Regulations and compliance · Practical guide

Part 11: detecting invalid or altered records

A readable record can support the wrong conclusion. A documented correction can be legitimate. To test whether a system can detect invalid or altered records under §11.10(a), define specific failure conditions and determine the expected outcome independently. An unusual result warrants assessment before anyone concludes that manipulation or an out-of-specification result occurred.

Table of contents

At a glance

How do you test whether a system detects invalid or altered records under Part 11?

Section 11.10(a) requires validation of systems for accuracy, reliability, consistent intended performance and the ability to detect invalid or altered records. Derive failure scenarios from the actual data path and test controls against predefined outcomes. FDA has exercised enforcement discretion for this Part 11 validation requirement since 2003. Underlying predicate-rule obligations remain. An unchanged hash does not prove substantive accuracy, and a recorded change is not necessarily unauthorised.

  • Invalidity, documented change, and unusual result differ.
  • Pre-test recognition to be predefined independently.
  • Identify, evaluate, and further use as separate steps.

1. Define the specific detection task

Within Part 11 scope, §11.10(a) addresses controls for closed systems. It calls for the ability to discern invalid or altered records without prescribing one universal technical mechanism. For open systems, consider the applicable conditions in §11.30.

Describe the record under consideration and its use. Should a measurement value be transferred, a calculation be verified, or a previously approved batch be re-opened? Depending on the data pathway, missing information, unexpected changes, incorrect sample, or inappropriate processing may be relevant. A general test line such as "data integrity functions" remains too vague for this context.

Sources: [1]

2. Do not confuse a change with manipulation

An authorized personnel can investigate and correct a deviation in a traceable manner. The record is subsequently amended, but not solely because of this amendment does it become invalid. Other changes may be inadvertent, technically caused, or unauthorized. Detection should enable a reasonable assessment, not presume the root cause without prior investigation.

Similarly, an unusual measurement value is not automatically an invalid record; it may represent a valid result. In the Drug-CGMP laboratory environment, the FDA requires for the invalidation of a test result a documented, scientifically sound justification. A technical alert or an out-of-range numerical value does not substitute for such an investigation.

Sources: [1] [2]

3. Separate technical integrity from meaning in the process

A checksum can detect changes compared to a reference under defined conditions. It does not confirm that the reference was correctly assigned to the intended sample or that the formula used was technically appropriate. Even a formally complete data format may contain an erroneous content.

The following matrix distinguishes these error patterns. The examples assume the presence of relevant electronic laboratory records within the appropriate Drug-CGMP scope; the specific detection pathways must be defined for actual implementation.

Own error image matrix for the considered record
Observed failureWhy pure readability is insufficientAppropriate test question
Reference calculation value unexpectedly changed The file can still be opened correctly. Is the deviation from the expected status identified and attributed?
Unit missing The numerical value is present; its significance is unclear. Does the processing lack the required context?
Incorrect sample identification Format and checksum may be technically correct. Is the assignment verified against the intended order?
Formula uses incorrect factor The calculation pathway may be technically error-free. Does the result align with an independently calculated expectation?
Authorized Correction A change is present, but may be acceptable. Are the initial value, modification, and necessary context traceable?

Sources: [1] [3] [2]

4. Define the reference prior to error entry

Create a controlled output dataset with a unique identifier, expected content, and required context. Document the specific change introduced by each individual test case. Use an appropriate test environment and approved test data; actual production records will not be intentionally modified for this work aid.

The intended observation must be defined prior to testing. It may, for example, be a specific error message, a detectable status, or a logically justifiable rejection of use. It is important that the evidence does not merely repeat the same implementation logic. An independent calculation pathway or a separately prepared mapping list may provide a different perspective on the expected content.

Sources: [1] [4]

5. Review errors and authorized changes jointly

Calculation example: The input value 50.0 is divided by the intended divisor 2.0; the expected result is 25.0. A mistaken divisor of 5.0 would yield 10.0. Both values are formally valid. Professional judgment distinguishes between them. These are computational results without reference to a real product specification.

The matrix describes target behavior, no observed test results. Especially the case of a permissible correction prevents detection from being mistakenly equated with a flat-rate blocking of any change.

Example: controlled intended omissions
Test caseTargeted VariationPre-established expectationLimit of detection
T1: Initial scenario 50.0 / 2.0 25.0 with correct identification and unit. Applies only to the selected normal case.
T2: Incorrect divisor 50.0 / 5.0 10.0 is recognized as a deviation from the intended divisor and result. No general mathematical invalidity of 10.0.
T3: Missing unit Value remains 50.0; mandatory context is missing. Use shall be clarified according to the established procedure prior to a misinterpretation occurring. No specific product function specified.
T4: Authorized Correction Documented correction of an input error. Change detectable; professional assessment remains possible. A change alone is not sufficient as evidence of tampering.
T5: Incorrect assignment Same value under foreign sample identification. The discrepancy from the expected order is detected. A checksum alone would not answer this question.

Sources: [1] [4] [2]

6. Assess the findings prior to further use

An alert is only practically useful when the relevant process and responsibility are clearly defined. Specify who evaluates the report, which record is involved, and how the process is managed until clarification is achieved and the outcome is documented. Not every technical observation requires the same action; however, a relevant uncertainty must not be left unaddressed in the log.

Retain the full context of a deviation. A successful repeat run does not automatically explain why the prior run deviated. In the laboratory setting, complete data and a justified evaluation remain required. A repeat test is not a general substitute for investigating an invalid result.

Sources: [2] [3]

7. Do not conclude that all records are valid based on a single approved case

Maintain documented versions, configurations, data flow, and relevant interfaces. Successful import testing does not automatically validate the integrity of subsequent exports or privileged modifications. Verify whether different pathways actually employ the same control or introduce distinct gaps.

The FDA explicitly references § 11.10(a) and the corresponding requirement in § 11.30 within the validation requirements, and it exercises a policy of non-enforcement since 2003. This does not repeal the validation and record-keeping obligations established under the underlying predicate rules. Therefore, the actual scope of validation evidence must be justified based on the intended use, applicable requirements, and associated risks; this policy neither supports a blanket exemption nor calls for a uniform, comprehensive validation.

Sources: [1] [5] [4]

8. Make the identification test reproducible

For a small team, a concise list of relevant failure modes with clear expected observations represents a practical starting point. The department defines significance and decision logic, while IT outlines technical pathways and existing controls. The responsible reviewers assess whether the evidence supports the respective assertion.

When changes are made to the data model, rules, or interfaces, the affected detection is re-evaluated. The number of test cases is not a general quality metric. What matters is whether the relevant error paths and appropriate responses are adequately covered.

  • Determine applicable record and intended data pathway.
  • Distinguishing invalidity, change and conspicuous result.
  • Controlled reference and independent set of desired specifications.
  • Introduce errors exclusively in an appropriate test environment.
  • Detection, assignment and technical response separately.
  • Make test boundaries and untested paths explicit.
  • Incorporate relevant changes into the evaluation of documentation.

Sources: [1] [4]

Project experience and collaboration

Test detection against predefined failure scenarios

With IT and process owners, we specify which errors a data path must detect, such as a wrong sample identifier, missing unit or unexpectedly changed value. We prepare controlled data with independently defined expected observations. A focused test can demonstrate an effective existing control. A detection gap calls for assessment of additional technical or business controls. We record which error was detected, under which conditions and how your team handles the finding.

Frequently asked questions about Invalid records

What does §11.10(a) require regarding the identification of records?

Section 11.10(a) requires a system validation that ensures, in addition to accuracy, reliability, and consistent performance as intended, the ability to detect invalid or altered records. The paragraph does not prescribe a universal detection mechanism. The specific errors and controls to be demonstrated depend on the relevant record and its data pathway; the FDA's enforcement policy must be considered separately.

Sources: [1]

Is any retrospective change unacceptable?

No. Authorized and traceable corrections may be required. The change must be evaluated in the appropriate context. Detection does not automatically imply deletion or rejection, but rather initial visibility and a well-founded decision regarding the significance of the change.

Sources: [1] [2]

Does an unaltered hash prove technical accuracy?

No. Technical similarity to a reference does not establish that the reference corresponds to the correct sample or uses the correct formula. Integrity and scientific relevance must be assessed separately. The selected technical validation evidence must not be interpreted beyond the scope of its support.

Sources: [1] [3]

Is an outlier value automatically considered out of specification or an invalid record?

No. The classification depends on specifications, data quality, and investigation. In the Drug-CGMP laboratory setting, the invalidation of test results requires a scientifically justified, documented evaluation. A technical alert or an adverse result alone is insufficient for this purpose.

Sources: [2]

May we intentionally corrupt production data to test detection?

Not based on actual required production records. For such error detection checks, we recommend controlled test data within an appropriate test environment. The starting condition, intentionally introduced variation, and expected detection are predefined and documented in advance. The check must not falsify required records or inadvertently disrupt the normal operation.

Sources: [1] [4]

Must every identified deviation trigger an immediate technical shutdown of all systems?

Not mandatory. Section 11.10(a) does not specify a particular lockout reaction. Define the appropriate response for the specific failure and its potential impact. A critical unverified record must not be used without review; in the case of a limited indication, an alternative documented assessment may be appropriate.

Sources: [1] [2]

Does a successful repeat run constitute validation?

Not automatically. It does not necessarily explain the cause of the first out-of-specification result. Obtain the required data and assess the relationship. In particular, repeated review must not be used to remove undesirable data without credible justification.

Sources: [2] [3]

How many error incidents do we need?

The cited sources provide no universal figure. Derive cases from data pathways, relevant failure patterns, and existing controls. More cases are not automatically more indicative if they consistently follow the same simple pathway and omit important variants.

Sources: [5] [4]

Can the Audit Trail perform all identification tasks?

It can make changes and associated information visible, but it does not automatically verify the technical accuracy or incorrect assignment. Assess the actual capability it provides and determine what additional controls are required for the intended use.

Sources: [1] [2]

Does the inspection bypass occur due to FDA Enforcement Discretion?

The FDA explicitly references § 11.10(a) and the corresponding requirement in § 11.30 within the validation requirements, and it exercises a policy of non-enforcement since 2003. This does not repeal the validation and record-keeping obligations established under the underlying predicate rules. Therefore, the actual scope of validation evidence must be justified based on the intended use, applicable requirements, and associated risks; this policy neither supports a blanket exemption nor calls for a uniform, comprehensive validation.

Sources: [5] [1]

Sources and review date

Source date: 7 October 2026. eCFR notice period until 5 October 2026; separate consideration of binding regulatory text and non-binding FDA Guidance. Examples and tables are internal work aids and do not represent completed product testing.

  1. 21 CFR Part 11 – Electronic Records; Electronic SignatureseCFR display until 05.10.2026; §§11.1–11.300.
  2. FDA: Data Integrity and Compliance With Drug CGMP – Questions and AnswersFinal Guidance December 2018; in particular Questions 1–3, 9–12, 14 and 17.
  3. 21 CFR §211.194 – Laboratory recordseCFR notice until 05.10.2026; paragraphs (a)(1)–(8), (b)–(e).
  4. 21 CFR §211.68 – Automatic, mechanical, and electronic equipmenteCFR notice until 05.10.2026; paragraphs (a)–(c).
  5. FDA: Part 11 – Scope and ApplicationFinal guidance. The PDF is dated August 2003 and the status page September 2003. See sections I and III.A–C.

Sources checked: .

Daniel Herrmann, Daniel Herrmann Consulting

Daniel Herrmann Consulting

Jointly develop the specific evidence process

We support you in translating relevant failure patterns into measurable expectations and in evaluating existing controls based on their actual effectiveness. The assessment remains aligned with your intended data and decision pathway.

  • 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 →