#6 E2E + Reliability

[diptimoyghosh9593-pixel]

Milestone. The 10 golden scenarios + reliability proofs + multi-day demo.

Scope (needs #2-#5 in place):

• 10 golden scenarios as parametrized E2E tests: stalled enterprise opportunity, blocked product work, stale SOP, undocumented decision, ownership gap, KPI deterioration, cross-system contradiction, human redirect, failed tool execution, infrastructure restart. Each defines: input events -> expected evidence -> context -> mission -> employee -> allowed capabilities -> expected tool calls -> human gate -> outcome.
• Reliability proofs: duplicate-event idempotency (re-deliver same event = no duplicate work); missions survive restart (replay); continue-as-new preserves state; tool retries do not duplicate actions; unavailable connector degrades gracefully; malformed LLM output rejected (Pydantic); tenant isolation (data cannot cross boundaries).
• Multi-day simulation run (mockoon :3001-:3004 + Temporal + real LLM where available) -> auditable timeline as north-star artefact.

Final acceptance statement (single criterion): A simulated B2B SaaS runs for multiple days while OntologyAI continuously ingests events from 4 connected systems, updates organizational knowledge, wakes appropriate missions, lets its 3 AI employees investigate and execute bounded work using sequential/parallel/branching behavior, handles human interruption and replanning, survives retries/restarts/replay, records an auditable timeline, verifies outcomes, and continuously maintains the founder operational workspace. If it works, STOP.

Tests (TDD): parametrized golden-scenario suite + reliability suite.