Re-freezing a third-state boundary ends one trial and starts a new operating baseline. The candidate rule has passed its shadow and limited-release gates, but it now faces arrival mixtures, staffing cycles and missing-data patterns that the trial could not exhaust. Monitoring must therefore protect the new reference without treating every ordinary fluctuation as evidence that another change is needed.
This guide continues boundary-thaw shadow trials and fallback-capacity reallocation. It asks three practical questions: what should be monitored after re-freeze, how can recurring capacity stress be separated from a boundary mistake, and what evidence is strong enough to open a second-thaw review?
The answer is a linked system of changed-set surveillance, service-demand evidence, recurrence diagnosis and governed action gates.
A practical orientation
Keep the boundary, the arrival stream and the service plan on separate but connected lines. A re-frozen rule can remain correct while capacity fails; capacity can remain stable while the new boundary misclassifies a changed population.
The worked example follows two school-term peaks after re-freeze and shows why a recurring delay signal does not automatically justify another boundary change.
- Define the re-frozen baseline — for owners starting post-release monitoring.
- Diagnose capacity recurrence — for repeated delay or overflow.
- Follow the worked example — for two term-peak cycles.
- Set second-thaw gates — for governance decisions.
- Transfer to a new service window — for different timing and case mix.
Chapter contents
Open a route that matches the decision in front of you, or read straight through. Each route returns to the same evidence chain.
Start with the operating question
Build the evidence chain
Define capacity recurrence precisely
Work the example
Stress-test the decision
1. Declare the new reference and its evidence envelope
Record the re-frozen rule, effective date, eligible population, service plan and evidence that supported release. Keep the previous boundary and trial record accessible. The new reference is not timeless; it is valid inside the stated arrival, information and capacity envelope.
Name the changed set created by the re-freeze. These cases moved relative to the old rule and deserve focused surveillance because any mistaken assumption is likely to appear there first.
- Re-frozen assignment rule and version.
- Changed-set definition and reason codes.
- Expected arrival and service-time ranges.
- Protected outcome and delay gates.
- Owner, backup, rollback and second-thaw authority.
2. Monitor assignment, demand, service and consequence
A reliable stack includes assignment shares, reason-specific fallback entries, service minutes demanded, waiting and completion, and final outcomes. Counts without service time can hide complexity; waiting without arrival denominators can confuse volume with performance.
Compare the whole queue and the changed set. A stable overall outcome can coexist with deterioration among the cases affected by the new boundary.
| Layer | Measure | Question |
|---|---|---|
| Assignment | State migration and reason share | Is the new rule behaving as released? |
| Demand | Expected and observed service minutes | What work is being created? |
| Service | Delay, completion and overflow | Can the route absorb that work? |
| Consequence | Outcome by changed set and protected group | Is the released decision still acceptable? |
3. Define capacity recurrence precisely
Capacity recurrence is a return of a previously controlled service shortfall under comparable conditions. It may appear as repeated overflow, rising upper-tail waiting, unresolved backlog or displacement of another queue. A one-off absence or exceptional surge is not automatically recurrence.
Compare mechanism, timing and scope with the earlier event. If the same time band, reason mix and service bottleneck return after the repair, recurrence becomes plausible. If the case mix or service-time distribution changed, treat it as a new demand problem until evidence connects it.
4. Separate recurrence from boundary failure
Replay the re-frozen assignment rule on comparable observations and examine changed-set outcomes. If assignment and outcomes remain stable while service time or staffing deteriorates, capacity is the first mechanism. If comparable arrivals move unexpectedly or changed-set errors rise, the boundary deserves review.
Do not move the boundary simply to reduce a queue that is under-resourced. That can remove difficult cases from review while exporting risk to the parent decision.
| Pattern | First hypothesis | First action |
|---|---|---|
| Assignment stable; delay repeats in same peak | Capacity recurrence | Restore or redesign sustainable capacity |
| Assignment rises with missing fields | Input integrity | Repair fields; keep boundary under observation |
| Changed-set outcomes deteriorate | Boundary suitability | Open evidence review and protect affected cases |
| Overflow shifts to another service | Displaced capacity | Measure both queues and reverse hidden transfer |
5. Use recurrence triggers with duration and consequence
A recurrence trigger should combine magnitude, persistence and consequence. For example, two relevant peak cycles above the upper-tail delay gate may justify escalation, while one severe protected-outcome event may require immediate restriction.
State the response attached to every trigger: inspect, add temporary protection, restore capacity, restrict scope, roll back or open a second-thaw proposal. Alerts without authority produce observation without control.
6. Worked example: two term peaks after re-freeze
The re-frozen rule sends an expected 260 of 1,600 weekly arrivals to fallback. In the first school-term peak, arrivals rise to 1,760 and fallback entries reach 292; upper-tail waiting breaches the gate for two days. Extra sustainable capacity restores service. In the second comparable peak, arrivals and reason mix are similar, but the breach returns because the added capacity had been reassigned elsewhere.
Frozen-rule replay and changed-set outcomes remain stable. The evidence supports capacity recurrence, not boundary failure. The immediate action is to restore the released service plan and review capacity ownership. A second thaw is not opened.
| Evidence | Peak 1 | Peak 2 | Interpretation |
|---|---|---|---|
| Eligible arrivals | 1,760 | 1,748 | Comparable volume |
| Fallback entries | 292 | 289 | Stable reason-adjusted share |
| Upper-tail delay | Gate breached | Gate breached | Recurring service failure |
| Changed-set outcome | Within released range | Within released range | No boundary failure signal |
| Capacity plan | Temporarily restored | Resource reassigned | Ownership recurrence |
7. Repair sustainable capacity and ownership
Restore the released capacity or replace it with an evidenced equivalent. Name who owns the resource, what competing work can displace it and which approval is required. A capacity promise that can disappear without a decision is not a control.
Test the repair through the same peak mechanism, not an easy low-load day. Measure effects on any donating queue so the repair does not simply move delay.
8. Set second-thaw gates
A second thaw should open only when evidence suggests the re-frozen boundary itself no longer fits: sustained reason-adjusted assignment change, changed-set outcome deterioration, a new population outside scope, a mechanism change, or repeated disagreement with the authorised parent decision.
Capacity recurrence alone is not a second-thaw gate unless the service constraint is an unavoidable property of the boundary design and cannot be met sustainably. The proposal must name what would change and how shadow evidence will test it.
- Boundary-related signal persists after operational repair.
- Changed-set consequence is material.
- New channel or population lies outside the release envelope.
- Candidate change has a causal rationale.
- Shadow, protection and rollback plan are approved.
9. Preserve stability between governed changes
After a repair or second-thaw decision, allow relevant cycles to accumulate before further tuning. Rapid micro-adjustments destroy the reference needed to attribute outcomes. Keep versioned rules and effective dates.
Use a change ledger that links signal, diagnosis, decision, implementation and evidence. It lets future owners distinguish recurrence from a new problem.
10. Read statistical signals in operational context
CUSUM and related charts can reveal sustained small shifts in waiting or assignment. Their reference, sampling unit and response plan must be fixed. A signal indicates departure from the statistical reference, not its cause.
Pair the signal with reason codes, service minutes, staffing, data completeness and outcomes. The same chart movement can arise from very different mechanisms.
11. Misconceptions and practical questions
Common errors include declaring every repeated delay a boundary failure, counting temporary overtime as restored capacity, monitoring total averages without the changed set, and opening a second thaw before the first repair is tested.
Does recurring fallback delay prove the re-frozen boundary is wrong?
No. Diagnose assignment, reason mix, service demand and capacity. A correct boundary can produce delay when the released service plan is not sustained.
Can a second thaw be opened while capacity is being repaired?
Only if independent evidence points to boundary failure or unacceptable consequence. Otherwise repair and test the operational mechanism first so the shadow trial is not contaminated.
12. Changed-context transfer, evidence check and sources
Transfer the method to a new evening service window. Build a window-specific baseline because arrivals may be more concentrated and service times different. Retain the same diagnostic separation: assignment, demand, service and consequence.
NIST describes organised continuous monitoring in SP 800-137. The NIST engineering handbook explains CUSUM charts for detecting sustained shifts. These sources support the monitoring discipline; local second-thaw gates require local queue and outcome evidence.
Continue through the Gold Standard ecosystem
Return to the Queueing Theory subject root, the Gold Standard of Education parent, or the How X Works hub.
Immediate predecessor: boundary-thaw shadow trials and re-freeze gates · Earlier guide: release surveillance and first boundary-thaw triggers · Queueing Theory root
These routes are grouped by learning purpose: prerequisite model, immediate predecessor, and the wider transfer framework. They are not substitutes for the evidence and limits stated in this guide.