Find your next learning step
A repaired topology eligibility rule may pass a clean reinstatement trial and still fail later. Monitoring must therefore remember both its new prospective evidence and its earlier revocation. The difficult part is repeated opportunity: if a rule can be revoked, repaired and restarted many times, choosing the most favourable restart can make weak evidence look strong. A defensible system predeclares re-revocation boundaries, keeps wealth and data lineages separate across restarts and limits how many renewed claims can consume the error budget.
Open the chapter contents · Topology 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. Define the reinstated monitoring state
Record the active rule version, eligible channels, exposure cap, starting evidence account and the exact monitoring clock. Earlier revoked versions remain historical comparators, not pooled observations.
Monitor validity, operational loss and topology-specific boundary conditions. A rule can remain statistically supported while its active-set interpretation stops matching the problem.
CHAPTER 2 OF 12 · DEFINE THE DECISION
2. Predeclare re-revocation boundaries
Use boundaries for sudden harm, cumulative underperformance, lineage break and contamination. A threshold should say whether it pauses exposure, narrows scope or revokes the rule.
Make the boundary predictable from past information. Changing it after a loss sequence appears can invalidate the evidence interpretation.
CHAPTER 3 OF 12 · WORK THE LEDGERS
3. Build a restart-lineage ledger
A topology optimiser has undergone two repairs and three prospective monitoring eras.
| Era | Rule version | Starting wealth | Eligible rounds | Outcome |
|---|---|---|---|---|
| Initial promotion | R3 | 1.00 | 46 | Revoked at loss boundary |
| First reinstatement | R4 | 1.00 | 38 | Re-revoked after lineage break |
| Second reinstatement | R5 | 0.50 | 52 | Active with capped exposure |
| Independent audit | R5-audit | 0.50 | 52 | No boundary crossed |
Every era keeps its own filtration: which outcomes were available when each decision and allocation was made.
CHAPTER 4 OF 12 · WORK THE LEDGERS
4. Worked multiple-restart error budget
The programme permits an overall false-unlock budget of 0.05 across three prospective restart attempts. It chooses fixed allocations before results arrive.
| Attempt | Allocated alpha | Evidence outcome | Unused budget | Governance decision |
|---|---|---|---|---|
| R4 | 0.020 | Fails lineage boundary | 0 | Revoke |
| R5 | 0.020 | Passes primary gate | 0 | Capped reinstatement |
| R6 reserve | 0.010 | Not started | 0.010 | Held for future |
| Total | 0.050 | — | 0.010 | Budget respected |
An e-process formulation can manage repeated looks within an era, but moving between eras still requires a declared capital and lineage rule.
CHAPTER 5 OF 12 · WORK THE LEDGERS
5. Separate within-era looks from between-era restarts
Anytime-valid monitoring allows repeated inspection under its stated protocol. It does not automatically license unlimited redesign and restart with the same data.
When a repair uses outcomes from the failed era, those outcomes become development evidence. Confirmation begins with new outcomes or a formally justified independent stream.
CHAPTER 6 OF 12 · TRANSFER THE METHOD
6. Use quarantine during uncertain lineage
Delayed outcomes can arrive after re-revocation. Tag them to the version and exposure that generated them before deciding whether they inform diagnosis, audit or a later prospective claim.
Quarantine wealth affected by uncertain eligibility until the lineage question is resolved. Do not choose the more favourable inclusion rule after seeing the outcome.
CHAPTER 7 OF 12 · TRANSFER THE METHOD
7. Changed-context transfer: active-manifold acceleration
Transfer the method to an optimisation rule that accelerates once iterates appear to share an active manifold. Two previous rules locked too early under correlated noise.
The transfer check assigns a new error budget to the repaired detector, limits initial acceleration and uses an independent audit stream before widening the eligible topology class.
CHAPTER 8 OF 12 · TRANSFER THE METHOD
8. Connect reinstatement, promotion and restart accounting
Begin with rule-reinstatement trials and wealth-restart protocols. Use the promoted-rule monitoring and revocation guide for the first operational boundary.
Return to sleeping channels and restart accounting and e-process mixtures and channel allocation when evidence streams activate at different times.
CHAPTER 9 OF 12 · CHECK LIMITS AND EVIDENCE
9. Limits and common misconceptions
Splitting an error budget is one valid teaching design, not the only possible method. A formal procedure must match the statistical object, dependence structure and stopping rule.
Re-revocation does not prove that reinstatement was irrational. It shows that governance responded to new evidence under a declared boundary.
CHAPTER 10 OF 12 · CHECK LIMITS AND EVIDENCE
10. Practical evidence checklist
Archive every rule version, repair input, freeze time, eligible channel, outcome delay, filtration, starting wealth, alpha allocation and monitoring boundary.
For each restart, save contamination decisions, spent and reserved budget, exposure cap, re-revocation event, quarantine ledger, audit result and conditions for any future attempt.
CHAPTER 11 OF 12 · ASK AND CONTINUE
11. Frequently asked questions
Can an anytime-valid test be restarted after failure?
Only with a protocol that accounts for the new attempt, its capital and the information used to design it. A reset button alone is not a guarantee.
Must every re-revocation end the programme?
No. It can end the current claim while preserving a bounded route to diagnose, repair and test a genuinely new version.
Why keep an independent audit stream?
It reduces shared tuning and lineage risk, although independence must be designed rather than assumed.
CHAPTER 12 OF 12 · ASK AND CONTINUE
12. Sources and next reading
The W3C PROV-O Recommendation supplies a vocabulary for entities, activities, agents and derivations. Ramdas and colleagues explain safe anytime-valid inference with e-processes. Choe and Ramdas discuss why combining evidence across different filtrations requires explicit care. The ledgers are original teaching examples.
Continue through the Topology root, the Gold Standard of Education and How X Works.
Top · Contents · Previous · How X Works