VALIDATION.md

v0.1.0 validation record

Completed in repository construction

• Current OpenAI Agents SDK release researched (0.23.1, 2026-10-02)
• Native TracingProcessor selected for model evidence
• Native decision-scoped RunHooks selected for actual local tool invocation evidence
• Native human approval flow reviewed (interruptions + RunState.approve/reject)
• Same LoopGrid evidence contract preserved
• Decision-scoped correlation via trace metadata; no global current-decision variable
• Privacy-safe commitments by default
• Background transport worker so tracing callbacks do not perform blocking HTTP
• Explicit transport health surface (flush + assert_healthy)

Technical release gates

• install exact openai-agents==0.23.1
• run unit/semantic tests with real dependencies
• Python compile check
• deterministic ScriptedModel Runner + real function-tool runtime test
• real LoopGrid Core 0.8.1-design-partner auto-allowed E2E
• require evidence_complete
• require applicable coverage 100%
• require verify.valid=true and failures=[]
• real needs_approval=True interruption + RunState.approve resume E2E
• verify reviewer evidence precedes execution evidence
• build wheel/sdist
• twine check dist/*
• inspect wheel contents
• fresh venv install built wheel
• public import test
• GitHub Actions Python matrix
• exact tested commit tagged v0.1.0
• PyPI Trusted Publishing
• fresh public PyPI install test
• website/docs integration page
• OpenAI Agents SDK ecosystem submission only after public validation