The same banking quarter can have different data cutoffs
Retain population, cutoff, publication date and version separately from the reference quarter for SI, LSI and system-wide statistics.
A reference quarter is not enough
Two banking statistics labelled 2nd quarter need not describe the same information state. Different cutoffs can place later corrections in one population but not the other. Truthound can check dates and populations in a release-admission table; economic interpretation remains with the analyst.
Two populations in the calendar
The ECB calendar distinguishes a 2nd quarter of 2026 SI cutoff of August 30, 2026 and publication on September 15, 2026. LSI and system-wide statistics show an October 4, 2026 cutoff and October 15, 2026 publication. Population and collection timing differ despite the shared quarter.
Separate the collection clocks
Keep reference quarter, population, cutoff, publication and acquired file version separately. The collector's observation time is another field. Publication time should not replace the data reference period, nor should collection time be presented as publication.
Conditions for a system-wide total
A system-wide aggregate needs the component cutoffs explained. Choosing each population's latest same-quarter file is not sufficient approval to combine them. Record the expected populations and the component versions used for the total.
The date and set policy in Truthound
Truthound Core Python SDK 3.3.4, operating on a prepared Polars table, provides DateOrderValidator for defined cutoff/publication relationships and DistinctValuesEqualSetValidator for a supplied expected population set. Owners supply the catalogue and exceptions. The validator cannot independently know which institution population is missing.
Compare the existing exceptions
An existing Pandera 0.35.0 pandas predicate can maintain the admission rules. Compare a same-quarter file from a different cutoff, a missing population and a cutoff recorded after publication. Matching operating decisions requires aligned definitions, not generic support for date checks.
When a later file revises an earlier one
Do not overwrite a file silently when a later release arrives. Preserve which report used which version and compare revisions. A changed indicator may reflect revised data rather than a new economic development; the history should allow that distinction.
Divide collection and analysis ownership
Collection owners can confirm the file-metadata relationship while analysis owners assess comparability of components. Successful collection should not become aggregation approval automatically. Give unresolved exceptions a clear destination before the next release.
Leave a reusable admission record
The reusable rule is an explainable record of which population and vintage were accepted, rather than simply choosing the latest date. That record supports the next quarter's comparison. Align the data clocks before adding the numbers.
Conditions
Use the public ECB calendar and verified population definitions for the extract.
Alternatives
Keep existing Pandera 0.35.0 pandas date and population predicates where they preserve the same admission policy.
Limitations
A valid timing contract does not determine economic interpretation or an individual bank's soundness.
Evidence
- Event checked
- Subject checked
- Modified
