A net response in an AI lending survey is not loan-volume growth
Keep bank-survey net responses separate from AI infrastructure loan amounts and their growth rates.
Numbers can measure different things
A statement about stronger AI infrastructure financing can tempt an analyst to put bank-survey numbers into a loan-growth chart. A bank's reported direction of change and the amount it actually lends are different quantities. Truthound cannot infer that distinction. Its useful role is to check a table whose measure types and periods have already been defined.
What the October survey covers
The ECB's October 27, 2026 bank lending survey covers 159 banks. Fieldwork ran from September 21, 2026 to October 6, 2026, separating 3rd-quarter observations from 4th-quarter expectations. Its AI questions describe most banks as not yet having dedicated quantitative criteria and anticipate gradually increasing demand for digital-infrastructure financing.
The denominator behind a net response
A net percentage subtracts the share reporting a decrease or easing from the share reporting an increase or tightening. Different underlying response distributions can produce the same net result. The loan amount of one large bank and the opinions of several small banks are not combined into one monetary quantity by that statistic. A column labelled only growth hides this denominator.
Try to disprove the interpretation first
The first hypothesis to challenge is that a higher survey net response means the same proportional increase in AI lending. Testing it requires actual loan amounts, a definition of AI financing, the reporting-bank population and a matching period. Without those data, retain the survey as a directional response measure. Filling a missing amount with a net response changes the measure rather than completing it.
Define the checks in Truthound
An illustrative preparation table could include measure_kind, observed_or_expected, reference_quarter and respondent_population. Truthound Core Python SDK 3.3.4, operating on a prepared Polars table, provides BetweenValidator for operator-defined numeric bounds and InSetValidator for allowed measure categories. A value within bounds can still be incorrectly classified as loan growth. Keep the definition's owner and evidence alongside the rule.
Keeping an existing GX Core workflow
GX Core 1.24.1 is also a defensible choice for an existing survey workflow. Its range Expectation provides minimum, maximum, strict boundaries and mostly settings, with pandas, Spark and SQL backends documented. Moving a prepared Polars table to Truthound is an operating-path decision, not an inherent measurement improvement. Align boundaries before comparing failure counts.
Check the population for each question
Knowing the survey's total respondent count does not establish that every question has the same interpretation or response population. Preserve the AI-specific question, general enterprise-lending question and next-quarter expectation as distinct records. Do not fill a missing detailed response with a percentage taken from the overall survey.
Do not merge expectations into observations
Treat a mixture of observed and expected periods as a separate failure. Joining a reported 3rd-quarter change to an expected 4th-quarter change can make a forecast appear to be an outcome. Preserve the survey round if the purpose is to compare earlier expectations with later observations; do not overwrite the former with the latter.
What evidence would change the explanation?
Even a valid table needs monetary evidence before the explanation becomes loan-volume growth. If the amount series covers a different population, show the two indicators alongside each other instead. The deciding evidence is a comparable population and period, rather than the reassuring impression that both lines move in the same direction.
Conditions
Prepare separate measure types, respondent populations and observed versus expected quarters.
Alternatives
Retain GX Core 1.24.1 range checks when they maintain the same bounds and null policy in the existing workflow.
Limitations
Survey net responses alone do not establish growth in actual AI loan amounts.
Evidence
- Event checked
- Subject checked
- Modified
