Geyserich / System architecture
One source trail. Four connected responsibilities.
Geyserich is developing historical intelligence infrastructure for a connected historical knowledge layer. The aim is to structure relationships and context across archival sources while retaining their evidence, uncertainty and researcher review.
Archive Network is a partnership framework; StirpeAI’s current focus is reading, transcription and translation; Hesperides — Historical Knowledge Graph — is a development target; Ladon is a planned architecture layer. These are distinct responsibilities, not separate finished products.
01 / System responsibilities
The application is one part of the system.
Archive Network → StirpeAI → Hesperides — Historical Knowledge Graph → Ladon identifies the four responsibilities. Read from source collaboration through interpretation to knowledge; Ladon’s intended integrity, provenance and access principles apply across that workflow.
01 / Institutional source acquisition & collaboration
Archive Network
Partnership framework
Archive Network provides the proposed framework for institutional source acquisition and collaboration. Its intended role is to bring collection context, custody and agreed permissions into the knowledge workflow; participation would be agreed with each holder.
- Receives
- Collection scope, access conditions and archival descriptions
- Intended contribution
- Addressable records with source references and agreed use conditions
No connected archive network or institutional partnership is claimed here.
Explore Archive Network ↗02 / Application and interpretation layer
StirpeAI
Current application focus
StirpeAI’s current focus is AI-assisted reading, transcription and translation for genealogical research. Its intended role in the infrastructure is to turn readings into source-bound mentions and interpretations for researcher review, with extraction kept separate from identity matching.
- Receives
- Historical record and its language and collection context
- Intended contribution
- Readings and, in the intended workflow, source-linked entity mentions
The wider extraction, evidence-graph and review workflow is a development target. The public demonstration is authored and illustrative; it does not run the live product or an extraction model.
Explore StirpeAI ↗03 / Historical Knowledge Graph Layer
Hesperides
Development target
Hesperides — Historical Knowledge Graph — is the development target for connecting entities, evidence and historical context. Its intended model structures claims, people, events, places and sources while preserving time, uncertain geography and competing identity hypotheses.
- Receives
- Source-bound mentions, dated events and proposed relationships
- Intended contribution
- Evidence-linked exploration with explicit uncertainty and review state
This site does not expose a production graph database, search API or cross-archive identity service.
Explore Hesperides ↗04 / Integrity, provenance & access principles
Ladon
Planned architecture layer
Ladon’s purpose is to carry integrity, provenance and access principles across the architecture. Its intended role is to retain source lineage and decision history, with collection access conditions applying to original records and derived data.
- Receives
- Access policy, source lineage and review decisions
- Intended contribution
- Intended permission-aware access and traceable revisions
These are design requirements, not a security certification or a claim of implemented controls.
Explore Ladon ↗
02 / Shared design contract
Context should survive every transformation.
The target flow is collection → record → reading → source-bound mention → claim → human decision. A durable historical data model should retain evidence and context through each step. The complete workflow remains a development target, with implementation and validation still required.
Preserve the source
Keep the original reference, text and image region alongside every interpretation.
Separate observation from inference
An observation is not an inference; extraction is not identity. A proposed graph connection is not confirmation or historical truth.
Retain context through time
Event dates, place interpretations and later review decisions must remain distinguishable.
Make review reversible
A corrected reading should not erase the record or the reason for an earlier decision.
Design documentation