Event clusters and overlapping excursions
Keep eligible observations, threshold events and independent sampling units separate when one displacement persists.
Many threshold rows can describe one excursion
A price can remain stretched across several adjacent observations. Counting each threshold match as an independent opportunity may substantially overstate the information in the record. The source population, event rule and any cluster definition need prior declaration. The original lab reports eligible next-close observations and a stretch subset; it does not claim that each is an independent trade. The new path lens measures one supplied path rather than separating a complete genuine event population. A larger row count cannot by itself justify narrower uncertainty or a stronger trading conclusion.
A cluster rule changes population ownership
Possible event definitions include every eligible threshold observation or an excursion-start rule with a stated reset. Neither is universally correct for every hypothesis. Choose before observing outcomes and preserve which rows belong to which event. A reset rule based on future success would introduce another selection boundary. Retain unavailable inputs, zero-scale windows, repeated triggers and unresolved endpoints. If an intervention suppresses later actions, keep its actual ownership distinct from the descriptive event count. The teaching cards do not implement a genuine clustering estimator or silently choose the financially favorable definition.
Overlapping labels need their own endpoints
Two decisions can use intersecting future observations to resolve their labels. This dependence is different from a label crossing a train/validation boundary. Both deserve explicit treatment. A fixed horizon can support software endpoint checks without establishing statistical independence. Variable targets require their own label ends and native policy. Do not manufacture an effective sample size from an unexplained formula. The publication estimates no confidence intervals from its authored cases. A genuine financial process needs reserved evaluation and its independently required master under the actual event and overlap contract.
Selection can also hide repeated attempts
Repeatedly changing a threshold or reset to favor inspected events consumes the consulted sample. Preserve all attempted candidate definitions, not just the eventual winner. Raw causal entry ranking remains separate from the later chosen native exit and account result. Software fixtures can test that a cluster rule behaves as specified, but their known answers are not blind financial evidence. A documented no-model master can be valid completion; an incomplete event grid cannot be relabelled a complete rejection. State the actual preserved stage and missing artifact rather than overstating what counts alone prove.
Authored reference and excursion case
An authored run has stretched observations at steps ten, eleven and twelve, all within one unresolved displacement. Under an every-observation rule there are three events. Under a predeclared excursion-start rule there might be one, provided its reset conditions are specified. The two denominators cannot be silently mixed. These indices are fictional and provide no estimate of real event frequency, dependency or profitability. The example asks for population ownership before a confidence claim is made.
| Object | Record | Bounded finding |
|---|---|---|
| Step 10 | Stretch starts | Possible cluster start |
| Steps 11..12 | Still stretched | Repeated observations |
| Every-row population | Three events | Not automatically independent |
| Start-only population | Reset rule required | Separate frozen definition |

Primary references and scope
Source review: 14 October 2026. Official mechanism descriptions are not this desk's paid-account test, independently captured fills or vendor-return audit.