Find your next learning step
A queue surrogate earns its place by shortening the route to a sound decision. It does not earn permanent authority. When demand, routing logic, service distributions or decision stakes move beyond the evidence that supported the approximation, the safe response may be to narrow its use, escalate to detailed simulation or retire it. This guide turns that choice into a visible, reviewable process.
Open the chapter contents · Queueing Theory root · Gold Standard of Education · How X Works
Chapter contents
Choose the route closest to the decision, or read from Chapter 1. The chapters remain open; these grouped routes are a navigable index.
Route 1 · Define the decision
Route 2 · Work the ledgers
Route 3 · Transfer the method
Route 4 · Check limits and evidence
Route 5 · Ask and continue
CHAPTER 1 OF 12 · DEFINE THE DECISION
1. Treat retirement as a decision, not deletion
Retirement means the surrogate can no longer support a declared use without disproportionate risk. Its files and results stay in the evidence trail; only its operational authority changes.
A model may be retired for one task and retained for another. A response surface can remain useful for teaching broad utilisation effects while being withdrawn from capacity approval near a service-level boundary.
CHAPTER 2 OF 12 · DEFINE THE DECISION
2. Define the fidelity ladder before trouble appears
Write the available levels in advance: closed-form approximation, fitted surrogate, coarse discrete-event simulation and full event model. Each level needs a cost, turnaround time and decision scope.
Escalation should follow evidence, not embarrassment. Typical triggers include extrapolation, rank instability, residual growth, changed routing rules, new tail penalties and a narrow gap between competing actions.
CHAPTER 3 OF 12 · WORK THE LEDGERS
3. Build a model-risk register
A contact centre records the assumptions that can invalidate its staffing surrogate.
| Risk item | Current evidence | Trigger | Response | Owner |
|---|---|---|---|---|
| Arrival profile | Weekday data | Weekend share above 18% | Escalate to detailed simulation | Planning |
| Service tail | Lognormal fit | 99th percentile error above 12% | Refit or narrow use | Analytics |
| Routing logic | Single skill pool | Overflow route activated | Retire current surrogate | Operations |
| Action ranking | 19 agents preferred | Regret gap below 0.4 | Independent confirmation | Review lead |
The register separates a known limitation from an unmanaged surprise. It should record evidence dates, model version, decision scope and the last independent challenge.
CHAPTER 4 OF 12 · WORK THE LEDGERS
4. Worked fidelity-escalation decision
A surrogate screens four staffing actions after weekend arrivals increase. The team compares decision separation with estimated model error.
| Action | Surrogate loss | Model-error bound | Decision gap | Fidelity decision |
|---|---|---|---|---|
| 18 agents | 16.1 | ±0.7 | 2.0 | Reject at surrogate level |
| 19 agents | 14.4 | ±0.8 | 0.3 | Escalate |
| 20 agents | 14.1 | ±0.7 | Reference | Escalate |
| 21 agents | 14.8 | ±0.6 | 0.7 | Retain as fallback |
The detailed simulator confirms 20 agents only after modelling overflow routing. That structural change also retires the old surrogate from weekend approval, even though its average error looked modest.
CHAPTER 5 OF 12 · WORK THE LEDGERS
5. Separate recalibration, restriction and retirement
Recalibration changes parameters while keeping the model form and use. Restriction narrows the domain or decision. Retirement ends the declared use because the form, data or governance no longer supports it.
Do not disguise structural failure as frequent recalibration. If every new regime needs a patch, the accumulated evidence may favour a redesigned model and a clean version boundary.
CHAPTER 6 OF 12 · TRANSFER THE METHOD
6. Preserve a safe fallback
A retirement decision needs an operational substitute: detailed simulation, a conservative rule, a previously validated capacity or manual review. Removing a model without a fallback can create a larger risk than continuing it briefly under controls.
State the sunset date, residual users, blocked interfaces and conditions for emergency reuse. Emergency reuse should be time-limited, logged and independently reviewed.
CHAPTER 7 OF 12 · TRANSFER THE METHOD
7. Changed-context transfer: hospital triage queues
Transfer the method to an emergency department where a surrogate estimates waiting-time effects of roster changes. A new streaming protocol changes patient routes, so historical predictive accuracy no longer proves decision validity.
The transfer check asks whether the new protocol changes only arrival rates or the queue network itself. Network change requires fidelity escalation and may require retirement of the old surrogate.
CHAPTER 8 OF 12 · TRANSFER THE METHOD
8. Connect screening, confirmation and recalibration
Start with the surrogate screening and confirmation guide. Use the nested Monte Carlo guide when numerical error clouds a trigger.
Return to the value-of-information guide when choosing whether another detailed experiment is worth its cost.
CHAPTER 9 OF 12 · CHECK LIMITS AND EVIDENCE
9. Limits and common misconceptions
Age alone does not retire a model, and recent deployment does not make one safe. Evidence quality, domain fit, decision consequence and control strength matter more than the calendar.
A register is not a substitute for testing. It is a control surface that makes assumptions, triggers and actions inspectable.
CHAPTER 10 OF 12 · CHECK LIMITS AND EVIDENCE
10. Practical evidence checklist
Archive model identifier, code and data versions, intended uses, prohibited uses, fidelity level, error evidence, trigger history, validation date, open findings and decision owner.
For retirement, save the reason, affected decisions, fallback, access changes, communication record, final outputs and conditions under which a successor may be approved.
CHAPTER 11 OF 12 · ASK AND CONTINUE
11. Frequently asked questions
Can a retired surrogate still be used for exploration?
Yes, if clearly labelled and kept outside operational approval. Exploratory use must not quietly feed a live decision.
Does every trigger require the highest-fidelity model?
No. Escalate only far enough to resolve the material uncertainty, while keeping a conservative fallback.
Who decides retirement?
The accountable decision owner should act with independent technical review and documented operational input.
CHAPTER 12 OF 12 · ASK AND CONTINUE
12. Sources and next reading
The Federal Reserve’s 17 April 2026 revised model-risk guidance describes inventories, validation, monitoring and governance in its regulated context. NIST’s AI RMF Playbook covers ongoing risk management and safe decommissioning governance. The queue ledgers are original teaching examples.
Continue through the Queueing Theory root, the Gold Standard of Education and How X Works.
Top · Contents · Previous · How X Works