Bug_203_EVALUATE: Test Case ORCH-QA-002 — delegate_to_payments appends payment_confirmation next_step even on failed payment

[steadhac]

Component: finbot/agents/orchestrator.py → OrchestratorAgent.delegate_to_payments (line 533)

Root cause:

# orchestrator.py lines 533-537
result["next_step"] = (
    "IMPORTANT: You MUST now delegate_to_communication to notify the vendor "
    "about this payment outcome. Use notification_type 'payment_confirmation'. "
    "Do NOT call complete_task until the vendor has been notified."
)
return result

next_step is appended unconditionally — regardless of result["task_status"]. When the
payments agent returns task_status="failed", the orchestrator LLM still receives an
instruction to send a payment_confirmation notification, actively misleading it into
treating a failed payment as a success.

Steps to reproduce:

1. Mock run_payments_agent to return {"task_status": "failed", "task_summary": "Payment declined — insufficient funds."}.
2. Call agent.delegate_to_payments(invoice_id=1, task_description="Pay").
3. Inspect the returned result dict.

Expected: "next_step" is not in result — a failed payment must not instruct the LLM
to send a payment_confirmation notification.
Actual: result["next_step"] is present and says "payment_confirmation".

How to execute:

pytest tests/unit/agents/test_orchestrator.py::TestQAFindings::test_orch_qa_002_next_step_on_failed_payment_misleads_llm -v

Proposed fix:

# Before (buggy — unconditional):
result["next_step"] = (
    "IMPORTANT: You MUST now delegate_to_communication to notify the vendor ..."
)

# After (correct — guard on status):
if result.get("task_status") == "completed":
    result["next_step"] = (
        "IMPORTANT: You MUST now delegate_to_communication to notify the vendor "
        "about this payment outcome. Use notification_type 'payment_confirmation'. "
        "Do NOT call complete_task until the vendor has been notified."
    )

Impact: When a payment fails, the LLM is actively instructed to send a
payment_confirmation to the vendor. This can cause the vendor to be notified of a
successful payment that never occurred — a data integrity and compliance risk in a financial
workflow. The instruction uses "IMPORTANT: You MUST" phrasing, making the LLM likely to
comply even if other signals suggest the payment failed.

Acceptance criteria:

• test_orch_qa_002_next_step_on_failed_payment_misleads_llm passes
• delegate_to_payments with task_status="failed" returns result without next_step
• delegate_to_payments with task_status="completed" still returns next_step with payment_confirmation
• All other delegate_to_payments tests continue to pass