On a Friday at 11:40 p.m., a patrol unit brings a man to the booking desk for taking part in a fight. He has no ID. He says his name is Luis Hernández and the clerk types Luis Hernández. He pays his fine and leaves.
In March, the same man had come through that desk as Luis Hernandes. In June, as José Luis Mora. In the system they are three people with no prior record, and the civic judge who sees him for the third time rules as if it were the first.
Now the reverse case. A different Luis Hernández, who has never set foot in a holding cell, is stopped at a checkpoint and the system shows two case records that are not his. He shares a name with someone else and nobody has a way to tell them apart.
These are two failures with one cause: the municipality stored a name and treated it as a person.
Mexico's National Detention Registry Law also applies to civic justice. Article 2 includes administrative arrest among the grounds for being a detained person, and Article 3 covers the administrative sanctioning procedure before a municipal or civic judge.
Article 17 requires the detention to be registered immediately. Article 18 sets out what that first record must contain, and the first item is the name. At that moment, in the field or at the door, the name is usually whatever the person declares.
Confirmation comes later. Article 23 orders administrative institutions to update the record with place and date of birth, address, CURP (Mexico's national population ID code), fingerprints, a photograph and other means that allow the person to be fully identified.
So the law provides two moments: an immediate data point and a later identification. The problem appears when operations only fulfill the first, or fill in the second without anyone comparing anything. Article 29 adds a point worth reading closely: the truthfulness of the information is the responsibility of the authority that generates it.
The 2025 National Census of Municipal Governments by INEGI, Mexico's statistics agency, reports 1,256,198 people handed over to an authority by municipal public safety institutions in 2024. Of those, 1,079,982 (86.0%) went before the civic justice authority.
In other words, most of the people municipal police hand over enter through civic justice, a fast, high-volume circuit. A badly captured data point can travel through the entire procedure there without anyone questioning it.
The same census shows that only 508 municipalities (20.6%) reported a formally constituted civic court. In 1,747 (71.0%) another authority performed the function, most often the municipal trustee (822 municipalities) or the public safety institution itself (498). In many places, then, the person deciding has no specialized area behind them to verify who is standing in front of them.
When the name fails at that scale, the loss does not stay in a single file.
A badly resolved identity can fail in two directions, and they deserve separate treatment because they are corrected differently.
One person with several names. This is the split record. Mexico's National Model of Police and Civic Justice calls for monitoring people who have been in conflict with the law to prevent reoffending or escalation into criminal activity. That monitoring cannot work if every visit creates a new person. A community service program gets assigned without anyone knowing a previous one was abandoned, and the municipality's repeat offense figures fall below reality without anyone noticing.
Several people with one name. This is the merged record, and it is the more serious error. If two histories are joined because they look alike, a person carries someone else's record before a judge, a commander or an officer in the field. In a country where many surnames repeat, searching by name alone makes this confusion likely.
That is why the answer cannot be to merge everything that looks similar. An aggressive rule fixes the first error and manufactures the second. We explained this in A search hit is not enough: linking records adds context only if the link can be defended.
Fingerprints and faces solve a good part of the problem because they do not depend on what the person chooses to say. They also have limits an institution should understand before buying.
The evaluation by the U.S. National Institute of Standards and Technology (NIST), published in 2019, analyzed 189 facial recognition algorithms from 99 developers using more than 18 million images. It separated two tasks that are often confused. One-to-one verification asks whether two images show the same person. One-to-many identification looks for someone across an entire database.
It found demographic differences in most of the algorithms, and also that those differences vary widely between products. The report's primary author summed it up in one line: "Different algorithms perform differently." The report warns about the cost of each error: in a one-to-many search, a false positive places an innocent person on a candidate list and can lead to a false accusation.
Fingerprints are not exempt either. In 2004, the FBI wrongly identified Brandon Mayfield as the source of a print found on a bag connected to the Madrid bombings. The review by the U.S. Justice Department's Office of the Inspector General concluded that three FBI examiners and a fourth, outside court-appointed examiner were confused by an extraordinarily rare degree of similarity. It also found circular reasoning: details visible in the known print suggested features that were not in the questioned print. The reforms and recommendations include blind verification and documenting what was observed before comparing.
A fingerprint taken at the booking desk, with the person present, is far cleaner than a latent print on a bag. The operational lesson still carries over: a comparison produces a result that someone must be able to review, and the reviewer should not begin by believing the machine.
A municipality does not need to resolve identity in the first minute. It needs to stop pretending it already did. That translates into concrete capture rules:
- Declared name and confirmed identity are separate fields. The first is recorded immediately, exactly as the person said it. The second is filled in only when there is support for it.
- Visible status. Anyone opening the file should be able to tell "declared" from "under review" and "confirmed."
- Confirmation method. Document, CURP, fingerprint, photograph or another means, because they do not all carry the same weight.
- Who and when. Who confirmed it, in what role and at what time.
- Candidates instead of doubtful automatic merges. If the resemblance falls short, the system proposes and a person decides.
- Reversible merges. Every record merge should keep what was joined, who approved it and how to undo it.
- A correction path. Article 24 of the law requires a record of each update, and Article 30 gives access to the person deprived of liberty and their legal representative. A namesake clarification should leave the history corrected and explained.
The person without documents deserves a separate note. Not carrying ID is common and proves nothing. The file must be able to move forward with an unconfirmed identity, without blocking the person's medical certificate, hearing or release, and without that condition being read as suspicion.
The reform to Mexico's General Population Law published on July 16, 2025 established that the CURP with fingerprints and a photograph will be the mandatory national identification document (Article 91 Bis) and created a Single Identity Platform to look up and validate codes (Article 91 Quater).
It is tempting to think this makes the municipal problem disappear. It is better not to plan that way. Article 91 Bis itself states that biometric data will be integrated with the prior consent of the holders. Article 91 Ter says the biometric CURP may be integrated into other national registries and systems under applicable law. Those articles do not by themselves settle how a municipal booking desk will run a lookup on a Friday night, or what happens with someone who does not have that data on file.
In the meantime, the municipality is still required to record fingerprints and photographs under Article 23 and to handle them with care. Mexico's General Law on Personal Data Protection Held by Obligated Parties defines as sensitive any data whose misuse may give rise to discrimination or entail a serious risk. Its Article 17 requires data to be kept accurate, complete, correct and up to date. A record merged because of a shared name is, on top of an operational problem, a data quality problem with consequences for a specific person.
Tribuna treats identity as part of the operational flow rather than a separate procedure.
Fingerprint capture is built into the case record, by hand and finger, from the web or with a reader on a tablet. Prints are compared against existing records with a dedicated biometric engine to confirm identity, even when the person gives a different name. There is also face search using the intake photograph, and every comparison is logged.
In the field, officers can search for people by text or fingerprint from Android tablets, even without a signal, and the record syncs when the connection returns.
For split records, Tribuna identifies duplicates by CURP, date of birth, address and phone number, and allows high-confidence record merges to keep one identity per person. That identity supports the file with repeat offenses, frequent neighborhoods and hours that opens from 360° Search.
Controls run alongside all of this. Results are filtered by the role of the person searching, every read and every change is recorded, and opening a sensitive file requires a stated reason. Technology compares and leaves a trail. The institution defines in its role and permission configuration who confirms, and the judge still rules with all the information at hand.
Before tendering for a fingerprint reader, the civic justice office should answer with data:
- How many case records from the past year have a confirmed identity, and by what method?
- Does your system distinguish the declared name from the verified identity?
- How many people appear duplicated in your database, and how do you know?
- Who reviews a biometric match before it changes a case file?
- Can a record merge be undone, and can you tell who approved it?
- What procedure does a namesake follow to correct a record that is not theirs?
- Do you know your vendor's error rates and the population they were tested on?
If several answers depend on the clerk's memory or on the judge recognizing a face, your repeat offense record rests on data nobody checked.
A name is not an identity. It is the first thing a person says to the authority, and it deserves to be recorded exactly as said. Turning it into an identity takes a method, a responsible person and a way to correct it.
That discipline protects in two directions. It protects the institution, which stops treating as a first time what has already happened three times. And it protects the person, who does not pay for the history of someone they only share a surname with.
If your institution is reviewing how it identifies and registers people at the booking desk, explore Tribuna, see how Agora organizes municipal citizen services, or request a demo.
- Chamber of Deputies of Mexico, Ley Nacional del Registro de Detenciones, DOF May 27, 2019, Articles 2, 3, 17, 18, 23, 24, 29 and 30
- INEGI, Censo Nacional de Gobiernos Municipales y Demarcaciones Territoriales de la Ciudad de México 2025, results report, March 19, 2026
- National Public Security Council of Mexico, Modelo Nacional de Policía y Justicia Cívica, Annex 1 of Agreement 05/XLVI/20, DOF January 26, 2021
- National Institute of Standards and Technology, NIST Study Evaluates Effects of Race, Age, Sex on Face Recognition Software, December 2019
- Office of the Inspector General, U.S. Department of Justice, A Review of the FBI's Handling of the Brandon Mayfield Case, conclusions
- Chamber of Deputies of Mexico, Ley General de Población, Articles 91 Bis, 91 Ter and 91 Quater
- Chamber of Deputies of Mexico, Ley General de Protección de Datos Personales en Posesión de Sujetos Obligados, Articles 3 and 17