Can a correction be undone?

Every organisation that keeps records corrects them, all the time. A surname is misspelt at a counter. A date of birth arrives as 31 June. The same company is entered twice, once in full and once in short. Nobody argues about whether such errors should be fixed. The real question is what happens to the value that was wrong.

Correcting is not erasing

The quickest fix is to type over the mistake. The next morning the record looks clean, and the old value has gone. Yet that old value was not worthless. It may explain why a letter never arrived, why two files were never matched, or why a decision was taken on the facts known at the time. Wipe it out and the record becomes tidy, and silent.

Good bookkeeping does not work like that. A wrong entry is not scratched out: a new entry cancels it and says why. The accounts end up right, and the mistake can still be read. We think data deserves the same care. A clean value whose history has been lost is harder to trust than a flawed one that can be explained.

Figure 1

Correct the record

Interactive

Customer record 0417

Three errors, as received

• Surname, first nameMARTN, ClaireMARTN, Claire
  As received
• Date of birth31 June 198431 June 1984
  As received
• CityParsi, 75011Parsi, 75011
  As received

History of changes

Nothing has changed yet.

Try itCorrect the record and watch what happens to the old values.

Correct the record

The record as it arrived: a misspelt surname, a date that does not exist, two letters swapped in the city.

What some rules already say

Some regulators have written this down. The US Food and Drug Administration, in its rule on electronic records, asks for a dated history of every entry and action that creates, changes or deletes a record. It adds one plain sentence: “Record changes shall not obscure previously recorded information.” In Europe, the guidelines for computer systems used to make medicines say that, for any change or deletion of data that matters to the quality of a medicine, “the reason should be documented”.

Both texts come from health rules, where a wrong value can harm a patient. Their reasoning holds elsewhere. A civil register, a customer file or the maintenance history of a turbine raises the same two questions each time a value changes: what did it say before, and why was it judged wrong?

The people who know the records

A platform is good at finding values that may need correcting: two spellings of one name, a date no calendar has, two letters in the wrong order. Deciding that a value really is wrong is another matter. The people who keep a register know its habits, its abbreviations and its exceptions. They will spot a correction that looks right but is not.

That is why, as question 2 explains, your own teams review the first corrections before any rule is applied to a whole register.

What Datasight keeps

In Datasight, every correction is stored together with the value it replaced and the rule behind it. Whoever reviews the record can see what was changed, and why, and put the earlier value back in a single step. Correcting a record never means losing what it said before.

How to check it

1. Pick a record that has been corrected, and ask to see what it said before and why it was changed.
2. Have that correction undone in front of you, and check that the earlier value comes back exactly as it was.
3. Ask to see the full history of a record that has been corrected more than once.

Next on the record · #8Who sees what, and who decides?

Sources

1. 21 CFR § 11.10(e), “Controls for closed systems”, Part 11, Electronic Records; Electronic Signatures, United States Food and Drug Administration. Cornell Law School
2. European Commission, EudraLex Volume 4, Good Manufacturing Practice, Annex 11: Computerised Systems, section 9, “Audit Trails”, in operation since 30 June 2011. European Commission