These analytical notes explore a field I want to develop in. The examples are hypothetical and do not describe inspections I have carried out.
Preserve context
A conclusion table is insufficient if its source export is unknown. Record source, date, version, field meanings and filters. Sensitive information also requires compliance with access and handling rules.
Separate facts from interpretation
A document may establish that approval exists without proving the quality of the assessment. I would distinguish documented facts, the bank’s explanation and the analyst’s conclusion. Readers can then see what is directly verifiable and what involves judgement.
Hypothetical example
Two team members obtain different case counts because exports were taken at different times and data were corrected. Identifying both versions explains the difference without silently overwriting the earlier result. The conclusion should specify which version supports it.
Writing for another reader
A useful working record lets a colleague repeat the essential step without an oral explanation. Include the source, scope, transformation and link to the result. Not every technical detail is necessary, but omitting a decisive filter can change the meaning entirely.
Record what could not be verified as well. Missing evidence may limit the conclusion or require another request. An open-items log helps distinguish completed analysis from work awaiting information.
Versions are part of the explanation
For differing exports, compare changed records and correction reasons as well as counts. If source data changed, explain whether the finding changed too. A new version may correct current data while confirming that an earlier report used an incorrect input.
A strong evidence trail enables substantive discussion: the bank can respond to a particular fact and colleagues can examine the method. Documentation is therefore part of analytical quality, not an administrative afterthought.
Questions for further review
Which records support the conclusion? What alternative explanation remains possible? What additional evidence would change the assessment? These questions help keep the analysis open to challenge and its scope clear.
A real case: a retrospective needs evidence
In December 2022, the UK FCA and PRA imposed combined penalties of GBP 48.65 million on TSB for failings connected with its April 2018 IT migration. The published explanation distinguishes successful data migration from subsequent platform failures. That distinction matters when constructing an evidence trail.
I do not infer undocumented details about the bank’s records beyond the published findings. My own question is whether our working papers could distinguish a partial success from the result of the whole process. “Migration completed” might mean transferred files, reconciled balances or an available customer service. Without a defined measure, the same statement can support several different interpretations and cannot settle a disagreement.
A hypothetical example: two extracts, two defensible totals
Imagine a portfolio review. Monday’s extract contains 10,000 contracts. By Wednesday, twenty records have been added and ten historical items corrected. A second analyst uses the newer extract and obtains a different exception count. The disagreement does not automatically mean one calculation is wrong. First we need to know whether the analysts examined the same population at the same reference point.
I would assign each extract an identifier, creation time, scope and description of its filters. Corrected items also need a reason and effective date. I would distinguish “information known on Monday” from “a corrected position relating to Monday.” Both can be useful, but they answer different questions. Silently treating them as equivalent would make the outcome harder to explain and could obscure an earlier reporting problem.
A comparison table would identify added, removed and changed records. I would then apply the same rules to both versions, separating data changes from method changes. If somebody also modified the query, its version needs to be retained. Otherwise the team could attribute apparent improvement to corrected data when the test has merely changed which situations it identifies as exceptions.
Linking a claim to its evidence
Every important claim would reference the relevant evidence and analytical step. The main conclusion need not contain a full extract or every line of code. However, a reader should be able to establish where a number came from and how it was derived. They should also see whether a statement is direct observation, the institution’s explanation or my interpretation connecting several pieces of evidence.
I would identify information that could contradict or materially change the conclusion. A located approval may not be relevant if it was created after the event under review. Conversely, an absent signature on one working copy does not prove that approval is missing from the original file. I would request the appropriate evidence and avoid strengthening the wording beyond what the available material can actually support.
Access, retention and handover
An evidence trail should not create uncontrolled copies of all sensitive information. I would use approved storage, proportionate permissions and links rather than repeatedly circulating attachments. A working example needs only the relevant information. Anonymisation should preserve analytical meaning while the authorised location of the original remains identifiable. Otherwise a privacy measure can inadvertently remove the very field required to check the conclusion.
Before handover I would ask a colleague to follow the record without my verbal explanation. Which filter is missing? Is the reference date clear? Can they distinguish original and corrected positions? This modest usability check can expose weaknesses more effectively than adding further pages. Documentation is sufficient when it enables the important steps to be understood and repeated, not merely when it is lengthy.
An open point needs an owner
For each missing item I would identify who is following it up, what information is required and which conclusion remains limited. When evidence arrives, the record should show whether it confirms, contradicts or refines the earlier assessment. Closing an administrative task and substantiating an analytical conclusion are different steps. Making that distinction visible improves both handover and progress reporting.
Summary and recommendations
I would maintain an explicit connection between question, population, method, evidence and conclusion. Versions and unresolved issues belong in that explanation rather than being treated as awkward additions. A usable record permits an informed response from the reviewed institution and independent examination by a colleague. That ability to challenge and reproduce the result makes documentation support accountable decisions instead of serving only as an archive of an already issued finding.
Sources and links
- FCA — TSB operational resilience failings, 20 December 2022
- ECB — Guide to on-site inspections and internal model investigations, September 2018
Foreign cases provide analytical lessons; their legal frameworks and measures are not automatically applicable to Czech banks. Hypothetical examples and proposed methods do not describe inspections I personally conducted.