MADAGASCAR HISTORICAL RESEARCH · 14 OCTOBER 2026
The supplied catalog is now imported into an auditable SQLite database. The original workbook, catalog, CSV files, scans and transcriptions are retained unchanged.
The supplied Bulletin holdings are now integrated as a scholarly periodical: 20 logical volumes in 12 PDF bindings, nominally covering 1902–1925. The full supplied OCR index contributes 5,907 page records. All page text and original page-row identifiers are preserved, including 101 empty-text rows. The 12 PDFs match their supplied SHA-256 hashes.
| New material | Count / status |
|---|---|
| Journal | 1 publication record |
| Logical volumes | First series 1–12; nouvelle série 1–8 |
| Physical bindings | 12; kept distinct from numbered volumes |
| Provisional article records | 70; 60 body-title anchors and 10 contents entries. Partial inventory, not exhaustive. |
| Text passages | 5,806 nonempty full pages; all unchecked OCR |
| Additional checked excerpts | 2 title blocks; citations to Tantara await editorial review |
| Combined document records | 1019 (892 original + 70 Bulletin anchors + 6 dictionaries + 51 added research documents/passages) |
The Bulletin tab supports article-title/byline search, full-page text search and a holdings view. Search is accent-insensitive. Every result links to its PDF page. Contents-only records link to the contents evidence; they have no invented article start or end page. Multivolume bindings retain unresolved article-to-volume assignments.
Bulletin_collection.zip into the same parent folder. It should contain sibling folders madagascar-kg-pilot/ and bulletin_collection/. Open madagascar-kg-pilot/START_HERE.html. The new ZIP includes Bulletin metadata and text, but does not duplicate the 487 MB of supplied scans. Online PDF links are also included from the supplied catalog; their availability was not checked.The assistant checked those title blocks against the supplied scans. Both now cite the abstract Tantara work. Neither is asserted to translate the catalogued 1908 edition: that requires edition identification and passage comparison.
The 70 article records were selected from title/byline pages and main contents pages; display titles are normalized and sometimes shortened. Original OCR and bylines remain available. Serial dictionary installments are separate provisional records. Proceedings, internal chapter headings and all page rows are not automatically treated as articles. Author authority records, exact article dates and complete extents remain to be reviewed.
The supplied inventory has 58 further rows that are acquisition leads rather than present holdings, plus special items. Reprint and date distinctions remain explicit: first-series volume 11 has a 1912 concordance/1913 title discrepancy; first-series volume 12 part 2 bears a 1930 imprint; nouvelle série volume 5 is a 1934 second edition. Four bindings have a difference between source scan-leaf counts and downloaded PDF pages. Completeness is unverified. The supplied index reports 527 pages with added French OCR and 203 pages below 80 characters.
| Imported material | Count | Meaning |
|---|---|---|
| Catalog records | 892 | 421 Annual articles; 60 Tantara chapters and 411 detailed contents entries |
| Original typed relationships | 1,429 | IDs, predicates, original evidence, status and confidence preserved |
| Annual index observations | 28,501 | Unchecked OCR/transcription-derived matches; overlaps and duplicate candidate/seed matches retained |
| Tantara index observations | 496 | Contents headings only; linked to existing relationships, not added twice |
| New historical pilot | 8 assertions | 8 selected excerpts checked by the assistant against the 1884 facsimile; all assertions await human review |
| Identity proposals | 8 | Exact normalized labels across Annual and Tantara; authorities remain distinct |
| Total assertions | 30,108 | Original corpus, Bulletin bibliography, two explicit Tantara citations and dictionary bibliography |
| Approved assertions | 0 | No automatic editorial approval |
The new pilot links the article aa-1884-001 to a proposed assembly event, Rainilaiarivony’s reported participation and office, the printed royal name Ranavalomanjaka III, and source-attributed statements about hasina, territorial authority and French demands. Every new assertion leads to an exact excerpt and a PDF page.
indexedSubject assignments. A word such as “delegates” can match a diplomatic search bucket while referring to a Bible committee. It is not automatically a historical relationship.candidate entities. Year and date strings stay date_literal entities. Neither becomes a person, place or dated event by import alone.possibleParallelDocument. No translationOf assertion was created.| Priority | Finding | Next action |
|---|---|---|
| high | The page index cites Antananarivo_Annual_1884_improved_transcription.txt, which is absent. | Attach the exact indexed transcription or regenerate it from the supplied scan. Imported index excerpts remain unchecked. |
| high | The farewell-speech correspondence remains an untested hypothesis. | Compare both texts and record evidence from each before asserting translationOf. |
| high | Tantara evidence consists of contents headings, not body-text passages. | Proofread headings and locate corresponding body pages before building historical claims. |
| high | Search buckets can generate contextual false positives: delegates can refer to a Bible revision committee. | Review the matched wording in context before replacing indexedSubject with a literal semantic relation. |
| medium | Same spellings occur in candidate and seed registers, and across Annual and Tantara. | Review identities and aliases. Possible matches are separate assertions; no authorities were silently merged. |
| medium | Annual 1875–1878 records use an 1885 revised reprint transcription. | Keep issue year and physical source edition separate when citing and comparing wording. |
| high | The eight pilot assertions have assistant-checked excerpts but no human editorial approval. | Review identity, chronology and source attribution before approving individual assertions. |
| high | Complete and review the article bibliography | 70 selected provisional records are a starting index. Review all contents pages, distinguish serial installments, establish author identities, exact titles, start/end pages and logical volume boundaries. |
| medium | Review OCR and blank-page coverage | 5,907 page rows preserved, including 101 with no searchable text. Check 203 low-text pages and 527 pages with added OCR against scans. |
| medium | Resolve coverage years, reprints and scan completeness | OS11 concordance 1912 versus title 1913; OS12 part 2 imprint 1930; NS5 second edition 1934. Four bindings have source-leaf/PDF-page differences. Preserve all distinctions. |
| medium | Acquire missing holdings separately | 20 logical volumes are present. The remaining 58 inventory rows and special items are leads, not imported holdings or claims that the series is complete. |
| high | Compare the two explicit Tantara extracts | Title blocks identify Tantara, but not a validated correspondence to the catalogued 1908 edition. Establish source edition and align passages before asserting translationOf. |
Open Diktionera from either the Catalog or Knowledge graph navigation. One Malagasy query compares Richardson (1885), Abinal–Malzac (1888), Callet (1902–1904), Malzac (1893), Malzac (1926) and Freeman–Johns (1835). Callet is a ten-installment extract covering A–ARINA; it is not the complete dictionary. English and French searches are also available. Direct headwords appear before reverse matches in translations/examples. Every result links to a PDF page.
The six sources contribute 4,554 PDF page records and 113,758 provisional OCR entry candidates. Selected short excerpts were checked against the scans by the assistant; these are labeled individually. Original OCR remains separate. Entry boundaries, spelling and completeness are unreviewed. Full-page search includes grammar and front matter. A missing result does not prove that a word is absent.
abinal-malzac-dictionnaire.pdf is Malzac’s Dictionnaire français-malgache, 1893, according to its title page. The separately supplied abinal-malzac-dictionnairemalgache-1888.pdf is the Abinal–Malzac Malagasy–French dictionary (1888), now indexed as its own source. The two Malzac editions are retained separately. French source wording is not silently translated into English.Six dictionary document records and twelve bibliographic relationships were added to the KG (1,019 documents and 30,108 assertions in total; none approved). Lexical candidates stay in dictionaries/dictionaries.sqlite; no automatic multilingual equivalence or person-identity assertion is created.
Download and start: extract the updated Madagascar_Knowledge_Graph_Pilot.zip. For local dictionary page links, extract Madagascar_Dictionary_Sources.zip into the same parent folder; its madagascar-kg-pilot/dictionaries/scans/ folder merges into the project. Open madagascar-kg-pilot/Madagascar_Catalog_Multisource.html and choose Diktionera. For the new 1888 volume, also extract Madagascar_Abinal_Malzac_1888_Source.zip into the same parent folder. Callet’s extracted PDF is included in the main package. Existing dictionary and Bulletin scans can be reused. Search works without the PDF bundles.
Rebuild dictionary search with python3 build_diktionera.py, using the preserved page OCR. Reimport dictionary bibliography with python3 import_dictionaries.py, then regenerate navigation with python3 export_browser.py. The tools use Python’s standard library; fresh OCR extraction additionally requires PyMuPDF and Tesseract. Functional checks execute the shipped search code, including language direction, source page links and accented/stress-mark queries. A live-browser layout check was unavailable.
56 curated entries enrich 5 existing records and add 51 documents/passages. There are 42 inspected source locations and 14 bibliographic leads. The main catalog offers research-category, verification and royal-speaker filters. Original imported rows, Callet data, assertions and editorial review events are preserved. The update adds catalog cross-references, not approved historical assertions. See research/bam_documents.json and import_bam_research.py.
Collection dropdown: open Madagascar_Catalog_Multisource.html for the combined catalog. Its Collection selector includes Bulletin de l’Académie malgache and displays 70 provisional articles plus 12 PDF bindings. The Catalog link in START_HERE.html opens this current view. The original_catalog/ directory remains a source snapshot.
Extract the complete ZIP and open START_HERE.html. It is a read-only view generated from pilot.sqlite and runs without a server. Source links require the extracted folder structure. Original catalog pages are included under original_catalog/.
The database is the imported working snapshot. The workbook remains the unchanged supplied editorial source. After editorial changes, run the controlled import into a new database and compare it before replacing a working version. Regenerate the browser after database reviews. Do not edit the HTML as another data master.
The included Python tools use the standard library. The database integrity, import reconciliation and evidence rules are checked by the included tests. JavaScript syntax and data references are checked; a live-browser interaction check was unavailable in this environment.
The initial starter is superseded by this populated package. This release is a local research prototype, with no hosted deployment or collaborative editing service. Full Tantara body OCR, complete article cataloging, human editorial approval and validated JSON-LD/RDF exports remain future work.