Run five-tester Community release candidate dogfood gate

[rosshd]

Objective

Prove the immutable Openlearn Community release candidate works for fresh users without maintainer intervention or shared credentials.

Authority

Follow manual-tests/public-release.md exactly.
Keep sanitized result records outside learner homes and never capture keys, prompts, course content, transcripts, or learner files.

Required coverage

• Five testers fresh to the candidate.
• At least one macOS, Windows, and Linux environment.
• Clean artifact hash verification and installation.
• Learner-owned hosted provider or local keyless Ollama.
• Technical Interview Prep placement defer and completion paths.
• Broad custom course, Quick Learn, and unavailable optional-runtime recovery.
• Provider rejection or unreachable-endpoint recovery.
• Resume, progress, provider settings, data inventory, and verified backup.
• Under five minutes from first launch to first teaching interaction, excluding external account setup or model downloads.

Acceptance criteria

• Exact release commit, version, artifacts, and SHA256SUMS are recorded.
• All five sanitized tester records pass the core and assigned supplemental journey.
• macOS, Windows, and Linux have human coverage.
• No tester requires maintainer intervention.
• No maintainer credential or credit is shared.
• Every finding is classified as release-blocking, follow-up, or documentation-only.
• Every release blocker is fixed and the affected journey is rerun against a new immutable candidate.
• Quick Learn Quick Learn dogfood pass with real study material #12 has the required real-source evidence.
• The final gate record in manual-tests/public-release.md is complete and the decision is GO.

Out of scope

• New learning tools or curriculum expansion.
• Subscription hosting.
• Recording private learner content.
• Treating this usability gate as learning-outcome calibration.