A person who was detained files a complaint months after the arrest. They say they were held for more hours than the law allows. The legal department opens the record in the system and finds the booking time: 23:55. With that time, the numbers add up.
The problem is that the officer who made the arrest remembers it differently. And the paper logbook at the booking desk says 22:40. Someone changed the time in the system, but the system only keeps the current value. It does not say who changed it, when, or why. It may have been a data entry error that someone fixed in good faith. It may have been something else. With what is available, nobody can prove either one.
The scene has smaller versions every day. A case changes department and nobody knows who reassigned it. An offense is reclassified and last month's dashboard no longer matches today's. In every case, the system saved the new value on top of the old one.
Correcting data is not just allowed, it is an obligation. But a correction is not an overwrite. Correcting means changing a value and leaving a record that it changed. Overwriting means changing it and erasing the evidence. The difference is invisible on the day of the change. It shows up when someone asks.
Mexico's General Law on the Protection of Personal Data Held by Obligated Entities (LGPDPPSO), published in March 2025, applies to authorities at all three levels of government, including municipalities. Its Article 17 requires the data controller to take the necessary measures to keep personal data "accurate, complete, correct and up to date", with an explicit purpose: that their truthfulness is not altered.
The same law recognizes, in Article 39, the right of the data subject to request the rectification of their data when it is inaccurate, incomplete or out of date. The law does not just allow corrections: it requires them.
But the law also defines what counts as a security breach. Article 32 includes, among others, "damage, alteration or unauthorized modification". A system where any user can change a value without leaving a trace cannot tell a legitimate rectification from an unauthorized modification. Both look the same: a new value where there used to be another one.
Mexico's General Archives Law completes the picture. Its Article 5 sets out the principle of integrity: ensuring that archival documents "are complete and truthful so as to accurately reflect the information they contain". It defines an archival document as one that records a fact or act produced in the exercise of the institution's functions, "regardless of its documentary medium". A row in a database, an electronic booking record or a citizen service ticket all fit that definition.
Article 46 of that law is even more specific. When it sets the basis for automated document management systems, it requires, at a minimum, "establishing the procedures to record the traceability of update, backup or any other process that affects the content of electronic archival documents". The word that matters is update. The law is not satisfied with the document being right today. It asks to know how it got that way.
Read together, the two laws say something simple: the data must be correct and its history must be verifiable. An institution that only meets the first half has clean data it cannot defend.
Saying "the system has an audit log" is not enough. Many logs record that someone signed in, but not what changed, or they say a record "was modified" without saying which field or what value it had before.
The NIST catalog of security and privacy controls, publication SP 800-53, used by United States federal agencies, offers a useful reference. Its AU-3 control requires that every audit record establish what type of event occurred, when, where, its source, its outcome and the identity of the individuals or entities associated with it. Mexican municipalities are not bound by that standard, but its logic works as a design baseline.
Applied to the correction of operational data, that baseline translates into seven elements:
- Which record and which field. Not "the booking record was modified", but "the booking time of record 4512 was modified".
- Previous value. Without it there is no way to reconstruct what the system said before.
- New value. What the system says now.
- Who. An identified person, not a generic account shared by ten operators.
- When. The system's date and time, not one the user types in.
- Why. A short reason: data entry error, rectification request, supervisor's decision, duplicate merge.
- Under what authorization. If the change required approval, who gave it.
The same catalog adds a point that is often forgotten. Its AU-9 control requires protecting audit information from unauthorized access, modification and deletion. A log that the administrator can edit proves nothing. The history of changes must be better protected than the data it describes.
There is also a limit. NIST itself warns that audit records can reveal personal information. Keeping the previous value of a sensitive field means that value still exists. That is why change history needs its own permissions: someone who can see a case file should not necessarily be able to see every previous version of every field.
Where history gets lost in daily operations
Overwriting is rarely a decision. It is almost always the default behavior of a system nobody designed to remember. There are four moments where it happens most often.
The most common case. An operator opens a record, changes a value and saves. The system updates the row in the database and the previous value disappears. If someone asks, the only source is the memory of whoever changed it.
When the same person appears three times in the system, with the name spelled three different ways, someone decides to clean up. They delete two records and keep one. The problem is that each deleted record had its own history: arrests, tickets, addresses. If the cleanup does not keep track of where each piece of data came from, the surviving record looks more complete than it is, and nobody can check whether the merge was right. A proper merge records which records were combined, on what criteria and who approved it.
A crash report, a medical certificate or a ruling goes through review and approval. If anyone can edit it after sign-off, the approval no longer means anything: it approved a text that may no longer exist. What has been approved must stay closed. If it needs correcting, the correction should be a new act, visible and with its own owner.
A migration, an import or a script that "normalizes" neighborhood names can change thousands of records in seconds. These are legitimate changes, but they tend to fall outside the audit log because no user made them from the screen. Article 46 of the General Archives Law speaks of "any other process that affects the content". An automated process is still a process.
Consider a monthly dashboard of administrative offenses. In October it reports 312 cases of disturbing the peace. In November, someone reclassifies forty of those cases as brawls. If the system overwrites, October's dashboard changes after the fact and the report delivered to the city council no longer matches the screen. In an earlier post we explained why a number is not an indicator when its definition changes quietly. The problem here is similar but harder to see: the definition did not change, the underlying data did.
With change history, the institution can answer two different questions: what October's data says today and what it said on the day the report was delivered. Both are legitimate. Without history, it can only answer the first.
Something similar happens with artificial intelligence. A model that suggests classifications, detects duplicates or drafts narratives learns and is evaluated with operational data. If human corrections are saved on top of the suggestions, the institution loses its best source for knowing where the model gets things wrong. Every recorded correction, with its previous value and its reason, is an example of something the system got wrong and a person fixed. As we explained in a suggestion is not a decision, that difference between what the system proposed and what the person decided has to stay visible.
Change history cannot depend on the discipline of each operator. It has to be built into the system's design.
In Agora, Intello's citizen service platform, each ticket's detail view includes a log of every change with who made it, when, and the value before and after. If a ticket changes status, priority or department, the change stays in its history, and when the ticket moves to In progress or Resolved the resident is notified by WhatsApp. Analytics include reopenings, which show how many tickets were marked as resolved and had to be opened again.
In Tribuna, the platform for public safety and municipal civic justice, the audit log records reads, writes and risk events, and every case file has its complete timeline. Crash reports go through review: the supervisor approves them or returns them with comments for correction, and the approved report is locked. Duplicate detection by CURP, date of birth, address and phone number helps keep a single identity per person, and record merging is reserved for high-confidence cases. INTELLO AI suggests narrative drafts and offense classifications, but the decision belongs to the person.
No platform decides for the institution who can correct which data or under what authorization. That decision belongs to each department, its transparency unit and its records management office. What a well-designed system does guarantee is that, once that decision is made, every change leaves a trace.
A quick review can be done with any record and five questions:
- If I change a value today, can I see tomorrow what it was before?
- Does the log say which field changed and who changed it, or only that "the record was modified"?
- Can an approved document be edited after approval without anyone noticing?
- When duplicate records are merged, is it recorded which ones were merged and on what criteria?
- Do imports and automated processes leave a trace, or only changes made from the screen?
If any answer is no, the institution has data it can correct but cannot defend.
For years, data quality in government was understood as cleanup: removing errors, filling gaps, standardizing formats. That work is still necessary, but clean data without history asks for faith. And public operations, especially when they touch a person's liberty or a response to a resident, cannot ask for faith. They have to show.
The question for leadership teams is not whether their data is right today. It is whether, in the face of a complaint, an audit or a change of administration, they can prove how it got that way.
If your institution wants every correction recorded with its previous value, its owner and its moment, see how Tribuna keeps the timeline of every public safety and civic justice case file, how Agora records the history of every citizen service ticket, or request a demo.