Find your next learning step
When a clean-lineage breach forces an automation cap back to zero, the safer state still needs monitoring. Advisory use can drift, operators can work around the boundary and supposedly fresh cases can be influenced by the same development history. Restoring automation therefore requires more than a good score. The decisive evidence must be genuinely fresh, the mathematical and operational claim must be frozen in advance, and the restored cap must match what the evidence supports. This article uses persistent-homology examples to show that chain.
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 reassigned-cap state
Record the successor rule, input space, filtration, summary, threshold, decision role, former cap, reassigned cap, advisory boundary, fallback and breach rationale. A zero automation cap can still permit human-reviewed advice when independent evidence supports it.
Name prohibited actions explicitly. Operators should not treat repeated acceptance of advice as informal automation or batch-confirm outputs without case review.
CHAPTER 2 OF 12 · DEFINE THE DECISION
2. Separate monitoring, qualification and restoration
Cap monitoring checks whether the reassigned boundary is obeyed and whether advisory performance remains stable. Fresh-evidence qualification evaluates a frozen claim on protected cases. Restoration decides what automatic authority, if any, that evidence earns.
A strong qualification result does not force restoration. Consequence, recovery, detection and human factors may justify a smaller cap or continued advisory use.
CHAPTER 3 OF 12 · WORK THE LEDGERS
3. Work a cap-reassignment monitoring register
A persistent-homology rule classifies scanned rings after a lineage breach removed its 20% auto-release cap.
| Monitoring signal | Required boundary | Month 1 | Month 2 | Month 3 | Decision |
|---|---|---|---|---|---|
| Automatic releases | 0 | 0 | 0 | 0 | Compliant |
| Batch-confirmed advice | 0 | 0 | 3 | 1 | Breach then repair |
| Advisory false intact | At most 1% | 0.8% | 0.9% | 0.7% | Stable |
| Operator override | Review above 12% | 10% | 14% | 11% | Investigate |
| Version-linked decisions | 100% | 100% | 99.6% | 100% | Repair complete |
The team disables the batch control, reviews the affected decisions and adds a technical prevention check before beginning fresh qualification.
CHAPTER 4 OF 12 · WORK THE LEDGERS
4. Build genuinely fresh evidence
Freeze the candidate, thresholds, strata, necessary limits, minimum exposure and analysis before opening decisive outcomes. Generate new shapes, collect future field scans or use a protected source not selected because of the failed cases.
| Evidence component | Protection | Exposure | Necessary result | Outcome |
|---|---|---|---|---|
| Synthetic intact rings | Independent generator seed held by reviewer | 600 | False reject at most 3% | 2.0% |
| Fresh single breaks | Future production stream | 420 | False intact at most 1% | 0.5% |
| Uneven sampling | Prespecified perturbations | 240 | False intact at most 2% | 1.3% |
| New firmware | Protected scanner rollout | 120 | At least 150 cases | Insufficient |
| Fallback drill | Blind injected fault | 4 drills | Recovery under 5 min | 3.6 min |
Freshness depends on influence, not simply date. A newly collected case is not clean if its inclusion, labelling or processing was chosen after seeing how the candidate performed.
CHAPTER 5 OF 12 · WORK THE LEDGERS
5. Restore automation within the qualified boundary
The board restores a 10% cap for ordinary intact and single-break cases on the qualified scanner version. Uneven-sampling cases remain human-confirmed, and new-firmware cases remain advisory until exposure reaches the prespecified minimum.
Attach the cap to rule, scanner, preprocessing, strata and fallback versions. “Automation restored” without those qualifiers invites unsupported inheritance.
CHAPTER 6 OF 12 · TRANSFER THE METHOD
6. Monitor topology and operations together
Track sampling density, metric scale, filtration parameters, barcode summaries, distance to threshold, abstentions, overrides, fallback use and downstream outcomes. A stable Betti summary can still feed a changed operational rule.
Use known-shape and perturbation tests after library, scanner or preprocessing changes. Compare mathematical invariants and implementation outputs rather than assuming one proves the other.
CHAPTER 7 OF 12 · TRANSFER THE METHOD
7. Changed-context transfer: coverage-hole dispatch
Transfer the method to a sensor network whose topology rule advises repair dispatch for persistent coverage holes. A lineage breach removes automatic dispatch while advice remains available.
A protected future deployment stream qualifies common grid layouts but not sparse rural layouts. Automation returns under a 5% cap for common grids, while rural alerts retain human confirmation and separate evidence collection.
CHAPTER 8 OF 12 · TRANSFER THE METHOD
8. Connect lineage breach and cap governance
Begin with successor-rule monitoring, clean-lineage breach and cap-reassignment governance. Use successor-rule commissioning, clean-lineage qualification and cap-inheritance prohibition for the original commissioning boundary.
Continue through re-promotion monitoring, cross-lineage recurrence and permanent authority caps when repeated breaches constrain future expansion.
CHAPTER 9 OF 12 · CHECK LIMITS AND EVIDENCE
9. Limits and common misconceptions
Fresh evidence cannot erase the historical breach; it supports a new bounded claim. A restored cap is not a reward and a zero cap is not punishment. Both are statements about supported authority.
Topology does not make the decision scale-free. Filtration, metric, sampling and operational thresholds still determine what the computed feature means.
CHAPTER 10 OF 12 · CHECK LIMITS AND EVIDENCE
10. Practical evidence and restoration checks
Retain the breach map, cap decision, technical enforcement, frozen candidate, evidence-source controls, labels, analysis plan, exposures, results, fallback drills, authority vote and production configuration.
Before restoration, test that excluded strata cannot enter automation, every decision records its version and cap, and rollback works under a blind drill. After release, monitor the first bounded window before any expansion.
CHAPTER 11 OF 12 · ASK AND CONTINUE
11. Frequently asked questions
Must fresh evidence come from the future?
No, but it must be protected from the influence that compromised the earlier claim and fit the prespecified qualification role.
Can automation return for only one stratum?
Yes. Partial restoration is often the clearest match to uneven evidence.
Does a lower cap reduce the need for necessary safeguards?
No. A cap limits exposure but cannot make an unacceptable failure mode permissible.
CHAPTER 12 OF 12 · ASK AND CONTINUE
12. Sources and next reading
The W3C PROV Data Model represents entities, activities, agents and derivations that help make evidence influence and responsibility explicit. The Statistical Science article on safe anytime-valid inference explains evidence processes designed for data-dependent stopping. The ring and sensor-network cases are original teaching examples.
Continue through the Topology root, the Gold Standard of Education and How X Works.
Top · Contents · Previous · How X Works