[cnicholas]

Release commit for 0.4.0, following CONTRIBUTING.md → Releasing.

• __version__ 0.3.3 → 0.4.0.
• CHANGELOG: [Unreleased] becomes [0.4.0] - 2026-10-14, and the compare links are updated.
• docs/reference/validation.md is regenerated by validation/e2e_bishop_report.py. Only its version line changes: 1,547 pass, 0 fail, 4 with no reference.

Checked on main (e982d32) before this commit

• Full suite: 2,551 collected, 2,544 passed, 7 skipped, 0 failed; coverage 90.95%.
• The release notes' figures, each checked against main and the "before" values against PyPI 0.3.3:
  • PM SDS 2 R2 limits ±1.94 → ±2.76.
  • Medicare sd(R2) 1,023 → 300; the organisation chart flags 8 → 19 of 24.
  • PM INERT SDS 2/3 interaction 0%.
  • ADS 3 recommends Xbar.
  • PM SDS 6 process mean 237.834 → 237.855 (VAS 237.86).
  • PM SDS 3/6 Xbar subgroups 783 → 800 and 642 → 772.
  • PM SDS 3 per-condition centres: four move by 0.013–0.044.
  • The R2-unavailable reason for a one-condition layout.

After merge: tag v0.4.0 on the merge commit and push the tag. That runs publish.yml (test matrix, then PyPI).