A number is not an indicator: why public data needs definitions that do not change silently

It is Monday and the security cabinet is meeting. The police department reports 412 arrests for administrative offenses this month. The civic court reports 371. The mayor's office asks which number is correct.

Both are. The police count every remittance that reached the booking desk. The court counts the case files that received a ruling, and leaves out people released earlier for lack of grounds or referred to another authority. Nobody made anything up. Each department counted something different under the same name.

The meeting spends forty minutes reconciling numbers and five minutes deciding. The following month the same thing happens, this time with average resolution time: one department measures it from intake and the other from the moment the judge receives the file.

A number is not an indicator. An indicator is a number plus a written definition: what is counted, what is not, from which moment, from which source and by which calculation rule. Without that definition, even the most sophisticated dashboard only shows numbers nobody can compare.

The national case: when the way of counting changes

The problem is not limited to municipalities. In 2026 Mexico's national crime statistics are living it.

Since 2015, the 32 state prosecutors' offices have reported their crime incidence every month to the Executive Secretariat of the National Public Security System (SESNSP) using a single instrument for registering, classifying and reporting crimes and victims. In December 2024, the National Public Security Council unanimously approved, through agreement 11/L/2024, an update to that instrument. The presentation of the new methodology to the National System of Statistical and Geographic Information (SNIEG) explains why.

Among the limitations of the previous scheme, the SESNSP notes that information was collected in aggregate form, that georeference, dates and times were missing, that there was no victim data for every crime, and that there was no certainty that states applied the instrument uniformly.

The new methodology requires daily, disaggregated information for each investigation file, a direct victim for every crime with sex, age, vulnerable group and nationality, a field that distinguishes attempted crimes, and the date, time and place of the events. It also splits categories that used to be reported together. What was "extortion" is now divided into in-person extortion and extortion by other means. "Small-scale drug dealing" is separated into simple possession and possession with intent to sell. And human trafficking is broken down by type of exploitation.

These are substantive improvements. But they carry a consequence every decision-maker should keep in mind: starting in January 2026, the series is built with rules different from those of 2025. The presentation itself plans for prosecutors to deliver retroactive data back to October 2024 under the new methodology, precisely to have months measured both ways and be able to compare them.

That is the operational lesson. Changing a definition is not a mistake; it is often necessary. The mistake is changing it without writing down what changed and from when, and without a period in which both versions are measured. Without that bridge, an increase in a category may be a real increase or an effect of the new way of classifying, and nobody can tell which one they are looking at.

Why it is more fragile at the municipal level

National statistics have committees, technical standards and formal presentations. Even so, changing the instrument required planning retroactive deliveries in order to compare. In a municipality, definitions tend to be even more fragile, because they live in places nobody sees:

• In the head of whoever builds the report. The analyst who has prepared the monthly report for three years knows that same-day re-entries are not counted twice. When that analyst moves to another department, the rule leaves with them.
• In an Excel formula. Average response time is calculated in a cell that someone changed to exclude canceled cases. Nobody documented it.
• In the system's catalog. A citizen report category was split in two last year. The dashboard still adds up only the original one, and those reports appear to have dropped by half.
• In each shift's judgment. One shift records a report as "attended" when the crew leaves; another, when the crew returns with evidence.

None of these cases is a technology problem or a matter of bad faith. It is a data governance problem: the indicator exists as a number, but not as an institutional agreement.

And the consequence is concrete. An incoming administration receives a dashboard with three years of series and does not know whether the changes it sees are real. An audit asks why response time improved 30% in one quarter, and the honest answer is that the starting point of the clock was changed. An analytics or AI model learns from a series that mixes two definitions and produces forecasts that inherit that confusion.

What the standards already say

None of this is new to official statistics. The United Nations Fundamental Principles of Official Statistics, endorsed by the General Assembly in 2014, ask in principle 3 that agencies present information on sources, methods and procedures according to scientific standards, so data can be interpreted correctly. Principle 9 calls for common concepts, classifications and methods so figures are consistent across systems.

In Mexico, CONEVAL's Manual for the design and construction of indicators, published in 2013, puts this into practice. It asks that the calculation method be a clear mathematical expression in which the variables used are understood, that variables be described precisely and that the time frame of the measurement be defined. And it warns of something any municipal analyst recognizes: although a percentage is simple, its real complexity depends on how the counted variable is defined.

That manual was written for social programs, but the principle applies just as well to a civic court or a public services department. An indicator without a written definition cannot be verified, and what cannot be verified is useless for accountability.

The sheet every indicator needs

In practice, every indicator that reaches an executive dashboard or a public report should have a short, living sheet. It takes six answers.

1. What it counts and what it does not

The exact unit: remittances, people, case files or rulings. And the explicit exclusions: same-day re-entries, canceled cases, duplicate reports that were merged.

2. From where to where

For any time indicator, which event starts the clock and which one stops it. Intake at the booking desk or assignment to the judge. Receipt of the report or assignment to the crew. Crew departure or closure with evidence.

3. Which source it comes from

Which system, which record status and which field. If the data comes from a system where every file has clear statuses, the rule can be written on top of those statuses. If it comes from a spreadsheet someone fills out at the end of the shift, the sheet should say so.

4. Which catalog

Which version of the offense catalog, the report type catalog or the neighborhood catalog is used. When a catalog changes, the indicator changes even if nobody touches the formula.

5. Since when this definition applies

The effective date and, if there was a previous version, what it said. This is what lets you read a three-year series without confusing a change in the rule with a change in reality.

6. Who approved it

The department responsible for the definition and the person who authorized the latest change. Without an owner, the definition changes every time someone needs the number to look different.

And one additional rule for changes: when a definition is modified, both versions are measured in parallel for at least one period, just as the SESNSP did with its retroactive delivery. That overlap is the bridge that makes the series explainable.

It is the natural continuation of what we argued in Structured intake is an operational control, not a form-design detail: if data is captured with structure, the indicator can be defined on that structure. If it is captured as free text, the definition ends up in the hands of whoever interprets the text.

How this connects to Tribuna and Agora

An indicator sheet only works if the data underneath is consistent. That is why the definition starts in operations, not on the dashboard.

In Tribuna, Intello's platform for public safety and civic justice, police, the medical examiner and the judge work on the same case file, with clear statuses: pending, assigned, sentenced and released. That makes it possible to write an indicator on events recorded in the system rather than on each department's interpretation. The intelligence map shows top offenses, repeat offending and average resolution time on that same base, and the people network detects duplicates by CURP, date of birth, address and phone to keep a single identity per person, which avoids counting twice someone who gave a different name. Implementation starts with a diagnostic that documents flows, catalogs and rules, followed by an institutional configuration where forms and naming conventions are defined: the natural moment to agree on definitions.

In Agora, the citizen services platform, reports from every channel arrive in a single system with deduplication and automatic classification, and traceability keeps the full history of each case. The public dashboard shows active cases, resolved cases and performance metrics. Precisely because those metrics are published, each one should have its sheet: what counts as resolved and from which moment the clock runs.

On both platforms, the value is not in calculating an average, which any spreadsheet can do. It is in calculating that average on records with structure, statuses and traceability, so the definition can be written, reviewed and sustained across administrations.

Questions to review your institution's indicators

Before the next report or the next dashboard purchase, it is worth asking:

• Does every indicator presented to leadership have a written definition stating what it counts and what it excludes?
• Do time indicators state which event starts and which event stops the clock?
• Is it known which system, status and field each figure comes from?
• When a catalog changed, did anyone review which indicators were affected?
• Do historical series state since when each definition applies?
• Is there a department responsible for approving definition changes?
• Do two departments reporting the same indicator get the same number?

If the answer to the last one is no, the problem is not data capture or visualization software. It is that the institution has not yet agreed on what it is measuring.

The institutional lesson

Governments invest in dashboards, observatories and, increasingly, AI-driven analytics. All of it rests on a precondition that cannot be bought: that every number has a written definition, with an owner and a date.

Mexico's national crime statistics are doing that work at country scale. Municipalities can do it at the scale of their own operations, one indicator at a time, starting with the five or six that reach the decision table every week.

The question for leadership teams is not how many indicators their dashboard has. It is how many of them they could explain, in writing, to an administration that arrives tomorrow.

If your institution wants indicators that hold up on consistent records, see how Tribuna organizes public safety and civic justice operations in a single case file, how Agora measures citizen services with full traceability, or request a demo.

Sources

• SESNSP, Nueva metodología para la recopilación de incidencia delictiva, presentation to the National System of Statistical and Geographic Information (2025)
• United Nations, Fundamental Principles of Official Statistics (A/RES/68/261)
• CONEVAL, Manual para el diseño y la construcción de indicadores. Instrumentos principales para el monitoreo de programas sociales de México (2013)