TownReporter 0.6.32 beta — evidence review polish
Pre-releaseKnown beta issue found in live-model evaluation: named Codex providers can refuse to start from an unpacked, non-Git installation. The adapter in this release omits Codex's documented non-Git execution option, and the Windows installation checks did not exercise a live writing account. This compatibility issue is corrected in v0.6.33; this dated v0.6.32 release remains unchanged.
TownReporter 0.6.32 beta packages the completed evidence-review polish. One available captured version appears once while retaining its artifact and capture-event identities; singular claim counts read correctly; manually selected records are clearly distinguished from draft citations; release headings use a single working link.
README, manuals, landing page, TODO and the development handoff are reconciled. The repository includes the Halo-local update prompt. The operator resolves release identity from this release, stages against a verified copy of Halo's data, and promotes only the accepted revision. This GitHub release does not deploy Halo or enable additional automatic publication.
This remains a beta. Stable Windows acquisition remains v0.6.30. The 60-minute editorial target, useful daily coverage and investigative accuracy are distinct acceptance questions; passing automated installation checks does not prove them.
Exact tag/package source: 592bfb16ce8104c67b0a9874c653751b15968105 (PR38 merge). All 16 application/package checks passed on this revision, plus GitHub Pages deployment. Application CI: 1,953 tests total, 1,899 passed, zero failed, 54 explicitly skipped. Windows packaged installation passed using the archive below. The separate real-PostgreSQL job passed.
The 54 aggregate skips are 44 opt-in PostgreSQL tests, six Windows/PowerShell tests, two build-dependent tests, one authenticated repository-setting check, and one opt-in live-model evaluation. Separate PostgreSQL, Windows and built-browser jobs cover their applicable paths; the repository-setting check and live-model evaluation were not run in CI.
Windows ZIP: TownReporter-0.6.32-windows-x64.zip, 7,112,424 bytes.
SHA-256: 56534d4fdfa50fd52788e3fc523e303a8b8c90fdf33c11b00746e1fab5d84339.
The JSON manifest binds this archive to the exact tag source. The archive was downloaded from the successful exact-source Windows workflow and independently hash-checked before publication.
Published assets were downloaded again and all three matched the tested package byte for byte. Published Windows ZIP verification passed: job 102089932990 took 13m04s and covered installation, browser workflow, stale-build refusal, restart persistence, ownership/ACL/lifecycle checks and a clean stop. This automated journey does not call a live writing model.
Historical replay remains incomplete: September 3–6 source packets were unavailable; one genuine September 7 Library capture was recovered, but the attempted replay worker did not start; September 8 was not run. See the attached historical-replay.md receipt. No six-day replay pass, useful-output result or editor-time result is claimed. Investigation evaluation is reported separately.