What a requirement can and cannot tell us
SREP, the supervisory review and evaluation process, connects assessment of an institution's risk profile with its ability to manage those risks. Pillar 2 adds an institution-specific dimension to standard requirements. My proposed analytical reading would begin with scope, timing and the category of the figure being reported, rather than a ranking of percentages. A single capital number cannot fully describe a bank or an individual process.
This article sets out an approach to public information and a hypothetical internal finding. It does not describe supervisory work I performed or reach a conclusion about a particular bank. Moving from aggregate results to an individual institution requires additional evidence. For figures applying in 2026, the assessment year, publication date and effective date of the decision also need to remain distinct.
Documented case: the Czech National Bank, February 2026
On 9 February 2026, the Czech National Bank published aggregate results of assessments carried out in 2025. The new Pillar 2 requirements applied from 1 January 2026. The weighted average P2R was 2.6%, down by 0.2 percentage points year on year. [1] This is a specific Czech supervisory publication, rather than a statement about every institution or every control process.
My inference is limited: an aggregate decline does not establish that a particular weakness has become immaterial. An average combines different institutions and weights. Explaining a specific finding still requires evidence about its own scope and context. The public report can help frame questions, but it cannot replace the evidence needed to answer a question about one internal decision or control.
Distinguish requirements from expectations
When reading European materials, I would distinguish P2R from P2G. The ECB explains that the Pillar 2 requirement is binding, whereas Pillar 2 guidance expresses supervisory expectations and is not legally binding. [2] These terms should not be interchangeable in a working paper. Nor should a specific Pillar 2 figure be confused with the total capital ratio or with capital buffers serving other purposes.
For each table, I would capture the source, reference date, unit and consolidation level. Figures become comparable only when they describe the same concept. A change in percentage may involve a different basis or population. Before drawing a trend, I would check the institution, period and category. Otherwise an arithmetically correct comparison can still tell an inaccurate story about what changed and why.
Hypothetical example: missing exception approvals
Imagine a fictional bank where exception approval cannot be found in three of twenty selected credit files. That observation does not establish that approval never existed or that fifteen percent of the whole portfolio has the same problem. I would first examine sample selection, document locations and the applicable rule. A missing link and a missing decision represent different conditions requiring different explanations.
Timing and recurrence matter too. A document created after the decision does not represent the same original evidence as timely consent. If all three cases share one system version, the boundary may be technically identifiable. If similar cases appear across products and teams, broader investigation becomes reasonable. The numbers and bank are hypothetical and are not findings taken from the Czech National Bank's publication.
Build a traceable analytical account
I would start with the rule and observed departure, separating both from any causal hypothesis. Important claims need evidence and limitations. I would then map the process: who prepares information, who decides, where the record is stored and which check detects absence. This makes it easier to distinguish a local filing problem from weaknesses in decision making or oversight of the process.
An interview can suggest a cause but should not establish it alone. If staff believe records disappeared during migration, I would seek a corresponding time boundary and affected population. If the evidence does not support that account, other explanations remain open. Confident language cannot substitute for verification, especially where the explanation will determine the design and cost of remedial action across the institution.
Assess significance in context
I would consider the affected decisions, reach, duration and ability to detect the weakness. Three sampled items can mean very different things. A defect preventing reconstruction of a material lending decision differs from a broken link to an otherwise accessible document. Calling the topic regulatory does not by itself establish the severity of the particular condition, and counting exceptions alone cannot resolve that judgement.
Interactions deserve attention as well. Unclear data ownership might affect reporting and approval evidence simultaneously. Several apparently small findings can therefore point towards a shared remedy. That relationship should be labelled as demonstrated or still hypothetical. A reader needs to distinguish an established impact, a plausible impact pathway and an area not yet examined, instead of receiving an undifferentiated list of alarming possibilities.
Separate correction from lasting change
For the fictional bank, I would distinguish restoring missing records from improving the process. Restoring individual files may recover traceability without ensuring that new cases will be complete. A lasting measure might involve an obligatory link to approval, a check before drawdown or clearer record ownership. The choice must follow the demonstrated mechanism. A mandatory field alone is weak protection if users can enter unverified placeholders.
Before closure, I would define effectiveness evidence. A fresh sample after implementation should establish more than a populated field: relevant authority, approval timing and accessible supporting information all matter. Completing a technical task and reducing the original risk are different outcomes. If a manual check remains necessary, its scope, workload and ownership should stay visible to the people responsible for ordinary operation.
Keep reporting useful to decision makers
I would organise the finding around a clear condition, its evidence, its significance and the next decision required. Specialist details can sit alongside that structure without burying the central issue. Management should be able to identify what must happen now, what requires further investigation and which conclusion is not yet supported. A precise limitation is more useful than a broad assurance that later proves unrepeatable.
A follow-up date should reflect the process cycle. Evidence gathered before new cases have passed through an amended control cannot show sustained operation. Conversely, waiting indefinitely for perfect information can leave known weaknesses unmanaged. I would propose interim protection and a later effectiveness review, explicitly distinguishing temporary containment from the evidence needed to close the underlying issue.
Summary and recommendations
I would use public SREP results as context for better questions. An individual finding still needs a chain from rule to evidence, scope, cause and remedy. Capital figures need explicit dates and categories. The Czech publication of February 2026 is a documented starting point, but it does not decide the quality of an arbitrary internal process or support extrapolating aggregate results to every bank.
I recommend avoiding an automatic translation from a finding to a capital percentage. That requires a broader supervisory assessment rather than a simple calculator. My proposed analytical contribution is a precise, reviewable account of what is known, what remains uncertain and how a measure will be shown to change actual practice. That account gives management and specialists a shared factual basis for their different decisions.