml (PR gate), physics-validation.yml (release-tag slow lane), and docs.yml (the docs gate — 0 broken .md links + 0 MyST content warnings, via make -C docs/website gate; added + validated locally 2026-06-20, including negative tests, but never run on GitHub). Confirm a green run on a real PR before tagging.

⚠️ Should fix before a polished public launch¶

• Fix the stale validate_king.py (D2/T11). (done 2026-06-18) Dropped the removed r_t= kwarg at both call sites; script exits 0 and writes all 5 figures.

• Document MichieProfile (D4). (done 2026-06-18) Full docstrings added to from_W0_rc, __init__, .sample_positions, .density.

• Fill TruncatedIMF.ppf and the IMF mass-ratio pdf/cdf/ppf Returns docstrings (D4). (done 2026-06-18)

• Decide the units-policy split (D4/A2). (done 2026-06-18 — breaking sweep) Chose the explicit-$G$ sweep: the five *VelocityDF.sample_velocities, MultiComponentCluster.sample_cluster, and sample_velocities_pipeline now require explicit $G$ (no G=None default); VelocityDF protocol contract aligned; CHANGELOG breaking entry added.

• Resolve the 40-howto/ stubs (D6). (done 2026-06-18) Authored the 4 progenax-only how-tos (each code block run-verified); interface-with-gravax backlogged out of the TOC until gravax matures.

• Fix the internal-doc links (D6). (done 2026-06-18; re-verified in the docs-hardening pass) Rewrote/removed the ../../plans/ and ../../notes/ links in the real-source pages. The docs-hardening pass added a dedicated link-integrity gate (scripts/check_links_and_counts.py) — MyST silently passes bad .md targets, so this catches what the build does not — and drove the broken-link count to 0 (the 7 broken .md links it first surfaced, including the 6 expecting a 99-bibliography/index.md landing, are all fixed).

• Fetch or downgrade two cited-but-not-held references (D6). (done 2026-06-18 — held + verified) B&M82, Baumgardt&Makino 2003, and Strigari+2007 PDFs ingested; per-paper notes written; project_dispersion/tidal.py citations corrected to the verified primaries (σ_los→B&M82 Eq.7; σ_pmR/σ_pmT→Strigari+2007 Eqs.2/3; jacobi_radius_isothermal→BM03 Eq.1; fill-factor BM03 mis-attribution removed). All formulas verified correct against the PDFs.

• Reconcile the repository URL (D7). (done 2026-06-18) myst.yml + generated API pages now use github.com/jaxstro/progenax.

• Hold the OED demos out of v0.1.0 (D10). (executed in the docs-hardening pass) The OED code is still held for the planned informax package. The docs were trimmed to a single public overview (60-science-demos/optimal-design/index.md), which states plainly that the OED tooling is prototyped and not part of v0.1.0; the five worked-example detail pages were set hidden: true (built and URL-reachable, kept out of the public nav, linked from the unlisted-pages index).

ℹ️ Minor / polish (non-blocking)¶

• Re-verify the scanned-PDF anchors for the methods paper (D6). EFF (1987), Plummer (1911), Michie (1963), Kroupa (2001) are scanned images; verify their load-bearing values visually before publication. (Demircan & Kahraman 1991 was re-verified during this audit.)

• Reconcile the experimental test count. Docs say 322; the suite collects 351 (343 non-slow). Prefer “see CI for the live count” over a pinned number.

• Trim the sdist (~20 MB). It bundles docs/ including built figures; optionally exclude _build/figures from the sdist.

• Scrub the public dev-log. (done in the docs-hardening pass) The 90-development-log/ section was curated and modernised, its absolute home paths and internal-archive pointers were scrubbed, and the whole section was set hidden: true (kept out of the public nav; linked from the unlisted-pages index). The companion absolute-path and internal-pointer leaks in the validation pages were scrubbed in the same pass.

• Optionally auto-execute docs/ code blocks the way the README blocks are, to machine-protect the tutorials.

✅ Verified green — no action needed¶

• Full test gate: 1561 passed / 2 skipped / 0 xfailed, exit 0; coverage 95.98%. (Audit-time was 1553 / 3 skip / 1 xfail / 95.96%; the OM-Plummer oracle closed the skip and the follow-on #4 arc closed the Michie-$W_0$ xfail — C¹ PCHIP back-interp, ADR 0016. The remaining skip is the env-gated strict-refs guard.)

• Physics anchors reproduce (Plummer/King/EFF $Q$, King $c(W_0)$, Engine-B, Kepler).

• Gradient integrity: 97 AD-vs-FD cases, 0 hazards; all 4 registries full.

• JAX-native discipline: no hardcoded $G$, no numpy/scipy on core paths, while_loop correctly fenced.

• myst build: 178 pages, 0 content warnings; no fabricated literature content. (172 at the 2026-06-17 audit; the docs-hardening pass added a bibliography landing, missing per-paper notes, and the unlisted-pages index while consolidating the gravoturbulence subsection.)

• Packaging: wheel clean, twine passes, LICENSE + metadata present, clean-venv import works.

• README examples execute (9 blocks) in CI.

• Nine of ten prior-audit blockers (R1, R3–R10) verified fixed.

📚 v0.1.0 documentation-hardening pass — release provenance¶

A dedicated pass made the docs site correct, honest, and warning-clean for the v0.1.0 public source release. It ran in two phases:

1. A hybrid adversarial audit — a serial reviewer plus parallel reviewers over the theory, validation, and demos sections — to surface every drift, fabricated claim, stale count, and broken link as leads.

2. A section-by-section hardening in which each lead was verified against the actual src/ symbol or the source PDF before any edit, then fixed:

   • Correctness: corrected API snippets (signatures, required G, return shapes, the King r_t differentiability statement) and citations against the primary sources.

   • Test-count drift → dashboard: replaced hardcoded current-tense counts with pointers to the test dashboard / “see CI for the live count”; dated changelog/audit counts were left frozen and annotated.

   • Link integrity: added a link/count gate (scripts/check_links_and_counts.py) and drove broken .md links to 0.

   • Scope & navigation: consolidated the gravoturbulence theory 10 → 5 pages and moved it to hidden: true; trimmed OED to one public overview with five hidden detail pages; curated, modernised, path-scrubbed, and hid the development log; added a 99-bibliography/index.md landing and the missing per-paper notes.

This entry records the pass itself as release provenance; the specific outcomes are checked off above. The genuine remaining blockers (CI re-enable, the jaxstro/PyPI dependency story) are unaffected and still pending.