Inspect
Examine the source in detail.
Evidence before interpretation Every symbol.
Every symbol.
Every claim.
Traceable.
A free to use community workbench for inspecting MS 408, creating structured transcriptions, and testing research hypotheses.
Currently training on Isambard-AI, with the research workbench expected to open publicly in September 2026.
Human-in-the-loop·Source-linked·Reproducible·Rights-aware
Transcribe
Create structured, source-linked transcriptions.
Compare
Compare patterns across the corpus.
Test
Test hypotheses with bounded claims.
Record
Record decisions for future researchers.
Transcription overlays
L1ϟ ʘ ꝏ · ʒ ϙ ʘ ʘ ꝏ
L2ʒ ꝏ ϟ · ʘ ʘ ϙ ꝏ ʒ
L3ꝏ ʒ ʘ · ϟ ϙ ꝏ ʘ
L4ϙ ʘ ʒ · ꝏ ʘ ϟ ʒ
L5ϟ ꝏ ϙ · ʒ ʘ ϟ ꝏ ʘ
Display: glyph+word-like groups·lines: 1—5·zoom: 100%
System architecture
Several models, working on the same evidence.
The workbench is designed to combine multiple OCR systems with visual recognition, sequence models, statistical analysis and cross-script search. Each examines a different part of the manuscript. Their results are compared and returned as research candidates for human review.
- System type
- Human-guided multimodal research ensemble
- Model families
- Multiple OCR engines, visual recognition networks, character-sequence models, clustering and semantic retrieval
- Comparative corpus
- Over 70 languages and writing systems; over 40 grimoires, manuscripts and ancient texts
- Historical range
- Approximately 10,000 BCE through the early modern period
- What it returns
- Transcription candidates, confidence scores, similarities and testable hypotheses
- Researcher role
- Review, compare and decide what enters the research record