Guides · Good documentation practice

Verifying transcribed GMP data: second-person checks that hold up

Copied critical data should be checked for accuracy against the original, by a second person or a validated system, with the depth set by risk. Record who verified which values, when and how; resolve each discrepancy at the original; keep every copied value linked to its source.

When a value is copied from a GMP record into another system, the copy has to be checked against the original, and the check has to be on record. EU and PIC/S guidance and the MHRA expect critical data entered by hand to be verified by a second person or a validated system, with the extent set by risk; the US rule, 21 CFR 211.68(b), requires computer input to be checked for accuracy, to a degree set by the system’s complexity and reliability. They leave the design of the check to you.

That design is where checks fail. A verifier who compares a spreadsheet with a printout of the same spreadsheet has checked nothing. A signature at the bottom of a long file says little about which rows were looked at. This guide sets out what the texts ask and how to build a check that answers an inspector’s questions.

What do the regulations ask of a transcription check?

  • 21 CFR 211.68(b) requires that input to and output from computer or related systems “be checked for accuracy”, with the degree and frequency of verification based on the system’s complexity and reliability. FDA’s 2018 data integrity Q&A (question 1f) reads “computer or related systems” broadly: hardware, software, networks, personnel and associated documents.
  • EU GMP Annex 11 §6: for critical data entered manually there should be an additional accuracy check, “by a second operator or by validated electronic means”, with criticality and consequences covered by risk management.
  • MHRA GXP data integrity guidance §6.11.2: “Where manual transcriptions occur, these should be verified by a second person or validated system.”
  • PIC/S PI 041-1 §9.7 says all manual entries of critical data should be verified by a second operator or validated computerised means, that the system should record who made each entry and when, and that changes should be in the audit trail. §8.8 tells inspectors to view original data where possible “to confirm that the correct data was transcribed for the calculation”. §9.10 asks for procedures and records at the interfaces of hybrid systems, including the manual input of manually generated data.
  • WHO TRS 1033 Annex 4 asks companies to design processes that avoid unnecessary transcription (§4.15). Its examples (Appendix 1, example 3) say an entry should, where appropriate, be verified by a second person or entered by technical means such as barcode scanning, and suggest locking critical entries once verified.

Whether a given spreadsheet or CPV database holds GMP data, and which of its values are critical, is a decision your quality system makes and documents. Once it is made, the expectations above follow.

Full or risk-based verification?

Annex 11 §6 ties the check to risk, 211.68(b) ties it to the system’s complexity and reliability, and PIC/S §8.8 says the need for and extent of a secondary check “should be based on quality risk management principles”. In practice:

  • Critical data (values used for release decisions, critical process parameters, the data behind CPV and the product quality review) is usually verified in full, every value.
  • Lower-risk data may be verified by sample where the quality system justifies it.

If you sample, write the rule down before you start: how many values, chosen how, and what happens when an error is found. A common rule is that any error in the sample triggers a full check of that batch or file. Record which values were in the sample, so the check can be reconstructed later. Some companies verify every transcribed GMP value whatever its risk class, to avoid the classification argument altogether. Both approaches are seen in practice.

What should the verifier compare against?

The original record, or a verified true copy of it. Not the transcriber’s notes, not an earlier export, not a printout of the destination file. A copy of a copy can only confirm that two copies agree.

For each value the verifier confirms:

  • the number itself, digit by digit, with its sign and decimal places;
  • the decimal separator (a comma on a European record is not a thousands separator);
  • the unit, and any conversion applied;
  • that the value sits in the right row: right batch, right step, right time point;
  • dates and times in the agreed format, with day and month not swapped;
  • that a corrected entry was copied as corrected, and a voided field was not copied as zero or left looking like missing data.

Calculations built on the copied values need their own check; PIC/S §8.8 expects secondary reviews to include “a verification of any calculations used”. The two mistakes that most often get through, decimals and dates, have their own guide.

Visual check or blind re-entry?

Visual check. The verifier reads the copy against the original and marks each value as checked. It is quick and needs no tooling. Its weakness is that the verifier sees the first person’s value before reading the original, and with ambiguous handwriting can read the expected value into the ink.

Blind re-entry (double data entry). A second person types the values again without seeing the first entry, and the two sets are compared automatically. Every difference goes back to the original to be resolved. This avoids the anchoring problem, costs a second round of typing, and still misses a value both people misread the same way. None of the GMP texts names double data entry; it is one way to perform the second-person verification they ask for, widely used in clinical data management.

Validated system. Barcode capture, an instrument interface or software that does the comparison can take the place of a second person, as Annex 11, MHRA and PIC/S all allow, if it is validated for that use. PIC/S §9.10 expressly lists “automated detection and transcription of printed data into computerised systems” among the hybrid-system interfaces that need procedures and records.

Many teams mix these: blind re-entry for critical values, visual checks for the rest. The choice belongs in the procedure, not in each verifier’s habits.

How should the check be recorded?

An inspector should be able to answer, from the record alone: who verified which values, against which original, when, by what method, and with what result.

  • Who: a named individual. FDA’s Q&A (question 5) says actions such as second person review must be attributable to a specific individual; shared logins do not meet this.
  • What: the scope, per value or per defined set (batch, page, file version). A single signature on a file should say exactly what it covers.
  • When: the date of verification, separate from the date of the event recorded.
  • How: visual, blind re-entry, sample (with the sample listed) or validated system.
  • Result: MHRA §6.15 asks that a documented data review include “a positive statement regarding whether issues were found or not”, the date and the reviewer’s signature.

In an electronic destination, the system’s audit trail should capture entries and changes (PIC/S §9.7). In a spreadsheet without one, this information has to be kept by procedure: a verification column per value, a locked version, a signed checklist. Your quality system decides which.

What should happen to a discrepancy?

  1. Go back to the original, not to the other copy.
  2. If the copy is wrong, correct it in a way that shows the old value, the new one, who changed it, when and why. Do not overwrite silently.
  3. If the original itself is unclear, incomplete or wrong, the person copying does not settle it. It goes to the record’s owner under the site’s procedure for corrections, and the copy waits or is marked as unresolved.
  4. Count discrepancies. A rising rate on one form or from one source points to a cause worth fixing, such as a cramped field, an ambiguous unit or a column order that differs from the page.

MHRA §6.15 asks for a procedure describing what to do when data review finds an error or omission, keeping the original visible and the correction traceable.

Give every copied value an address: product, batch, document identifier and version, page, and the field or step it came from. If the original is scanned, record which scan file, and keep that file unchanged. With the address, anyone can go from a point on a control chart back to the ink, which is what an investigation or an inspector will ask for. ALCOA+ on paper explains why the batch record stays the original even when everyone works from the copy, and batch records to CPV data follows the values into trending.

A checklist for designing the check

  • Decide and document which destinations hold GMP data and which values are critical.
  • Write down full or sampled verification for each class, with the sampling rule and the escalation rule.
  • Name the original the verifier compares against, and make it available to them.
  • Choose the method per class: visual, blind re-entry or validated system.
  • Record who, what, when, how and the result, per value or per defined set.
  • Route discrepancies back to the original; correct copies traceably; escalate unclear originals.
  • Give every copied value an address back to its page.
  • Review discrepancy counts periodically and fix their causes.

Sources

  1. 21 CFR 211.68 Automatic, mechanical, and electronic equipment, accessed 18 November 2026.
  2. EudraLex Volume 4, Annex 11: Computerised Systems, accessed 18 November 2026.
  3. MHRA 'GXP' Data Integrity Guidance and Definitions, Revision 1 (March 2018), accessed 18 November 2026.
  4. PIC/S PI 041-1 Good Practices for Data Management and Integrity in Regulated GMP/GDP Environments (1 July 2021), accessed 18 November 2026.
  5. WHO TRS 1033, Annex 4: Guideline on data integrity (2021), accessed 18 November 2026.
  6. FDA, Data Integrity and Compliance With Drug CGMP: Questions and Answers, Guidance for Industry (December 2018), accessed 18 November 2026.

General information, not regulatory advice. Check the current text of each regulation; your quality system decides how it applies to you.

Related guides