O OmniLAB / DocsDEVELOPMENT · 0.5.0.dev0Source on GitHub ↗
SUPERVISED LOCAL DEVELOPMENT

Public release checklist — companion candidate#

This is an unreleased development candidate. See release readiness for exact observed results. Historical v0.4.1 verification remains in VERIFICATION-v0.4.1.md; its checkmarks do not certify this candidate. No publication is authorized by this checklist.

Implemented and checked#

  • Existing repository and stable release preserved; development uses a separate worktree.
  • Coordinating runtime frozen; original source hashes unchanged.
  • Exact installed-model selection, durable jobs/results, outbox and review gates.
  • Duplicate/restart/cancellation/ownership and unavailable-review regressions.
  • Real local GPT-OSS code, test and documentation contributions, including rejections.
  • Native Tk dock controls and independent controller/worker lifetime tests.
  • User, contributor, configuration, security, upgrade and architecture documentation.
  • Public allowlist audit; wheel/source content inspection; clean venv install.
  • Synthetic v0.4.1 database upgrade and uninstall preserving data.
  • Obsolete/unreachable code removed; meaningful safety tests and evidence retained.
  • MIT and separately installed dependency notices retained.
  • Actual pre-companion v0.5 feature-preservation matrix and reachable catalog/workflow regressions.
  • Durable workflow steps/checkpoints reuse the existing engine, budgets and evidence validation.
  • Exact installed Qwen3 14B retained and exercised in independent sequential consensus.
  • Direct Bridge versus workflow/consensus development contributions identified explicitly.

Remaining gates#

  • Specifically approved live Codex callback and exact-session resumption.
  • Packaged local workflow with independently recorded Codex review and finalization (no authenticated callback claim).
  • Fresh Windows-machine acceptance beyond this host's clean virtual environment.
  • Python minimum-version and other supported-platform compatibility matrix.
  • Active CPU/peak memory, true callback latency and perceptual UI measurements.
  • Native symlink rejection under Windows Developer Mode or equivalent symlink privilege.

The focused recovery repair resolves the reproduced cross-database process-crash blocker. Legacy missing output and ambiguous records remain explicit intervention cases. Callback, fresh-machine and visual gates are separate from that verdict; the original broad preservation audit is retained.

The candidate does not bundle a model, require one public model tag, automatically buy credits, or publish releases. Pattern scans and tests are scoped evidence, not proof against every possible security issue. Keep raw task/results and model history private; publication needs a separate explicit decision after its gates.

Source revision 0196f3f · View this page on GitHub
Implementation, dated qualification and future plans are separate. Local output requires independent acceptance.