Methodology and preservation
2192 HTML/PDF source records → 2090 unique documents. Every source is accounted for in the ledger.
HTML identity uses conservative normalized full content and meaningful media, lists, quotes and table relationships, independently of filenames and titles. Only byte-identical PDFs merge. Translations, revisions, excerpts and uncertain matches remain separate.
- identity
- HTML: NFC/whitespace normalized complete ordered content, independent of title and formatting, with media/quotes/list/table relationships; ambiguous extraction retained. PDF: original bytes SHA256 only.
- citations
- Document-level source citations. Source-node locations are retained in structured blocks; no invented paragraph attribution.
- versions
- Substantive versions, translations, excerpts and shared passages remain distinct. Candidate similarity never merges.
- normalization
- NFC and whitespace only; case, punctuation and Ethiopic text retained.
- categories
- Representative editorial groups use recorded source-body evidence. All legacy source-category memberships and classification review flags are preserved in taxonomy.json. Non-English sources retain original text without invented translation; unresolved classification stays discoverable.
- presentation
- Pure empty unlabelled spacer fields are omitted visually; source-labelled missing values and all cell coordinates remain in the structured models and per-source projection evidence. Source-proven page-layout groups flow in source order.
- metadata
- Visible bylines, credited roles and dates use exact source-body excerpts pinned to original SHA256; catalog CMS creator/date declarations remain separate and do not imply authorship. Dates keep original spelling and precision.
- pdf text
- Preserved catalog text extracted with pypdf supports local search and advisory comparison; PDF identity and rendering use only unchanged original bytes.
Subject groups and source collections
Editorial subject groups organize actual content; original archive categories remain evidence in the complete membership mapping. Classification review flags remain visible and unclassified material stays discoverable. Original languages are not translated, and native-language approval is not claimed.
Subject mapping, rationale and review flags (JSON) · Every supplied record and its role (JSON) · Structured documents, citations and complete decisions (JSON)
Exact duplicate groups
- ELF-RC: Poised for Another Round of Persecution? · 2 source records, each retained in Sources.
- Human Rights Violations committed in 2006 · 2 source records, each retained in Sources.
- Remembering the forgotten Eritrean Revolutionary Hero · 2 source records, each retained in Sources.
- Over 50 Parents Jailed in Tsaeda Christian · 2 source records, each retained in Sources.
- Ona and Besikdira massacres · 2 source records, each retained in Sources.
- Open Doors (Deutschland) : Eritrea - Hadas und ihre Töchter · 2 source records, each retained in Sources.
- A New Opposition Needed Not A New Type Of Politics · 2 source records, each retained in Sources.
- Awate - Update on the “G-15”, the “G-1” & The People · 2 source records, each retained in Sources.
Source decisions
Inspect the complete source-to-document ledger. Meaningful attachments and unavailable references remain linked from archival source records; CSS, scripts and images are accounted as assets.
Comparison candidates
1310 advisory pairs remain separate. Their relationships and evidence are listed with the structured records; similarity never removes a document.