Methodology
Transparency note. The information on this page was processed with artificial intelligence. An AI model (Claude, by Anthropic) read the documents, extracted and classified their content, translated the summaries into English, drafted the text and built the page, working from PDF files supplied by the project owner, who set the scope, the questions and the formats. Prepared on 20 October 2026. The output can contain errors of reading, classification or summary. Check it against the linked PDFs before quoting it.
Steps
- Scope. The corpus is the joint declarations of the Proceso de Quito (I to XI), the Plan de Acción of November 2018, the Hoja de Ruta del Capítulo de Buenos Aires and the Joint Communiqué of 30 January 2025, as supplied. Other files in the project were not analysed: the Regional Contingency Protocol, the identification guides on trafficking and smuggling, the regional reports, the concept notes, the RAP booklets, the voluntary review of the Global Compact for Migration and the socioeconomic integration strategy. One exception: the XII Ronda concept note was consulted only to state that no XII declaration exists yet. No outside source was used for the content of the declarations.
- Reading. Files with a text layer were extracted as text. Scanned files (eight, listed above) were converted to images and read page by page by the AI model. No automatic text recognition was used, because the available tool had no Spanish model.
- Identification. For each file the title, date, place, participating governments and signatures were taken from the preamble and signature pages, not from the file name. This showed that two file names were wrong (see the gaps list), and the III Reunión was added when supplied.
- Extraction. Each document was read in full and broken down into items: a statement, appeal, recommendation, instrument, body or funding request that the text itself makes, with its paragraph reference. The rule was to record only what is explicitly written, with no inference and no outside knowledge. The result is 166 items across 13 documents.
- Classification. Each item was given one theme (13) and one type of intervention (6). Both lists were created for this page and do not come from the texts. Types, as used here: Stance (a position, recognition or reaffirmation); Appeal (a call or exhortation to Venezuela, the international community or other bodies); Recommendation to States (a measure States are invited to take); Instrument or product (a platform, guide, study, card or mapping); Institution or mechanism (a body, group, secretariat or follow-up arrangement); Funding call (a request for financial or technical cooperation). One sentence can produce more than one item.
- Quotations. The 57 quotations are in the original Spanish, with the paragraph or annex reference. Where the file has a text layer, each quotation was checked against the extracted text, automatically (three quotations from the table layout of the Plan de Acción were confirmed by hand). For the scanned files they were transcribed from the images and not machine-checked. Items in English are AI summaries of the text, not official translations.
- Matrix. A number is the count of items coded to that theme in that document. A dot means no item was coded, not that the topic is absent from other parts of the process. Counts depend on how sentences were split and classified, so they show where a document puts its attention, not how important a theme is.
- Periods and shifts. The three periods and the one-line “shift” of each card are an analytical reading made for this page. The texts do not use these labels.
- Seminar lens. The optional notes (off by default) relate the texts to the seminar on free movement and mobility regimes. They are interpretation and are kept apart from what the documents say.
- Outflow figures. The strip reproduces the totals as each declaration cites them. Bar length is proportional to the figure on a scale from 0 to 6.5 million. The sources and what they count differ (departures, people in the region), so the bars are not a single comparable series.
- Page checks. The page was opened in a browser at desktop and phone width to check that it renders, that the filters, tabs and hover texts work and that every PDF link points to an existing file.
Limits
- The classification is a judgement. A different reader could split or label some items differently.
- Reading scanned pages can produce slips in names, numbers or paragraph references. Numbers in the outflow strip were re-read against the pages or the extracted text.
- Annexes that are not in the supplied files (for example the project profiles of the Hoja de Ruta) are not covered.
- Only the supplied texts are covered. Whether a measure was applied in practice is outside this page.