Provenance & methodology
This is a research catalogue of publicly archived material. It is built to be honest about where each fact comes from and where the record is silent.
The source
Every record is drawn from the SABC / South African History Archive (SAHA) — Truth Commission Special Report. The collection was crawled in full: 6,869 documents, each preserved as Markdown with its original URL, publisher, category, download date and a SHA-256 content hash. Original provenance is never overwritten with inferred information, and every record on this site links back to its source document.
How records are built
A deterministic parser reads only the downloaded source files and extracts a standardised catalogue entry: title, document family and type, event dates, the people named and their roles, four separate geographic fields (province, municipality, area and venue), amnesty case identifiers and — for amnesty decisions — the outcome. Values are kept close to the source wording; historic place names are preserved rather than modernised.
A short plain-English summary is written for each record by an on-device language model working strictly from that record’s own text. Summaries are a reading aid, not official Commission text. 6,783 of 6,869 records carry one.
What is published here
This site publishes catalogue metadata and the short summaries only. It does not re-host the full transcripts; to read a document in full, follow its link to the original archive. Of 6,869 records, 4,137 name at least one person, 2,097 carry a case identifier, and 364 are archive index / landing pages rather than individual hearings.
Sensitivity
This archive concerns survivor, perpetrator and family testimony given to the Truth and Reconciliation Commission. The material is treated as sensitive. This site publishes catalogue metadata and short summaries only, and links back to the original documents on the public SABC / SAHA archive; it does not re-host the full transcripts.
How to read this data
- Automated standardisation is complete, but full manual QA is not claimed: every record's review status is “Not reviewed”.
- Fields are left blank rather than inferred. A missing person, date, place or outcome means the source did not explicitly state it.
- Summaries are evidence-bounded and were generated on-device from each source document. They are a reading aid, not official TRC text.
- Case IDs and matter descriptions are kept separate. A description is shown only when supported by an exact quotation from the source.
- Historic, source-era place names are preserved (for example Port Elizabeth, Pietersburg and Umtata) and are not silently modernised.
- A small number of records carry event dates outside the Commission's active period; these are retained as-registered and flagged for manual check.
Data version 1.0.0, built 2026-07-28. Generated by trc-publish-site-data.