Distinguish absence from withdrawal
The August fixture starts with growth 0.70 and inflation 0.10. Both are eligible by the fictional cutoff of 20 September 2024 at 16:00 UTC. Inflation is exactly on the inclusive boundary, so the initial label is Boundary. This is a deliberate equality test, not a report about an actual inflation announcement.
The later growth revision contains an explicit null. Null is a withdrawal marker in this fixture, whereas the absence of any eligible record means no record is available yet. Those situations share the Unavailable output but leave different input evidence. A reviewer should be able to distinguish them without reconstructing the database from memory.
Select the record before inspecting its value
Selection chooses the highest eligible revision number for each feature and period. Only then does classification inspect whether the selected value is usable. For the latest August growth record, revision two wins and its null value is preserved. Searching backward for the most recent non-null value would implement a different rule and defeat the withdrawal.
The second inflation score is 0.15. It is outside the boundary band, but one valid feature cannot repair the other feature's withdrawal. The latest label is therefore Unavailable. The record still retains the growth revision identifier and its release and ingestion timestamps, allowing a reader to see why the calculation was withheld.
Do not turn a data event into a market transition
Comparing Boundary with Unavailable records a changed output, but it is not evidence of an economic transition. The change comes from the explicit data-status rule. Treating all adjacent labels as economic states would muddle information quality with macro conditions and could produce meaningless transition frequencies.
Likewise, dropping withdrawn records from a historical archive can make apparent coverage improve. This edition keeps all twelve fixed decision periods and all forty-eight feature revisions. The complete ledger is downloadable, with both the initial and revised versions present. The original archive remains accessible even when a later view has less usable information.
Record what a future recovery would require
A third revision containing a valid value would need a new release timestamp, an ingestion timestamp no earlier than release, and a consecutive revision identity. An earlier cutoff still could not use it. Duplicate revision identifiers, reversed chronology and out-of-range feature values are rejected rather than silently sorted into a plausible record.
No such third revision is included in this edition. The whole example, including the withdrawal, was constructed on 16 October 2026 and is not a real operational incident. Its purpose is to make the treatment of nulls observable. It cannot certify a provider's data controls, predict the restoration of a source or establish an investment consequence.
Inspect the comparison
| View | Growth revision / value | Inflation revision / value | Label |
|---|---|---|---|
| At September cutoff | 1 / 0.70 | 1 / 0.10 | Boundary |
| Latest fictional revision | 2 / null (withdrawn) | 2 / 0.15 | Unavailable |
Primary references
References support the stated methods and data conventions. Original fictional examples are not observations from the cited sources.