How ThunderBay.Wiki Works
How It Works
ThunderBay.wiki combines automated research with human verification:
- Primary sources as anchors. Census data, birth certificates, marriage certificates, death certificates, military service records, and other historical primary documents, as held at their source (often FamilySearch).
- Obituaries for linking hints. Obituaries explicitly state life dates, family relationships, and migrations, although they cannot always be trusted, as the information is not always literally true. However, often, parents, siblings, spouses, children - are identified and their records can then be located and connected.
- AI-assisted record linkage. Automated research agents cross-reference newspaper indexes, census returns, probate records, and published local history to draft each biography, with every fact carrying a citation to its source. A fact without a source is left blank rather than guessed.
- Automatic entity resolution. Identity resolution — confirming that the "James Murphy" in an 1891 census return is the same James Murphy in a 1928 obituary — is solved probabilistically from all the sources in a CIDOC CRM knowledge graph. Doubtful or ambiguous matches are to be listed at [INSERT LINK] so interested humans can add additional evidence or attestations to move the entity resolver in the correct direction. However, all entity resolution is performed automatically - humans only intervene by adding more evidence.
- Privacy by design. Automated research is limited to individuals deceased 20 years or more, plus obituary-sourced memorial pages for recent passings. Living residents appear only by explicit opt-in. See our privacy policy and corrections process.
Other examples:
- Vienna History Wiki - A Semantic MediaWiki (SMW) website with 54,000+ articles about the City of Vienna, Austria.
- Liquipedia - A SMW website for esports with leaderboards and complex cross-referencing.
- WarSampo, 100,000 records about Finland in WWII, part of a larger "SampoSampo" program that uses CIDOC CRM built by Esko Ikkala [1] [2] [3]
- https://researchspace.org/
Sources
| Source | Type | Coverage | Notes |
|---|---|---|---|
| Public obituaries | Secondary | Late 19th century – present | Sourced from local newspapers and public memorial sites See [4] for prior research on this topic |
| Local newspaper indexes | Index | Late 19th century – present | Social, vital-statistics, and subject indexes to the Daily Times-Journal, News-Chronicle, Chronicle-Journal and predecessors |
| Papers and Records | Secondary | 1973 – present | Annual journal of the Thunder Bay Historical Museum Society; selected quotations cited with links back to the original documents available for purchase at TBHMS. |
| Works of F. Brent Scollie | Secondary | 19th – 20th century | Biographical dictionaries of early Fort William and Port Arthur residents; sources used but the actual text is not used as it is under the author's copyright |
| EliteProspects | Secondary | 19th – 21st century | Every ice hockey player from Thunder Bay tracked by a highly comprehensive player database of professional and junior players [5] |
| Probate records (Livio Di Matteo) | Primary / dataset | Late 19th – early 20th century | Thunder Bay District probate and wealth-holding data compiled by economic historian Livio Di Matteo |
| Census of Canada (1881) | Primary | 1881 | Full-count nominal data via IPUMS International |
| Census of Canada (1901, 1911) | Primary | 1901, 1911 | Volunteer-transcribed nominal indexes for 1901 Census and 1911 Census; transcribed by human volunteers circa 2018 by a project organized by New Brunswick programmer Lindsay Patten as AutomatedGenealogy.com |
| Thunder Bay BandWiki | Primary | 2008 - 2014 | Volunteeer-curated information about local bands |
| Gateway to Northwestern Ontario History | Primary / images | 19th – 20th century | Digitized photographs, maps, and documents managed by the Thunder Bay Public Library |
| Canadian Who's Who and Who's Who in Canada, Ontario Who's Who, etc. | Primary | 19th – 20th century | Biographies [6] On Internet Archive: [7] |
| FamilySearch | Linkage | Global | Each biography is linked to its FamilySearch Person ID, connecting local records to the global family tree (link coverage pending as of June 2026) |
| Henderson's Guides | Primary | 1907-2004 | Available at Brodie Library on Paper and Microfilm, also scanned by TBHMS |
| Vital statistics, land, and military records | Primary | Various | Ontario civil registration, land records, and service files as they become available |
Get Involved
- Search for your family. Use the search box to look for ancestors by name. Red links indicate people identified in the records who don't yet have a biography — they're our work queue.
- Submit a correction. If a source was misread or an identity wrongly merged, let us know on the article's talk page or via the contact address.
- Contribute records. Family bibles, letters, photographs, and privately held obituary clippings help anchor biographies that public records can't. Please email Michael B. Currie using the first initial of his first name, followed by his last name, followed by the at sign, followed by "gmail.com". Consult the User's Guide for information on using the wiki software.
- Living residents: you will not appear here by default. To document your own connection to the tree, see the opt-in process in our privacy policy.
Purpose
This website is another vain attempt to preserve information longitudinally, a task which has always been a low priority for any civilization:
What makes Encyclopedia-style articles so fascinating is how they look at subjects normally seen only in a snapshot at their present moment, and instead provide perspective by showing a little bit of information, the most salient information, spread over their whole lives. So instead of a LOT of information about the present moment, instead you get a nice even amount of information across their whole lives. Which creates a perspective that cannot easily be seen otherwise. It shows a more whole perspective, by showing the dimension of time. Another reason to do genealogy is to become habituated to the broad sweep of a human life and therefore become better at planning and thinking in one's own life. By studying one's own ancestors and seeing the mistakes they made, it's possible to contextualize events in your own life not as singular, special things but as variations on a theme established by dozens of your forebears.
— Michael B. Currie, October 2019
Newest Articles
Read, from ThunderBay.wiki's newest articles newest articles; or its longest ones. Some highlights:
- James Murphy, a Fort William coal merchant, left the largest estate of any Thunder Bay District resident probated before 1930 — inventoried at $750,851 in 1928 — one of more than 2,300 local decedents newly catalogued from the Canadian Regional Historical Wealth Micro-Data Collection.
- Three of Thunder Bay's seven "elite" Dance Studios were founded by students of the second matriarch of Thunder Bay elite dance training, Sylvia Horn. Who was the first?
- J.J. Carrick introduced Daylight Savings Time while he serving as Mayor of Port Arthur in 1908, making it the first jurisdiction anywhere in the world to use it.
- South Hill Street, Thunder Bay is 1.6 kilometres long and is the address for about 152 detached homes.
- Clyde Currie has at least 25 step-descendants through his children Marilyn and Allan.
- Ruth Newman won the top prize for her Royal Conservatory piano examination in 1941.
- Neill Currie's bomber's nose was shattered on 27 September 1944, over Germany.
- Sir Arthur Currie is not closely related to Michael B. Currie.
- The First Millennial Foundation, a non-profit utopian movement to create space colonies, was one of the first organizations to coordinate its activites over the World Wide Web?
Data Aggregation and Deduplication
The foundational dataset was built upon massive archives of primary local sources. This initial ingestion phase included:
- Over 500,000 local newspaper entries (digitized from PDFs), originally indexed by community volunteers.
- Thousands of regional estate and probate records.
An AI-driven pipeline was utilized to parse, clean, and deduplicate these primary sources, ensuring that varying name spellings, overlapping dates, and fragmented records were correctly attributed to single historical identities.
Data is saved to a PostGIS database that mimics a proper graph database by storing data using JSONB. The data schema follows the CIDOC CRM ontology. CIDOC CRM is strictly "event-centric", which lets all data represent claims which can be accepted or rejected at the conclusions layer presented to the user via the wiki.
The key benefit of a CIDOC CRM ontology is that it is flexible enough to allow sources to be ingested as an entirely separate process from drawing conclusions. In a non-CIDOC-CRM system, each additional source adds a nonlinear amount of new work, and the work must be managed by humans. By contrast CIDOC CRM allows both processes to be handled by increasingly automated means, and the conclusions improve in a monotonic non-decreasing fashion because no information is ever thrown away.
In the future we may implement a proper graph database the syncs with the PostGIS database, using Polyglot Persistence, or something simpler like Apache AGE or OnTop.
ThunderBay.Wiki is a reference implementation of a site that is well-behaved with respect to the Semantic Web. Its sources layer is mastered by datum.locker, which accepts new source submissions from casual users in an accessible Web UI, and which implements a PostGIS JSONB implementation of CIDOC CRM's knowledge graph ontology. Open Standards used:
- CIDOC Conceptual Reference Model (CRM), an ISO standard (ISO 21127)
- W3C's Resource Description Framework (RDF)
- Simple Knowledge Organization System (SKOS) (a W3C "recommendation")
- PostgreSQL (+PostGIS)
- Future: APACHE AGE (on top of PostGIS)
- Future: Formalize our CIDOC CRM implementation using W3C's Web Ontology Language (OWL), i.e. the Erlangen implementation of CIDOC CRM into OWL.
- Future: host SPARQL endpoints
It seems like the fully federated semantic web is infeasible because of the high fixed cost with developing a node and understanding how to use it. So a smaller number of highly user-friendly well-behaved nodes like ThunderBay.wiki and datum.locker could be adapted by many humanities projects to avoid heavy work on their part.
Entity Generation and Interoperability
Following the data cleaning phase, the system generated thousands of foundational "stub" articles, creating a unique MediaWiki entry for each identified individual.
To ensure the repository is interoperable with broader genealogical and academic databases, the data architecture prioritizes strict linking and provenance:
- Global Identifiers: Where possible, entities are mapped to the global FamilySearch PID system and Wikidata Q identifiers, embedding the local database within the global semantic web.
- Source Provenance: Every claim and entity maintains direct reference links back to the original historical newspaper entries and source documents.
Progressive AI Expansion via RAG
To elevate the stubs into comprehensive, narrative-style biographies, the project employs a progressive scanning AI loop.
This automated system continuously executes Retrieval-Augmented Generation (RAG) searches on existing entities. It crawls and synthesizes supplementary information from a vast array of secondary sources, including:
- Academic research and historical journals
- Online articles and digital archives
- Full-text digital obituaries
This continuous loop allows the database to autonomously expand its coverage and depth, generating rich historical narratives for a large fraction of the region's historical population.
Privacy and PIPEDA Compliance
The platform adheres strictly to the Personal Information Protection and Electronic Documents Act (PIPEDA). While the database maps familial linkages that connect historical figures to living descendants, all information regarding living individuals is strictly obfuscated unless the explicitly opt-in to having their information displayed publicly. Any information incidentally collected about living persons who have not opted in are shielded behind a secure gate, accessible only to explicitly approved researchers, until such time as a death record transitions them to the public historical domain.
Future Roadmap
The ongoing development of the repository will focus on deeper data integration and structural refinement:
- Semantic Indexing: Transitioning the MediaWiki structure to fully semantic data models (e.g., Semantic MediaWiki) to allow for complex, database-style querying by researchers.
- Primary Source Expansion: Integrating full-text archives of historical newspapers and historical census data to further refine and cross-reference existing biographies.
Academia vs Genealogy
Genealogy has had a poor reputation in Academia, since it tends to have low standards of evidence, consisting of "genealogy grandmas" entering information haphasardly without academic rigor. Academics have even invented a fancier term, "Prosopography", to distinguish the academic study of genealogy from the amateur phenomenon. The University of Oxford's Modern History research unit even notes on its home page, "If your interest is in tracing your family history, you should apply to the Society of Genealogists... We do not enter into correspondence about family history questions."[1]
The highest standard of humanities evidence-based ontology is the CIDOC CRM model, an ISO standard for cultural heritage information. No commercial genealogical site uses this ontology, however, since it is difficult to implement in an accessible manner for amateur genealogists.
Without a framework that can handle conflicting evidence as distinct historical events, a monolithic global tree inevitably becomes a "tree of lies." [8] It cannot overcome the weight of its own contradictions without requiring massive, unsustainable human effort to constantly referee disputes and untangle merged identities. By contrast, employing a proper data schema ensures that as more disparate sources are added, the overall database becomes stronger, not weaker. Each newly ingested census record or conflicting obituary acts as a cross-referenced node that builds a clearer probabilistic picture, rather than a destructive overwrite that breaks the tree.
Treating this architecture as a public good is crucial. When platforms attempt to monetize genealogical data, they inevitably rely on vendor lock-in or "treasure hunt" mechanics to drive subscriptions. This creates a perverse incentive to artificially hide historical records from users or force them to manually perform tedious, repetitive data-entry tasks that are far better handled by automated AI research agents.
An open, rigorously modeled database sidesteps these pitfalls and enables the best of both worlds. It harnesses crowdsourced wisdom, encouraging countless individuals with eyes on their own family histories to contribute invaluable local context—such as private photographs, forgotten community brochures, and old school yearbooks. At the same time, the underlying semantic architecture provides the necessary rigor to actually process, weigh, and use that information properly alongside the broader historical record.
Similar efforts
[1] William Addams Reitwiesner, who got a job as a cart-pusher at the library of congress to pursue his hobby of genealogy, and who published the biographies of thousands of royals, and others, before dying of cancer at age 56 in 2010.
[2] https://thundercitizen.ca by Eric Richardson
[3] https://citycouncilmatters.ca by Taren Francisco [9]
[4] https://justthemenu.ca by Stephen Garstin