Find your next learning step
A topology rule can survive many local repairs while its larger claim family quietly consumes data, analyst attention and error allowance. Family-wide stopping rules decide when testing must pause or end across all related versions. Independent audit capital protects a separate evidence stream from the development search. Claim retirement closes the scientific and operational authority of the family when another repair would no longer be a credible use of evidence.
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 what can stop across the whole family
The family is the set of rules linked by claim, data, repair lineage, selectable channels or decision target. Its stopping rule applies even if the current version has a new label. Define the family before outcomes accumulate.
State whether stopping freezes testing, suspends deployment, triggers independent audit or retires the claim. One trigger may govern evidence collection while another governs operational authority.
CHAPTER 2 OF 12 · DEFINE THE DECISION
2. Separate local failure from family exhaustion
A local version can fail because one threshold or channel is wrong. Family exhaustion occurs when repeated repairs consume the available confirmation pathways, violate a global budget, expose unstable mechanisms or make another clean test implausible.
Do not require a catastrophic last failure. Planned exhaustion limits protect against continuing until one favourable path appears. They also let a team retire a claim without pretending every mathematical idea behind it is false.
CHAPTER 3 OF 12 · WORK THE LEDGERS
3. Build a family-wide stopping ledger
An optimiser governs active-set unlocks through a repaired topology family.
| Trigger | Family threshold | Current state | Consequence | Owner |
|---|---|---|---|---|
| Repair count | Maximum 3 | 3 | No further internal repair | Method lead |
| Exposed rounds | Maximum 220 | 214 | Reserve audit only | Data steward |
| Global evidence spend | Maximum 0.040 | 0.038 | Stop primary testing | Statistician |
| Boundary crossings | Maximum 2 | 2 | Suspend deployment | Safety owner |
| Audit failure | Any confirmed failure | Not tested | Retire claim | Independent panel |
Because repair count and boundary crossings are already exhausted, a fourth internal patch is not authorised even though six rounds remain.
CHAPTER 4 OF 12 · WORK THE LEDGERS
4. Protect independent audit capital
The programme ring-fences 0.010 evidence capital for an independent audit and keeps its cases, allocation rule and analysis inaccessible to the repair team.
| Evidence account | Starting capital | Spent | Available | Permitted use |
|---|---|---|---|---|
| Development diagnostics | Not inferential | N/A | Open | Design and debugging |
| Family primary | 0.040 | 0.038 | 0.002 | Frozen: below minimum tranche |
| Independent audit | 0.010 | 0.000 | 0.010 | One preregistered audit |
| Future new claim | Separate protocol | 0.000 | Undefined | Only after boundary review |
Operational independence matters as much as arithmetic. If the repair team previews audit cases or changes the rule in response, the audit no longer answers the preregistered confirmation claim.
CHAPTER 5 OF 12 · WORK THE LEDGERS
5. Design predictable stopping and escalation
A valid sequential design can allow continuing looks when its update rule is predictable and matched to the evidence process. But governance still needs practical caps: time, exposure, repairs, decision losses and audit availability.
Escalate before the last unit is spent. A warning boundary can freeze deployment and preserve enough capital for an independent diagnosis. Record who can invoke emergency containment and what evidence may be viewed.
CHAPTER 6 OF 12 · TRANSFER THE METHOD
6. Retire the claim, not the archive
Claim retirement ends authority for a named proposition, such as this topology rule safely unlocks the active set under the declared operating conditions. Preserve code, provenance, failures, diagnostics and audit results for learning.
Retirement can be partial. One domain, endpoint or channel family may retire while a separable claim remains. Any successor must state whether it is a narrower claim, a new mechanism or merely another member of the exhausted family.
CHAPTER 7 OF 12 · TRANSFER THE METHOD
7. Changed-context transfer: adaptive screening graph
Transfer the method to a screening graph whose edges activate follow-up tests. Analysts have already removed two false branches and recalibrated thresholds three times. A family-wide rule now blocks another internal edit and reserves independent patient cases for audit.
If the audit confirms unstable edge selection, the unlock claim retires. A later graph built for a different endpoint with prospectively collected evidence may open a new family only after lineage and claim boundaries are reviewed.
CHAPTER 8 OF 12 · TRANSFER THE METHOD
8. Connect restart limits, reinstatement and provenance
Begin with restart-family limits and global evidence budgets. Use reinstated-rule monitoring and re-revocation boundaries to recover operational triggers.
Return to reinstatement trials, contamination barriers and wealth restarts when evaluating whether the audit stream remained independent.
CHAPTER 9 OF 12 · CHECK LIMITS AND EVIDENCE
9. Limits and common misconceptions
Anytime-valid evidence does not license unlimited redesign. Validity depends on the defined process; changing hypotheses, channels or data access can create a new selection problem. A numerical capital account cannot repair hidden contamination.
Retirement is not proof that the underlying topological idea is universally false. It states that the declared evidence path no longer supports operational authority. Audit independence also requires governance, access controls and separate judgement, not just a different random seed.
CHAPTER 10 OF 12 · CHECK LIMITS AND EVIDENCE
10. Practical evidence and retirement checks
Preserve the claim, family boundary, versions, repair graph, visible data, predictable update rules, starting and spent capital, stopping thresholds, boundary crossings, deployment exposure and audit reservation.
For retirement, record effective date, scope, reason, affected systems, fallback, archive location, public limitation, successor boundary, re-entry prohibition and the authority required to propose a genuinely new claim.
CHAPTER 11 OF 12 · ASK AND CONTINUE
11. Frequently asked questions
Can unused primary capital move to the audit?
Only if the protocol allowed that transfer predictably before results. Ad hoc rescue funding can compromise the audit’s intended independence.
Can a retired claim return?
Not as the same unmodified claim under the spent family. A substantively new claim needs a new boundary, evidence source and prospective protocol.
Who should decide retirement?
The named authority should include operational and independent scientific judgement, with conflicts and evidence access recorded.
CHAPTER 12 OF 12 · ASK AND CONTINUE
12. Sources and next reading
The W3C PROV-O Recommendation supplies terms for entities, activities, agents, derivation and invalidation that can preserve a repair and retirement lineage. Ramdas and colleagues explain safe anytime-valid inference with e-processes, including evidence processes designed for optional continuation. Vovk and Wang study merging sequential e-values via martingales. The ledger and capital amounts are original teaching examples, not a universal testing prescription.
Continue through the Topology root, the Gold Standard of Education and How X Works.
Top · Contents · Previous · How X Works