Agentic Fabric Failure-Mode View
Failure Mode View
A dedicated surface proving that degraded AI, agent, data, policy, SAP, and cloud conditions route through controlled fallback instead of uncontrolled automation.
Every path ends with accountability, evidence, and replay.
Autonomous write-back blocked in degraded mode
Human fallback required
Replay stored on every path
Policy gate remains authoritative
Normal Path
Business Event
v
Decision Request
v
Agent Runtime
v
Policy Validation
v
Evidence Collection
v
Action Execution
v
Replay Stored
Failure Path
Business Event
v
Decision Request
v
Agent Runtime
v
Agent Failure
v
Fallback Policy
v
Human Workflow
v
Business Continuity
v
Replay Stored
Failure Scenario Matrix
| Failure | Degraded Mode | Continuity |
|---|---|---|
| LLM service unavailable | Rules engine + human workflow | Decisions continue with deterministic policy checks |
| Agent runtime unavailable | Fallback policy + analyst queue | Work routed to accountable human owner |
| Databricks unavailable | Last known context + freshness warning | Autonomous actions blocked; recommendations allowed if confidence permits |
| SAP interface delayed | Read-only mode + validation retry | No write-back until current state is confirmed |
| Policy changed unexpectedly | Version lock + policy owner review | Historical decisions remain tied to prior policy version |
| Cloud region outage | Failover + recovery workflow | Checkpointed runtime state and replay store recover path |
Safe Degradation Modes
| Safe Degradation Mode | Behavior |
|---|---|
| Mode 1 - Full Autonomous | Agent, policy, evidence, approval, execution |
| Mode 2 - Human Approval Required | Agent recommendation, human approval, execution |
| Mode 3 - Read Only | Insights only; no writes and no actions |
| Mode 4 - Manual Operations | Human process with replay capture and audit capture |
Human Fallback Matrix
| AI Function | Human Fallback |
|---|---|
| Invoice Matching | AP Analyst |
| Journal Validation | Controller |
| Spend Decision | Budget Owner |
| Credit Decision | Credit Manager |
| Close Validation | Finance Lead |
| Policy Interpretation | Risk Owner |
Runtime Controls
| Control | Endpoint | Purpose |
|---|---|---|
| Resilience status | /api/iaf/control-plane/resilience/status | Current continuity posture |
| Kill switch | /api/iaf/control-plane/resilience/kill-switch/activate | Centrally disable unsafe agent action |
| Failure simulation | /api/iaf/control-plane/resilience/simulate/{scenario_id} | Prove fallback path before production incident |
| Integrity verify | /api/iaf/control-plane/resilience/integrity/verify | Verify recovered replay and evidence chain |