About this collection
In 1917 three shepherd children in Fátima, Portugal, said a Lady appeared to them once a month from May to October. On 13 October a crowd of tens of thousands reported that the sun spun, changed colour and seemed to fall towards the earth. Thousands of documents about these events were later gathered and published by the Sanctuary of Fátima as the Documentação Crítica de Fátima (DCF), a critical edition in 15 tomes (1992–2013), entirely in Portuguese.
This site collects every first- and second-hand witness account of the 1917 events found in the volumes we have: the passages where someone describes what they themselves saw or heard (first-hand), or relays what a specific, identifiable person told them (second-hand). That includes the three seers themselves, Lúcia, Francisco and Jacinta, whose words come to us through interrogations and later writings. Vague rumour (“they say that…”) is left out.
Sources
- DCF III. Das aparições ao processo canónico diocesano, tomes 1 (1917–18), 2 (1918–20) and 3 (1920–22). These hold almost every letter, newspaper report and testimony from the year of the apparitions.
- DCF V/4 (Jan–Apr 1930).
- DCF, Seleção de documentos (1917–1930) (2013), which reprints key documents from the other volumes, including the 1917 sworn depositions and interrogations of DCF vol. I and the canonical process of vol. II.
Not yet included: the full DCF vols. I, II, IV and V/1–3, 5–6. Accounts that appear only there will be added as those volumes are scanned.
Method
- Scanning and text recognition. The printed volumes were scanned and converted to text with OCR, keeping the editors' archaic spelling, footnotes and page breaks. The OCR turned out to “modernise” the old spelling silently (lugares for the book's logares) and to misread some words as other valid words (cavaleiro, horseman, for cavalheiro, gentleman). In a correction pass, every document with a witness passage was checked against the page scans, and the Seleção against the publisher's own digital text, restoring the edition's wording. The original writers' own spelling and grammar are kept as the editors printed them.
- Splitting into documents. The text was split into the editors' numbered documents (letters, articles, depositions…), each tied to its scanned pages.
- Finding the witnesses. Every document was read by an AI model with written criteria for what counts as first- or second-hand, which event it concerns, and where the passage starts and ends. A second, dedicated pass collected the seers' own statements (interrogations, Lúcia's writings, and words relayed by their families and visitors). Each passage is cut out of the source verbatim by a script, so the Portuguese you see is exactly the editors' transcription.
- Verification. A second model re-read every document with passages to fix attributions, events and passage boundaries and to add anything missed, and the result was checked against an earlier independent survey of the volumes. A final pass re-read every document, including those without passages, for missed testimony about the 1917 events, however late it was written down; each addition was confirmed by an independent reviewer.
- Who they are. Each witness is classified (seers, their families, local villagers, clergy, journalists, aristocracy, professionals, scientists and academics, officials, students, pilgrims, groups) with where they came from and, where the sources say so, whether they arrived as believers or sceptics. Short biographies come from the DCF editors' footnotes.
- Removing duplicates. The same letter or report often appears several times (a letter later printed in a newspaper, an article reprinted elsewhere, a document reproduced in the Seleção). Near-identical passages are merged into one, and the other printings are listed under “also printed in”. Different statements by the same person stay separate but share one witness page.
- Translation. Each passage, and each full document containing one, was translated into English paragraph by paragraph, so the two languages line up side by side. The English aims to be both faithful and readable for a modern reader: long 1917 sentences are untangled where needed, but the witness's register is kept. Where a reader needs context (an old office such as the regedor, a unit, a coin, a religious practice, the “old” and “official” clock times of 1917), a small * opens a short translator's note. Every paragraph was then re-read against the Portuguese to correct mistranslations, omissions and awkward English. The translation is machine-assisted. Where precision matters, read the Portuguese and the scan.
Classifications such as “saw it / saw nothing”, the phenomenon tags and the distances are reading aids, not verdicts. Every passage links to the full document and to the page scans, so you can check them yourself.
Images
The gallery shows the photographs, holy cards and drawings printed in the volumes, straightened and cropped. Where a cleaner copy of the same photograph exists (above all the Sanctuary of Fátima's own archive prints of the 1917 glass negatives), it is shown as the “archive print” and used as the basis for the AI versions; otherwise the book scan is used. Each image has an AI black-and-white restoration and an AI colourisation, made with OpenAI's GPT-Image 2.5 at maximum quality and checked by eye against the original. The photographs are colourised as if shot today on a modern digital camera, not in the style of an early colour process. These are interpretations, not historical records: fine detail and all colours are reconstructed and may be wrong. The unaltered originals are always one click away.
Credits
The documents were collected, transcribed and annotated by the editors of the Documentação Crítica de Fátima (Santuário de Fátima; scientific direction of the Centro de Estudos de História Religiosa, Universidade Católica Portuguesa). The map page opens on a modern map; the alternative base map is the Mappa de Portugal made for the newspaper O Século in 1909, from the Biblioteca Nacional de Portugal (public domain). Within about 50 km of Fátima it is AI-enhanced: redrawn at four times the scan's resolution by GPT-Image 2.5, given every printed name, and checked against the scan name by name. Parts of the rest are upscaled 2× and not yet checked. This project was prompted by Scott Alexander's “The Fatima Sun Miracle: Much More Than You Wanted To Know”.
Data last built 2026-10-09: 471 passages, 168 witnesses, 182 documents.