Methodology · version 2026.08
How we build the Observatory.
We monitor AI announcements, legislation and judicial cases across the Mexican state. Agents automate detection and the initial cross-check; every record declares its sources, its review level and its history. The method must be possible to read, question and reproduce.
Editorial commitment
- —Automation is never presented as human audit.
- —We publish the origin of every claim and distinguish between source, review and status.
- —No public record without an accessible source.
- —Trust labels are derived from the source; they are never hand-assigned by whoever captures the record.
- —Every editorial correction keeps its history and generates a public event.
- —When the method changes, it is versioned here — never silently patched.
01 · Process
From finding to public record.
Detect
Continuous monitoring of government, legislative and judicial sources (gob.mx, DOF — the Official Gazette of the Federation, parliamentary gazettes, SJF — the Federal Judicial Weekly).
Output: A candidate finding with source, date, actor and original link.
Normalize
Standardization of names, dates and agencies, plus duplicate detection against what is already published.
Output: A single record with structured fields.
Cross-check
Comparison against the primary document or verifiable public sources. Without sufficient evidence, the finding is not published.
Output: Linked evidence with verbatim quotes.
Classify
Application of the public status, topic and scope criteria (see section 04). Trust labels are never assigned by hand: they are derived from the source.
Output: Reproducible status and classification.
Publish
The record goes out with its sources, its review level and its public reference number. No public record without an accessible source.
Output: A public record with full traceability.
Monitor
Review of status changes, deadlines and corrections. Every material update generates a public event in the activity log.
Output: A visible history per record.
02 · Traceability
Two axes, so evidence and review never get conflated.
"Official source" describes where the evidence comes from — not who verified it. They are independent axes, and each record shows them separately. Today Axis A is derived automatically from each source’s domain; Axis B fields are being rolled out record by record, and in the meantime every record states honestly whether its human audit is still pending.
Axis A · Source origin
The evidence comes from gob.mx, the DOF, parliamentary gazettes, the SCJN/SJF (Supreme Court / Federal Judicial Weekly), state congresses or another public body. Derived automatically from the source domain — never assigned by hand.
A verifiable public outlet or document (e.g. a national newspaper), pending corroboration against an official primary source.
Not published. The Observatory keeps no public records without a source anyone can consult.
Axis B · Review level
An agent compared the record’s fields against its sources. This does not imply human audit, and the record never presents it as such.
A person reviewed the record against its evidence. To count as audited, the reviewer and date must be on file.
The record is published as provisional with its source in plain sight, labeled "pending human audit", and prioritized by sensitivity.
03 · Worked example
The methodology is best understood on a real record.
Anyone should be able to reconstruct what happened from the first finding to the current state. This is a real public record of the Observatory — including its pending human audit, because transparency also covers what is still missing.
AI regulation against digital violence and deepfakes — reform to the Código Penal Federal (Federal Criminal Code) and the Ley de Acceso de las Mujeres a una Vida Libre de Violencia (Law on Women’s Access to a Life Free of Violence)
- Primary source
- Gaceta del Senado (Senate Gazette) · LXVI Legislature
- Sponsor
- Sen. Karen Castrejón Trujillo
- Legislative status
- In committee
- Next action
- Confirm committee ruling and preserve the primary document
- Feb 4, 2026Detected in the Gaceta del Senado; the record is created with a public reference number.
- Feb 4, 2026Initial cross-check: sponsor, date and subject matter against the parliamentary source.
- 2026Published as provisional, with its source in plain sight and flagged for human audit.
04 · Criteria
What each status means.
These are the six statuses of the announcements catalog — the same rule regardless of topic or institution. Every total shown on the site is the sum of its visible parts; a status outside the catalog is shown as "unclassified", never hidden. For legislation and judicial cases, the status mirrors the parliamentary or judicial source (in committee, referred, resolved…), not our own interpretation.
Promised
Publicly announced by an authority, with no verifiable evidence of execution yet.
In development
There is concrete, verifiable progress (a contract, works, a pilot, an open call), but it is not yet available to its target population.
Operating
There is public evidence of real use or availability for the target population. Operating does not equal fulfilling everything promised: it qualifies what already works.
Concluded
The commitment had a bounded scope (an event, a delivery, a publication) and it happened. Closed with its evidence.
Broken
An explicit, verifiable deadline passed with no evidence of fulfillment. Without a verifiable deadline no breach is declared: the record keeps its last provable status.
Abandoned
The authority cancelled the commitment, or the program sustaining it ceased to exist, with evidence of the abandonment.
"Operating" is not "compliance" either: recap percentages are labeled "in operation", and "compliance" is reserved for commitments with a verifiable scope and deadline.
The methodology has a history too.
This page is versioned (v2026.08). Status changes and corrections live in the public activity log; the data can be browsed as an accessible table; and any error can be reported in the project’s public repository.