Label-window overlap and boundary ownership
Trace fixed and variable outcome endpoints without equating an excluded decision count with independent executed trades.
A decision and its outcome occupy different times
A feature vector belongs to the declared decision boundary. A forward target may use several later observations before its label becomes resolvable. Both clocks need ownership. A development decision near the region end may depend on a label that finishes in validation; allowing that label into fitting would cross the intended boundary. Earlier context can be valid feature warmup without authorizing future target access. Record decision index, label start, label end and allowed fit region rather than relying on a single timestamp for every role.
The planner handles one fixed horizon
The existing planner assumes each outcome uses exactly the entered number of subsequent rows. Under its default, development ends at row 760 while eligible development decisions end at row 755. A decision at 755 can resolve a five-row label within development. The final five rows remain observations but are excluded as development decisions. The same arithmetic applies at validation and holdout ends. These exclusions compute boundary eligibility, not independent trade count or a complete causality guarantee. The source's actual arrival and transformation scope still require separate evidence.
Variable targets need individual endpoints
A native exit or recovery label can resolve at different observations. One fixed gap cannot automatically represent every such outcome. Record each label's actual permitted end under its frozen policy and determine boundary ownership from that record. Do not extend the allowed observation window retrospectively to obtain a favorable result. Overlapping labels also create dependence questions distinct from leakage: removing a boundary-crossing outcome does not make the remaining outcomes independent. The site's arithmetic does not estimate effective sample size or certify a statistical test for a real variable-exit model.
Keep financial and software completion separate
Invented row-index fixtures should test endpoint inclusion, equality and exclusion behavior. Their known expected answers belong to software correctness. Real research uses the authoritative factory's source route and full master, preserving native entry, exit, warmup and fill rules. A teaching planner should not replace those requirements with a compact split grid. When labels or sources are unavailable, state the exact missing artifact without claiming that the market failed the complete system. A documented boundary is useful preparation but is not itself an independently validated strategy.
Fictional model evidence case
In the authored planner default, decision 755 has label endpoint 760 and remains eligible for development. Decision 756 would resolve at 761 and is excluded from development under the five-row contract. A different fictional target resolving after two rows would require its own policy and endpoint, not silent reuse of the five-row conclusion. The example contains no actual calendar, instrument or fill. Its purpose is to preserve ownership at an inclusive region end.
| Object | Record | Bounded finding |
|---|---|---|
| Decision 755 | Five following rows end at 760 | Eligible development |
| Decision 756 | Five following rows end at 761 | Excluded development |
| Variable label | Own endpoint required | No universal fixed-gap certification |

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.