A sixth thaw should answer the unsupported question found by fifth-re-freeze monitoring, not reopen the entire queue. In the mobile example, Q5 remained stable where its original evidence applied. The new problem was a late-field state created by an operating-system update, plus the specialist workload that state donated.
This guide turns that diagnosis into a bounded experiment. It preserves Q5 for supported arrivals, keeps Q0 as rollback, registers challengers that represent different late-field mechanisms, replays them on decision-time evidence and admits a new rule only when assignment, capacity and protected outcomes transfer together.
The Gold Standard is a thaw with a clear stopping rule. A useful challenger must explain what changes, remain executable with information available at the decision, and earn fresh confirmation after any tuning.
A practical orientation
Keep the operating state, causal claim, challenger, replay evidence, service consequence and governance decision separate. A candidate can fit the records yet fail because its holding time creates an unacceptable queue elsewhere.
The worked example compares immediate fallback, bounded waiting, predicted-arrival routing and an invalid later-field rule. It then combines a safe information window with a capacity guard, but treats that combination as a new candidate requiring later confirmation.
- Bound the sixth thaw — for owners protecting the supported fifth re-freeze.
- Design new-state challengers — for analysts expressing competing mechanisms.
- Follow the late-field replay — for a full numerical example.
- Apply re-freeze gates — for safe release and rollback.
- Transfer the design — for another channel or information delay.
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
Build challengers from distinct late-field mechanisms
Work the example
Stress-test the decision
1. Translate the sixth-thaw gate into one operating question
Write the question in full: for updated mobile devices whose required field arrives after the decision, can a bounded delay or alternative parent route reduce unsupported assignments without breaching specialist wait limits? Exclude complete-field arrivals and the original missingness states where Q5 remains supported.
Freeze Q5, Q0, the field-delivery definition, capacity envelope and protected outcomes. A sixth thaw changes only the declared late-field boundary unless later evidence proves that boundary wrong.
- Affected devices, channel and decision-time field state.
- Mechanisms still plausible and alternatives already checked.
- Rule components allowed to change.
- Capacity guard, protected outcomes and rollback.
- Screening, confirmation and decision authority.
2. Preserve the fifth-re-freeze reference and lineage
Keep Q5 executable for supported strata and store its output for every thaw record. Preserve Q1 and Q4 component outputs where possible so a new challenger can be compared with the mechanisms that built Q5, not only with its final label.
Lineage must show which candidate came from which rule, what information it uses, and why it is eligible. Do not rename a tuned Q5 as the old rule or revive a leakage candidate under a cleaner title.
3. Build challengers from distinct late-field mechanisms
Register a small set. Q6 can route the new state immediately to Q0. Q7 can hold only when the field is expected inside a fixed window. Q8 can predict whether the field will arrive from decision-time metadata. Q9 can use the eventual field and is included only as a diagnostic upper bound because it is not executable at the decision.
Each candidate needs an expected changed set and service translation. A holding rule consumes time and queue space; a fallback rule donates work; a prediction rule creates error costs. Those consequences belong in the candidate definition.
| Candidate | Mechanism | Eligibility |
|---|---|---|
| Q6 immediate fallback | Route late-field state to Q0 | Executable reference |
| Q7 bounded hold | Wait only inside declared window | Eligible if service guard passes |
| Q8 arrival prediction | Use decision-time metadata | Eligible with calibration and audit |
| Q9 eventual-field oracle | Use information that arrives later | Diagnostic only; not deployable |
4. Reconstruct decision-time records and common clocks
For every arrival, preserve the device state, field snapshot, predicted delivery time, candidate decision, arrival time, service start, completion and mature outcome window. Run every eligible challenger on the same records and order.
Do not let Q7 see actual arrival before deciding to wait, or let Q8 train on the confirmation window. A fair replay preserves what was knowable and separates screening, tuning and confirmation evidence.
5. Translate decisions into queues and capacity
Count donated and received jobs by service class, then translate them into minutes with the same service-time assumptions. Simulate ordinary and declared stress arrival patterns. Report upper-tail waiting and abandonments as well as averages.
A rule that improves assignment can still fail if it synchronises releases from the holding lane and creates a burst. Model the release process, not only the number held.
6. Worked example: a late-field challenger replay
On 1,000 updated-device arrivals, 180 enter the new late-field state. Q6 sends all 180 to Q0. Q7 holds 120 predicted to receive the field within eight minutes; 96 receive it in time, 24 time out, and the release pattern adds a short burst. Q8 predicts 110 timely arrivals, with 88 correct and 22 false holds. Q9 shows that 128 fields eventually arrive but remains ineligible.
Q7 removes more unsupported assignments than Q8, but its unrestricted release would push the specialist 95th-percentile wait beyond the declared limit. Adding Q4’s burst guard creates Q10, a new combined candidate. Q10 is screened now and reserved for later-cycle confirmation rather than declared successful on the same 1,000 records.
| Candidate | Assignment evidence | Capacity result | Decision |
|---|---|---|---|
| Q6 fallback | Stable baseline | Inside envelope | Retain |
| Q7 bounded hold | 96 timely fields | Tail breach without guard | Modify |
| Q8 prediction | 88 correct, 22 false holds | Inside envelope | Keep as comparator |
| Q9 oracle | 128 eventual fields | Not executable | Exclude |
| Q10 Q7 plus guard | Best screened trade-off | Inside screened envelope | Confirm later |
7. Use shadow and limited release to test replay limits
Run Q10 in shadow on current arrivals with reason codes, predicted arrival windows and capacity-guard activations. Then release to a bounded device cohort while Q5 remains active elsewhere and Q0 remains executable.
Observe staff behaviour, upstream retries and actual holding-lane releases. Replay cannot recreate those feedback effects. Exercise rollback before expansion and measure how quickly held cases return to a safe route.
8. Set sixth re-freeze gates
Re-freeze only when Q10’s late-field mechanism is supported on fresh cycles, the changed set matches the declared state, specialist tails remain inside ordinary and stress limits, protected outcomes do not deteriorate, and reason-code integrity supports reconstruction.
Name the frozen scope. The rule may be approved for one device family and field-delivery window while other devices remain on Q5 or Q0. Partial re-freeze is more honest than pretending a local result is universal.
- Fresh mechanism-linked confirmation.
- Explainable changed set and information timing.
- Ordinary and stress capacity pass.
- Protected outcomes and maturity window pass.
- Rollback exercised with current infrastructure.
- Unsupported scope and next monitoring triggers recorded.
9. Re-freeze without erasing challenger evidence
Store Q10’s implementation, window, predictor version, capacity guard and monitoring plan. Preserve Q6, Q7 and Q8 results because they explain why the combination was chosen and provide future diagnostics.
Do not delete Q9. Mark it ineligible and retain its upper-bound role. Future readers need to see that the apparent best assignment score depended on unavailable information.
10. Know what the sixth-thaw evidence cannot prove
The replay is conditional on recorded arrivals, candidate set, service-time model and no unmeasured interference. Limited release reduces some uncertainty but still may not cover rare bursts, seasonal mix or future device behaviour.
Selecting Q10 among declared candidates does not prove global optimality. The defensible conclusion is that it met the stated gates for the tested late-field scope while protected references remained available.
11. Misconceptions, practical questions and transfer
Common mistakes include expanding the thaw to supported arrivals, treating the oracle as deployable, tuning and confirming on one window, and checking only assignment scores. The repair is bounded scope, decision-time eligibility and joined service evidence.
Can Q7 be selected if it has the best assignment result but breaches the wait limit?
No. It can be redesigned, for example with a burst guard, but that creates a new candidate that needs its own confirmation.
Does a sixth re-freeze replace Q5 everywhere?
Only if evidence supports that scope. A device-specific Q10 can coexist with Q5 for supported states and Q0 as rollback.
12. Changed-context evidence check and authoritative sources
Transfer the design to a partner feed whose field delay is caused by batch uploads rather than mobile delivery. Rebuild expected-arrival windows and release bursts from that mechanism; do not copy the eight-minute device rule. Keep common clocks, candidate eligibility and fresh confirmation.
NIST SP 800-137 supports continuous monitoring and timely risk response. NIST’s handbook explains cumulative-sum control charts and their sensitivity to sustained shifts. These sources support monitoring discipline; local queue, information and capacity assumptions determine the thaw and gates.
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: fifth-re-freeze monitoring and sixth-thaw gates · Earlier guide: fifth-thaw scope and portfolio-reset replays · 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.