Compare experimental service indices without losing base year and revision state
A service-industry comparison needs its base year and provisional or revised state attached to each value.
Read the state of the index first
When comparing IT-service activity with other industries, collecting index values alone can erase their meaning. A table may look complete even when it mixes base years or values from before and after a revision. Truthound can check the version and category contract in a prepared statistical extract. It does not explain the causes of industrial growth or forecast a company’s future revenue.
Preserve the experimental-series qualification
On 29 October 2026, India’s MoSPI released August 2026 service-production indices for 19 subsectors using base year 2024–25. The release calls the series experimental and identifies railway, banking and insurance observations as provisional monthly data subject to annual revision. Even within one publication, different subsectors can therefore carry different data states. Those qualifications belong in the comparison, rather than being lost when values are copied.
Retain classification and release identity
A proposed extract records subsector code, base year, observation month, release vintage and provisional state. Release date and observation month need separate fields. If a readable industry label changes, retain a reference classification explaining the code. When an older value is updated from a later release, preserve enough version information to reconstruct which value supported an earlier planning decision or chart.
A revision is not simply a duplicate
Two rows for one subsector and month may represent an earlier release and a revised value. Keeping the first encountered row would discard the revision without a reason; averaging both would invent a new observation. Choose the release vintage intended for the comparison first. Then test key uniqueness and categories within that version, instead of treating every repeated month as an accidental duplicate.
Put statistical meaning into the contract
Truthound Core Python SDK 3.3.4 (open-source SDK, Polars backend) documents ColumnPairInSetValidator for allowed pairs supplied by an operator. It could check approved subsector and data-state combinations. MultiColumnUniqueValidator can separately test a chosen composite key inside the selected release vintage. Define key columns and null policy explicitly. A unique, well-classified observation still does not establish that the experimental methodology is representative or appropriate for the analyst’s economic question.
An existing schema may cover the task
Pandera 0.35.0 (MIT open-source Python DataFrame validation library; pandas, Polars and other supported backends) offers declarative and custom checks for the same prepared extract, and it also documents Polars support. If an established statistical workflow already maintains that contract, identify a real gap in version selection or failure handling before replacing it. Selecting Truthound does not move the interpretation of an experimental index into the validation library. The statistical owner remains responsible for the meaning and suitability of the comparison.
Make one comparison reproducible
Begin with a year-on-year comparison for one subsector and record the exact vintage and base year used. Hold rows whose state cannot be established from the source. If a revised value changes the conclusion, describe that separately from a calculation error. A reproducible comparison gives an industrial-outlook discussion a stronger foundation than a larger table whose revisions and reference periods cannot be explained.
Conditions
Fix subsector classification, base year and release vintage, retaining provisional states.
Alternatives
Use an existing Pandera 0.35.0 statistical contract that manages vintages and exceptions.
Limitations
A unique observation does not establish representativeness or a valid economic interpretation.
Evidence
- Event checked
- Subject checked
- Modified
