Rolling Planning Horizons

1. The real problem: one fixed calendar creates false certainty

When every item appears on one calendar with one commitment level, this week's work and a three-month possibility look equally certain. A rolling horizon corrects that by separating NOW / NEXT / LATER and DEFERRED; commitment should weaken as the horizon extends, while review requirements become more explicit.

2. Lesson boundary and owned artifact

The owned output is Rolling Planning Horizon Decision Ledger v1. For each item it records evidence cutoff, signal window, confidence, commitment, capacity cost, dependency state, review trigger, replan reason and provenance. Content Gap Detection belongs to PLN-010 and inbox privacy belongs to PLN-008.

3. A horizon is a decision state, not just a date

NOW is near-term work with enough evidence, capacity and dependency readiness for operational commitment. NEXT is plausible but less committed or waiting on bounded evidence. LATER contains low-commitment possibilities. DEFERRED means current evidence, capacity or dependencies do not justify movement.

4. Record the evidence cutoff before moving an item

Every placement must say which evidence, up to what cutoff, supported it. Without a cutoff, the next cycle cannot distinguish a new decision from silent history rewriting. Evidence arriving after the cutoff may trigger review, but it should not erase the prior rationale.

5. Search Console provides time-bounded signals, not priority

The Search Console Performance report supports date ranges and comparisons, so query/page signals can feed a bounded review window. Those signals do not determine business priority, capacity or commitment. Record the range, granularity and provenance with the planning row.

6. Google Trends is not absolute demand

Google Trends uses sampled search data and normalizes it by time and location. A value of 100 or a short spike is therefore not a publication mandate. Treat Trends as one contextual signal inside the selected window and geography.

7. Confidence bands must be explainable

Use HIGH / MEDIUM / LOW with a visible reason rather than a decorative score. HIGH can mean fresh, corroborated evidence plus clear dependencies; MEDIUM can mean one bounded evidence or dependency gap; LOW can mean weak, transient or conflicting signals. Do not coerce UNKNOWN into LOW.

8. Match the commitment gradient to the horizon

NOW should carry stronger commitment than NEXT, and NEXT stronger than LATER. If every horizon is frozen at the same commitment level, the system is merely a fixed calendar with three labels. COMMITTED / PROVISIONAL / OPTIONAL is one possible pattern, not a universal threshold.

9. Apply the capacity gate before promotion

Every promotion into NOW must fit the capacity budget. If available capacity drops by 40%, placements must be recomputed; keeping the same NOW list and squeezing the schedule is not an auditable response. Record each item's cost next to the available budget.

10. Dependency state blocks optimistic promotion

Separate READY, PENDING and BLOCKED. A BLOCKED item cannot enter NOW because of a trend spike. PENDING work may remain in NEXT with a finite trigger. If a critical dependency is unknown, preserve REVIEW_REQUIRED.

11. A review trigger must be finite

"Review later" is not a trigger. Valid examples include dependency X becoming ready, the end of a seven-day signal window, a capacity threshold breach, new cohort evidence, or the next monthly review date. The trigger must be testable.

12. Replanning requires a reason

Moving LATER→NOW, NEXT→DEFERRED, or any other horizon transition needs a replan reason such as new evidence, capacity change, dependency change, context change or expiry. A silent move is REVIEW_REQUIRED even when the destination appears sensible.

13. Worked example: four items, four commitment states

Assume NOW capacity is five units. A has high confidence, cost=2 and a ready dependency, so it enters NOW. B has medium confidence and a pending dependency, so it remains NEXT. C has only a short-lived spike, so it stays LATER with a seven-day review. D costs four units when only two remain, so it is DEFERRED. The ledger stores the reason independently from the label.

14. Deterministic failure probes

Reject four shortcuts: (1) one commitment level for every horizon; (2) a 40% capacity cut without recomputing placements; (3) allowing a short-lived trend spike to override stronger evidence; (4) moving LATER→NOW with no new evidence or replan reason. In all four cases, preserve history and constrain the decision.

15. Transfer: a documentation program with monthly releases

A monthly documentation program should not copy the dates of a weekly plan. Recompute horizon lengths, commitment levels and review triggers for the new release rhythm. With 40% less capacity, the contract stays stable while its context-specific parameters change. The final ledger must show what changed, why, and which evidence supports the change.