Practical guides · Practical guide

Validation test deviations: recording the finding, correcting it and retesting

A failed test followed by a configuration fix and a successful rerun leaves the original cause and impact unresolved. Follow a decimal error in a human medicinal product GMP test: the literal input “1,25” becomes 125. Keep the original evidence, investigation, effect on other tests and closure decision linked.

Table of contents

At a glance

How do you handle a deviation during validation testing?

Preserve the actual test result and the version and conditions under which the test ran. Compare the finding with the approved criterion, assess the affected processes and investigate the cause. Implement the justified correction under control. Define the retest and necessary regression checks. Close the finding on assessed results with clear ownership, retaining the original run.

  • A subsequent test result supersedes the original finding and does not negate its investigation.
  • Software, configuration, test data, test script, and environment may have different root causes.
  • Post-test verifies the correction; regression evaluates the affected additional functions with valid rationale.
  • Final assessment, transition to the next test stage, and productive use are distinct decisions.

Describe the deviation before assessing its significance

A test deviation arises when the observed behaviour falls outside the specified criterion or when the planned test could not produce meaningful evidence. Describe the difference first. “Import incorrect” gives little to investigate. “The source file contains ‘1,25’, while the target record shows 125 for the same sample ID and unit” identifies what to follow.

For computerised GMP processes, Annex 11 Section 4.2 addresses the documentation of validation deviations. Complementarily, Annex 15, within its qualification and validation scope, covers the investigation of unmet criteria. The current EudraLex registry references Annex 11 as amended in January 2011 and Annex 15 from 2015. Draft versions are not considered enforceable as a compliance obligation.

Not every software testing error automatically constitutes a laboratory out-of-specification (OOS), a critical operational incident, or a standalone CAPA action. The key factors are the nature of the issue and its impact. The FDA's 2022 OOS Guidance, applicable to the US context for chemical-based pharmaceutical laboratory testing, does not equate a hypothetical test dataset within an isolated CSV environment. Actual laboratory results or previously used production data may require additional evaluation and handling as needed.

Sources: [2] [3] [1] [6]

Protect the first run before making any changes

Maintain contemporaneous records of test identity, step, expected result, and observed result. Preserve the relevant input file, target state, and available protocols. Add user role, execution timestamp, software version, and configuration. If the display alone is insufficient, also secure the underlying relevant documentation. Distinguish between technical result data and purely visual representations.

Avoid spontaneous changes that may alter the system state without detection. A repeated import could overwrite existing data. A system restart might result in the loss of transient diagnostic information. Agree on limited technical safeguards and subsequent actions. The investigation should not compromise the real operating environment; a suitable test environment is often the appropriate setting for targeted reproduction efforts.

Later additions shall be traceable. A previous "failed" status must not be overridden to create the appearance of consistent success. A screenshot must include sufficient context to establish its relevance. Archiving a document does not imply that any arbitrary log file must be retained permanently. Define the scope relevant to investigations and traceability.

Sources: [4] [2]

Assess the impact and temporary containment separately from the root cause.

You can often narrow down the impact before the root cause is identified. Check whether the same processing is used for the same requirements, functions, and data. Does the affected interface exclusively process test data? Has the same configuration already been used in a production environment? These questions determine whether only a test segment is impacted or whether additional responsible parties need to be involved.

State a specific interim action. "Caution during import" is not a reliable restriction. In the example, the relevant import in the test environment could be locked for further technical evaluation, while independent testing continues. For a production process, a separate decision would need to be made regarding which data to retain and which previous results to investigate.

Justify the classification based on potential impact and existing evidence. The pressure to meet a milestone does not imply any meaning in the process. Similarly, a minor visible display error should not be automatically classified as a critical data error without proper investigation. If the scope remains unknown, document this uncertainty and assign clarification to the responsible party.

Sources: [5] [2]

Distinguish test errors, configuration, and software root causes

Investigate competing explanations rather than recording the first assumption as the root cause. For decimal values, consider at least one of the following: a defective source file, an inappropriate format setting, a incorrect field mapping, or an error in the expected test value. Compare the unaltered source with the intermediate steps and the target. This enables identification of the first point at which meaning is lost.

A targeted reproduction exercise deliberately alters only the condition relevant to the hypothesis. Document it as a test run and distinguish it from the subsequent released follow-up test. If only the surface displays "125" while the stored value is correct, a different assessment of impact arises compared to a case where the actual value has been genuinely changed.

The statement "Operational errors" does not end the investigation by itself. Check whether guidance, roles, input control or test preparation have favored the error. If a clear cause is not detectable, document this status and the remaining hypotheses. A non-reproducible result will not automatically be meaningless; further measures are based on its possible impact and the available evidence.

Sources: [5] [3]

Worked deviation record for a decimal-comma import that changes 1.25 to 125

This example describes a fictional interface in an isolated test environment. A semicolon-delimited file must transfer the result “1,25” mg/mL, numerically 1.25 mg/mL, unchanged. Requirement R-08 allows a decimal comma in this field. It prohibits thousands separators and empty mandatory fields. The sample ID and unit must be preserved.

In this example the investigation identifies a normalisation rule in mapping M-2. It removes the comma before the application parses the number, multiplying the value by 100. The cause in this example is a configuration error. Seeing 125 on screen in a real system would not establish that cause.

The sheet shows the entire process path with assumed results. The identifiers are placeholders. In your project, actual documents, persons, dates and decisions would be entered. The result is only valid for the described import scope and does not anticipate general system or productive release.

Example: Find B-12 with completed action sequence
StepEntry in the example
Finding and applicable version T-14, step 3, application V2, mapping M-2 and the import test role. Expected 1.25 mg/mL from the literal input “1,25”. Observed 125 mg/mL.
Original record Original source file Q-14, target extract Z-14, and processing trace L-14 assigned to the first run.
Impact Factor 100 in the result value; meaning altered. Example exclusively in test environment, no production data affected.
Provisional Limitation Do not mark the import function within the test scope as completed; continue independent testing separately.
Cause Accepted diagnostic record D-12 shows removal of the comma by M-2; source and required value are correct.
Correction Change C-12 replaces blanket character removal with validation against the agreed decimal-comma format. The new mapping version is M-3.
Scope of evidence T-14-R1 repeats the original case; REG-12 examines small values, whole numbers, blank entries, and invalid format.
Approved conclusion T-14-R1 and REG-12 meet the criteria defined below; specialist review confirms R-08 for M-3. The original report remains valid.
Open item and decision authority No open action is assumed for this example scope; named responsible reviewers document the decision. Other import profiles remain outside the scope.

Sources: [2] [3] [4] [5]

A correction is performed under control, without making the required adjustment.

Link the correction to the finding and the updated status. In the example, the configuration changes from M-2 to M-3. Document the relevant rule and the environments in which it is applied. If a shared component has been modified, also verify other import profiles that use this component.

A supplier notification marked "resolved" serves as an entry point for evaluation. Clarify version, affected function, prerequisite, and existing evidence. Some corrections are customer-specific, while others form part of a general update. Only through the correct assignment and required assessment in the intended usage environment can a professional conclusion be reached. The customer is not required to blindly reperform each appropriate supplier test.

Even a defective test script can be corrected. Ensure that the incorrect expectation and the justified change are clearly documented and traceable. Do not retrospectively alter acceptance criteria to justify an observed value. A legitimate "should" change must be substantiated by technical rationale, impact assessment, and controlled release; it constitutes an independent decision within the result evaluation process.

Sources: [2] [3]

Plan after-testing and regression with different questions

The retrospective test answers: Has the original defect been resolved in the corrected state? The regression test asks: Has the change impacted other relevant processes? In the example, the new format validation addresses the interpretation of numeric inputs. Therefore, defining various acceptable values and specified error conditions is appropriate. A full reexecution of all role, archive, and print tests does not follow automatically.

Re-run the original case with unchanged statement using a controlled copy of the original input and a known initial state. For any differences in test environment or data, assess whether the comparison remains valid. Check not only the visible value but also the assignment, unit, and stored value, to the extent that these determine the acceptance criteria.

These cases apply to the example interface and its stated format. Other interfaces may legitimately accept decimal points or thousands separators. Define the expected result against your own specification. An empty input must never silently become a valid zero.

Example: Post-test and regression selection for mapping M-3
Test inputExpectation in the exampleReason for testing
“1,25” mg/mL Numeric target value 1.25, with the sample ID and unit unchanged. Retest the original finding.
“0,05” mg/mL The target is 0.05. Neither 5 nor 0 is correct. Small decimal values and a leading zero.
12 mg/mL Target value 12 without artificial decimal shift. A permitted integer input is preserved.
Empty mandatory field Reject this data set; no implied zero or partial release. Defined error handling following new format validation.
Literal “1.25” or “1.250,00” in the numeric field Upon the agreed example contract, reject and clearly label. Do not ambiguously normalize invalid format variants.

Sources: [2] [5]

Complete the results only on the basis of evaluated results

Summarize the root cause identified, the corrected status, and the evidence supporting the decision. Identify the affected requirements and relevant dependencies. If results are inconsistent, a professional clarification is still required. A green test overview alone does not confirm that the correct criterion was evaluated.

Distinguish a closed finding from an outstanding action or limited use. In example B-12, all defined tests are assumed to pass and no residual actions remain for that scope. If retesting is still pending in a real project, record the affected evidence as open. A due date cannot establish closure.

Annex 15 §2.10 describes the authorized transition to the next qualification or validation stage under the conditions mentioned there. This does not result in a general permission to start productively with critical open points. Even a technically completed import certificate does not automatically confirm the entire operational readiness. Keep the respective decision and its scope apart.

Sources: [3] [2]

Maintain feasibility of QA, department, IT, and supplier processing activities

A person should keep the findings together. The tester ensures observations and answers questions about the process. Department and IT clarify meaning and technical reach. The supplier can provide diagnosis and correction. Whoever approves a measure or the completion depends on the tasks and powers in your quality system, not on a universal signature list.

A dedicated CAPA may be appropriate or required for systemic root causes or significant implications, as per your procedure. Not every typing error in a test script necessitates a separate major process. Conversely, repeated data processing errors should not be treated as isolated incidents if the same root cause affects additional processes. Justify investigation depth and escalation pathway.

For efficient processing, precise evidence is more effective than numerous follow-up queries. Provide the technical point of contact with the relevant incoming data, expected output, version status, and a reproducible procedure. Separate investigation runs, formal post-test evaluations, and review decisions. The actual effort primarily depends on scope, reproducibility, shared logic, and available comparative data.

Sources: [2] [5] [3]

Checklist for a Traceable Investigation Conclusion

Before closing the deviation, trace the finding from the first run to the final decision. The following questions are our practical guide to the technical review. Follow your established deviation procedure.

  • Expectations and actual observations are specifically described.
  • Original input, results, and relevant execution statuses are secured.
  • Test environment and potential production relevance have been clarified.
  • Impacts, provisional limitations, and remaining uncertainties are evident.
  • Alleged causes are supported by appropriate investigation results.
  • Corrected and amended versions are controlled and documented.
  • The post-test and additional regression scope is technically justified.
  • The actual results are evaluated and linked to requirements.
  • Open actions and limitations are not obscured by the closure.
  • The authorized decision remains clearly linked to the original finding.

Sources: [2] [3] [4]

Project experience and collaboration

Document the rationale for closing test findings

Start a failed test investigation by documenting what happened. With your team, we identify the affected function, information that must be retained and any other results that may be affected. An isolated fault may need a targeted correction and retest. Signs of a broader problem prompt a fresh assessment of requirements and test strategy. The responsible people decide on follow-up and closure. A later successful repeat does not resolve the original finding on its own.

Frequently asked questions about Deviations during validation testing

Is every failed software test an out-of-specification?

No. Assess a CSV test finding against its own acceptance criterion first. The FDA OOS guidance discussed here concerns chemical laboratory testing of pharmaceuticals within its scope. If real laboratory results or product decisions are affected, also apply the relevant investigation procedure. A fictional software test result is not automatically a laboratory OOS.

Sources: [6] [2]

Can we perform the test again immediately?

First, preserve the original state and assess whether a repeat run modifies records or overwrites data. A targeted investigation run may then be appropriate. Clearly define its purpose and conditions. Do not repeatedly perform the test until a green result appears. The formal retrospective testing requires a known corrected state and a traceable link to the findings.

Sources: [4] [3]

What constitutes at least a valid test result?

Describe the test step, expected and observed behavior, as well as the relevant data, configuration, and execution status. Assign original documentation to specific activities and document potential impacts. Later supplement with investigation, correction, validation scope, and decision. The exact scope depends on the findings. A screenshot without context or a ticket marked "not applicable" is often insufficient for a credible technical assessment.

Sources: [2] [4]

What is the difference between post-test and regression?

The retest evaluates the originally failed case in the corrected state. Regression analysis examines additional relevant functions that could be impacted by the change. A new number formatting validation may affect integers, small decimal values, and error handling. Select additional cases based on actual dependencies. A complete reexecution of all tests or reliance solely on the original error case is not always sufficient.

Sources: [2] [5]

Can a incorrect acceptance criterion be corrected retrospectively?

Yes, if the change is professionally justified and controlled. Retain the original criterion and the observed finding. Assess whether the requirement, risk analysis, or any other tests are affected. A new expectation must not merely be tailored to fit an adverse result. Document the valid current status and determine the subsequent testing requirements.

Sources: [3] [4]

When does a test deviation require additional CAPA?

An additional CAPA may be considered, particularly when the investigation identifies systemic causes, recurring patterns, or potential impacts on other processes. Decide in accordance with your quality system. An isolated, traceable correction of a typing error does not necessarily require a separate CAPA process. Document the rationale such that no systemic root cause remains unaddressed.

Sources: [2] [5]

Does the supplier's confirmation that the defect has been corrected suffice?

She is a key entry point, but her actual scope of validation must be clearly demonstrable. Review the corrected status, the affected function, and the relevant test documentation. Existing evidence may be utilized where appropriate. Customer-specific configurations and remaining questions may require additional evaluation. A general ticket response without reference to version and results does not automatically constitute a complete professional assessment.

Sources: [2] [3]

Must we reject the entire test run as invalid?

Not automatically. Clarify which specific steps and outcomes are actually affected. A local formatting error may invalidate individual pieces of evidence; an inappropriate shared environment could impact a broader scope. Ensure that valid, restricted, and re-requirement-based portions are distinguishable. Do not delete or overwrite original records merely to achieve a clean overall overview.

Sources: [2] [5] [4]

Can we proceed to the next test stage with open deviations?

A conditional transition may be permissible. According to Annex 15, Section 2.10, the next qualification level may be advanced despite open acceptance criteria or deviations, provided a documented assessment demonstrates that the subsequent activity is not materially affected. The responsible parties must authorize the transition. This is not a blanket release. Identify open items, boundaries, and prerequisites; if a finding impacts the next inspection, a subsequent correction date is insufficient.

Sources: [3]

What do we do when the error is not reproducible?

Obtain the original result and investigate differences regarding data, timing, environment, and operator procedure. Assess existing logs and potential scope of impact. A single successful repeat run does not invalidate the previous observation. Document remaining uncertainties and justify additional investigation, monitoring, or usage limitations. The conclusion must reflect the appropriate level of significance, not an invented singular root cause.

Sources: [5] [2]

Sources and review date

Primary sources were opened on October 6, 2026. The focus is on computerized system validation within human medicinal product GMP. The FDA source is used exclusively for limited differentiation from laboratory out-of-specification results. Test result sheets, identifiers, and test outcomes are examples; they do not indicate any identified product defects.

  1. European Commission: EudraLex Volume 4 – Human Medicinal Products-GMP
  2. EU-GMP Annex 11: Computerised Systems – Validation Reports, Modifications and Incidents
  3. EU GMP Annex 15: Qualification and Validation – §§2.7–2.10
  4. EU-GMP Chapter 4: Documentation – Timely Records and Detectable Changes
  5. ICH Q9(R1): Quality Risk Management – Assessment and Reassessment
  6. FDA: Investigating Out-of-Specification Test Results for Pharmaceutical Production, May 2022 – Definition and Scope

Sources checked: .

Daniel Herrmann, Daniel Herrmann Consulting

Daniel Herrmann Consulting

Test results support a traceable final release decision

Daniel Herrmann Consulting supports QA, IT, and functional departments in the evaluation of test deviations, determining the appropriate extent of evidence, and establishing a clear link between corrective actions, results, and decisions.

  • Computer system validation →

    Bring your GxP systems into validated operation, from laboratory software, MES and eQMS to ERP. We define the scope, plan risk-based tests using GAMP 5 and hand the evidence to QA for release.

Send a project enquiry Customer testimonials →