Keep the decision fixed
The invented May observation has a fixed decision cutoff of 20 June 2024 at 16:00 UTC. Growth is available, but inflation is deliberately delayed until 22 June. Moving the decision to a later date would answer a different question. The archived result must represent the declared cutoff, even when a more complete record arrives soon afterwards.
All dates and values in this example were created in October 2026. They are neither a reconstruction of an actual agency delay nor evidence that a real publication withheld a signal. The illustration isolates a common data-handling choice: whether a missing input stays missing or is converted into an economically meaningful number.
Inspect the absent feature
The initial growth score is -0.40. At the fixed cutoff there is no eligible inflation revision for May. The engine returns Unavailable and keeps the selected inflation record empty. It does not invent a score of zero, borrow April's observation or search forward for the eventual June ingestion.
Zero would have a specific meaning under this rule: it lies inside the inclusive 0.1 boundary band. Replacing absence with zero would therefore change the result to Boundary and conceal the coverage failure. Carrying a previous period forward would introduce a separate stale-data policy, which this edition explicitly does not implement.
Advance time without rewriting the archive
The delayed first inflation score, 0.05, becomes eligible at 22 June 2024 at 13:05 UTC. A replay with that later cutoff selects it and returns Boundary, because 0.05 remains within 0.1 of the zero inflation threshold. The fixed decision archive stays Unavailable; a later interactive replay is a new query, not a correction to the earlier information set.
The latest fixture revision, notionally ingested in March 2025, contains growth -0.35 and inflation 0.20. Both are outside the boundary band and the resulting label is G- / I+. That third state is retrospective relative to June. Retaining the three stages explains why a changed label need not mean that the underlying economy moved at each timestamp.
Separate coverage from classification
A coverage report should count the missing pair before computing state proportions. Otherwise a provider might exclude unavailable dates from a denominator without making the resulting selection visible. In this small fixture, May contributes one of twelve fixed decision records and remains explicitly unavailable. It is not silently removed from the comparison total.
A production specification would additionally need release-calendar ownership, lateness thresholds, holidays, time zones and an escalation route. None is inferred from this fictional twentieth-day schedule. The teaching rule establishes a conservative and reproducible behavior for missing observations, not an operational response standard or an investment instruction.
Inspect the comparison
| Information cutoff | Growth | Inflation | Result |
|---|---|---|---|
| 2024-06-20T16:00:00Z | -0.40 | Not available | Unavailable |
| 2024-06-22T13:05:00Z | -0.40 | 0.05 | Boundary |
| Latest fictional revision | -0.35 | 0.20 | G- / I+ |
Primary references
References support the stated methods and data conventions. Original fictional examples are not observations from the cited sources.