Internal R&D / Workflow design experiment
Internal workflow design reference.
These fictional examples explain the intended workflow behind StirpeAI: archival input, reading, source-bound mentions, evidence graphs and researcher review. They illustrate design decisions, not a completed infrastructure implementation.
Observation ≠ inference. Extraction ≠ identity. Proposal ≠ confirmation. Graph connection ≠ historical truth. No model runs in this demonstration.
Design illustration · Fictional passages · No verified historical claims. The separate manual pilot uses short real-source excerpts, with its own limitations. The complete workflow remains a development target.
00 / Intended system flow
Context across the source trail.
Inspect the proposed transformations and their boundaries. This diagram and workflow describe design responsibilities; the separate manual pilot does not implement AI processing or a production graph.
03 / Interpretation layer
AI extraction
AI-assisted transcription and language processing propose readings of historical handwriting. Researchers assess the interpretation against the source; ambiguous passages remain marked for review.
OutputTranscribed text · language · readings for review
Illustrative passage / 1842
Die XIV Maii, anno MDCCCXLII,
Joannes Martin et Maria Laurent
matrimonium contraxerunt.
Structured interpretation
- People
- Joannes Martin · Maria Laurent
- Date
- 14 May 1842
- Event
- Marriage
From record to relationship
- 01
Handwritten record
Preserve the page and its archival reference.
- 02
AI-assisted transcription
Propose readings; retain uncertain passages for review.
- 03
Entity extraction
Identify people, places, dates and events.
- 04
Evidence graph
Link claims to sources and retain their review state.
- 05
Researcher-reviewed output
Record decisions, citations and unresolved questions.
The research challenge extends beyond reading a page: it is connecting evidence without losing its provenance.
01 / The research workflow
From the historical record to a reviewed interpretation.
Follow five stages from source to review. Inspect the intermediate objects and their limits using two fictional records. No model is running in this demonstration.
01 / Source
Archival input
Begin with the record and its reference. These authored Latin passages stand in for documents; no real manuscript image is used.
Illustrative object
DEMO-001, fol. 12 · Latin · marriage passage
A real intake would also require custody, image region and permitted uses.
Read the complete workflow as text
- Archival input. Begin with the record and its reference. These authored Latin passages stand in for documents; no real manuscript image is used. DEMO-001, fol. 12 · Latin · marriage passage. A real intake would also require custody, image region and permitted uses.
- AI-assisted processing. Transcription and translation produce candidate readings. Here the text and translation are authored examples, not live AI results. On 14 May 1842, Joannes Martin and Maria Laurent contracted marriage.. Retain uncertain characters and competing readings. Do not assign an invented confidence score.
- Entities & evidence. Separate the names, event and date from interpretation. Each observation points back to the exact passage. Joannes Martin · Maria Laurent · marriage · 14 May 1842 → DEMO-001. The passage gives no parents, birthplace or event location.
- Knowledge graph. Connect the participants to their event and source. Compare the later residence mention without silently merging identities. 1842 marriage ↔ source; 1843 residence ↔ separate source; possible shared identity remains proposed. Catalogue location is not proof of event geography. Matching names do not establish migration.
- Researcher-reviewed output. The intended handover combines claims, citations, review decisions and open questions. In this example, image comparison and identity review remain pending. Marriage claim · DEMO-001, fol. 12 · review pending · place and cross-record identity unresolved. No completed researcher verification is simulated or asserted.
Fictional input / DEMO-001 · fol. 12
Die XIV Maii, anno MDCCCXLII, Joannes Martin et Maria Laurent matrimonium contraxerunt.On 14 May 1842, Joannes Martin and Maria Laurent contracted marriage.
Anlier is a proposed catalogue location, not an event location established by the passage.
Fictional input / DEMO-002 · fol. 3
Die II Junii, anno MDCCCXLIII, Joannes Martin in Arlon habitans memoratur.On 2 June 1843, Joannes Martin, residing in Arlon, is mentioned.
A matching name does not prove this is the participant from DEMO-001. No journey or migration is established.
02 / Explore the evidence graph
Relationships retain their evidence and review state.
The intended graph connects persons, places, events, documents, sources and time periods. Family relationships are one part of this historical knowledge system. This small illustration shows two persons, a dated event, a document reference and a proposed place.
Confirmed relationships
A researcher has checked the supporting sources and recorded a decision. Confirmation applies to the specific claim and evidence reviewed; it can be revised.
Proposed relationships
A reading or inference suggests a connection, but review is incomplete. Source-supported extraction alone does not make it confirmed.
Unresolved questions
Missing or conflicting evidence prevents a conclusion. The question stays open rather than becoming a relationship asserted as fact.
Event / Passage-supported / review pending
Marriage · 1842
The fictional passage records a marriage on 14 May 1842. Both persons are linked to this dated event by the same document. This extraction is awaiting source review, not a confirmed historical conclusion.
Connected evidence
- Joannes Martin → participant → Marriage · 1842
- Maria Laurent → participant → Marriage · 1842
- Marriage · 1842 → recorded in → DEMO-001 · fol. 12
- Marriage · 1842 → proposed location → Anlier · proposed
Read all relationships as text
- Joannes Martin → participant → Marriage · 1842. Supported by the fictional passage; researcher review pending.
- Maria Laurent → participant → Marriage · 1842. Supported by the fictional passage; researcher review pending.
- Marriage · 1842 → recorded in → DEMO-001 · fol. 12. Supported by the fictional passage; researcher review pending.
- Marriage · 1842 → proposed location → Anlier · proposed. Proposed; event location unresolved.
03 / Historical context
A graph unfolds through time and place.
Compare event dates and place evidence without turning a possible identity match into a migration story. Filter the timeline and exclude the proposed connection.
Event timeline
Marriage
Joannes Martin · Maria Laurent
DEMO-001 · fol. 12
Residence mention
Joannes Martin (identity unresolved)
DEMO-002 · fol. 3
Geographic context
DEMO-001 · fol. 12
Not stated in the passage
Anlier is a proposed catalogue location, not an event location established by the passage.
DEMO-002 · fol. 3
Arlon, as written in this fictional passage
A matching name does not prove this is the participant from DEMO-001. No journey or migration is established.
Textual place evidence only. No coordinates, route or historical boundary are inferred.
Identity connection / Proposed
Joannes Martin in DEMO-001 possibly the Joannes Martin in DEMO-002. This is an unresolved hypothesis, not evidence of movement from Anlier to Arlon.
Event time describes the historical observation. Review time would describe when a researcher accepted, rejected or revised a claim. The demonstration contains no completed review dates.
04 / Researcher-reviewed output
Review decisions remain traceable.
In a completed review, the researcher checks the reading against the source image, records their decision and keeps unsupported claims open.
Illustrative output / review format
Joannes Martin & Maria Laurent
- Claim
- Marriage on 14 May 1842
- Citation
- DEMO-001, fol. 12, fictional passage
- Review
- Pending comparison with the source image
- Open question
- Event location; identity across other records
Verification stays with the researcher.
An extracted name is a reading. An identity match is a hypothesis. A reviewed conclusion needs supporting records. The workflow keeps each decision linked to its sources. Human review remains essential to historical interpretation; unresolved identities and locations remain open questions.
Read the research notes ↗Development reference